python_code
stringlengths 0
1.02M
| repo_name
stringlengths 9
48
| file_path
stringlengths 5
114
|
|---|---|---|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Class CollectiveAllReduceStrategy implementing DistributionStrategy."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import copy
from tensorflow.core.protobuf import config_pb2
from tensorflow.core.protobuf import rewriter_config_pb2
from tensorflow.core.protobuf import tensorflow_server_pb2
from tensorflow.python.distribute import cross_device_ops as cross_device_ops_lib
from tensorflow.python.distribute import cross_device_utils
from tensorflow.python.distribute import device_util
from tensorflow.python.distribute import distribute_lib
from tensorflow.python.distribute import input_lib
from tensorflow.python.distribute import mirrored_strategy
from tensorflow.python.distribute import multi_worker_util
from tensorflow.python.distribute import numpy_dataset
from tensorflow.python.distribute import reduce_util
from tensorflow.python.distribute import values
from tensorflow.python.distribute.cluster_resolver import SimpleClusterResolver
from tensorflow.python.distribute.cluster_resolver import TFConfigClusterResolver
from tensorflow.python.eager import context
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import collective_ops
from tensorflow.python.platform import tf_logging as logging
from tensorflow.python.util.tf_export import tf_export
# TODO(yuefengz): support in-graph replication.
@tf_export("distribute.experimental.MultiWorkerMirroredStrategy", v1=[])
class CollectiveAllReduceStrategy(distribute_lib.Strategy):
"""Distribution strategy that uses collective ops for all-reduce.
It is similar to MirroredStrategy but it uses collective ops for reduction.
By default it uses all local GPUs or CPU for single-worker training.
When 'TF_CONFIG' environment variable is given, it parses cluster_spec,
task_type and task_id from 'TF_CONFIG' and turns into a multi-worker strategy
which mirrores models on GPUs of all machines in a cluster. In the current
implementation, it uses all GPUs in a cluster and it assumes all workers have
the same number of GPUs.
It supports both eager mode and graph mode. However, for eager mode, it has to
set up the eager context in its constructor and therefore all ops in eager
mode have to run after the strategy object is created.
Args:
communication: optional Enum of type
`distribute.experimental.CollectiveCommunication`. This provides a way
for the user to override the choice of collective op communication.
Possible values include `AUTO`, `RING`, and `NCCL`.
"""
def __init__(
self,
communication=cross_device_ops_lib.CollectiveCommunication.AUTO):
"""Initializes the object."""
super(CollectiveAllReduceStrategy, self).__init__(
CollectiveAllReduceExtended(
self,
communication=communication))
@classmethod
def _from_local_devices(cls, devices):
"""A convenience method to create an obejct with a list of devices."""
obj = cls()
obj.extended._initialize_local(TFConfigClusterResolver(), devices=devices) # pylint: disable=protected-access
return obj
@tf_export(v1=["distribute.experimental.MultiWorkerMirroredStrategy"])
class CollectiveAllReduceStrategyV1(distribute_lib.StrategyV1):
__doc__ = CollectiveAllReduceStrategy.__doc__
def __init__(
self,
communication=cross_device_ops_lib.CollectiveCommunication.AUTO):
"""Initializes the object."""
super(CollectiveAllReduceStrategyV1, self).__init__(
CollectiveAllReduceExtended(
self,
communication=communication))
class CollectiveAllReduceExtended(mirrored_strategy.MirroredExtended):
"""Implementation of CollectiveAllReduceStrategy."""
def __init__(self,
container_strategy,
communication,
cluster_resolver=TFConfigClusterResolver()):
distribute_lib.StrategyExtendedV1.__init__(self, container_strategy)
assert isinstance(
communication,
cross_device_ops_lib.CollectiveCommunication)
self._communication = communication
self._initialize_strategy(cluster_resolver)
assert isinstance(self._get_cross_device_ops(),
cross_device_ops_lib.CollectiveAllReduce)
def _initialize_strategy(self, cluster_resolver):
if cluster_resolver.cluster_spec().as_dict():
self._initialize_multi_worker(cluster_resolver)
else:
self._initialize_local(cluster_resolver)
def _initialize_local(self, cluster_resolver, devices=None):
"""Initializes the object for local training."""
self._is_chief = True
self._num_workers = 1
if ops.executing_eagerly_outside_functions():
try:
context.context().configure_collective_ops(
scoped_allocator_enabled_ops=("CollectiveReduce",),
use_nccl_communication=(self._communication == cross_device_ops_lib
.CollectiveCommunication.NCCL))
except RuntimeError:
logging.warning("Collective ops is not configured at program startup. "
"Some performance features may not be enabled.")
self._collective_ops_configured = True
# TODO(b/126786766): TFConfigClusterResolver returns wrong number of GPUs in
# some cases.
if isinstance(cluster_resolver, TFConfigClusterResolver):
num_gpus = context.num_gpus()
else:
num_gpus = cluster_resolver.num_accelerators().get("GPU", 0)
if devices:
local_devices = devices
else:
if num_gpus:
local_devices = tuple("/device:GPU:%d" % i for i in range(num_gpus))
else:
local_devices = ("/device:CPU:0",)
self._worker_device = device_util.canonicalize("/device:CPU:0")
self._host_input_device = numpy_dataset.SingleDevice(self._worker_device)
self._collective_keys = cross_device_utils.CollectiveKeys()
super(CollectiveAllReduceExtended, self)._initialize_local(local_devices)
# TODO(yuefengz): remove num_gpus_per_worker from CollectiveAllReduce.
self._cross_device_ops = cross_device_ops_lib.CollectiveAllReduce(
num_workers=self._num_workers,
num_gpus_per_worker=num_gpus,
collective_keys=self._collective_keys)
self._cluster_spec = None
self._task_type = None
self._task_id = None
# This is a mark to tell whether we are running with standalone client or
# independent worker. Right now with standalone client, strategy object is
# created as local strategy and then turn into multi-worker strategy via
# configure call.
self._local_or_standalone_client_mode = True
# Save the num_gpus_per_worker and rpc_layer for configure method.
self._num_gpus_per_worker = num_gpus
self._rpc_layer = cluster_resolver.rpc_layer
self._warn_nccl_no_gpu()
logging.info("Single-worker CollectiveAllReduceStrategy with local_devices "
"= %r, communication = %s", local_devices, self._communication)
def _initialize_multi_worker(self, cluster_resolver):
"""Initializes the object for multi-worker training."""
cluster_spec = multi_worker_util.normalize_cluster_spec(
cluster_resolver.cluster_spec())
task_type = cluster_resolver.task_type
task_id = cluster_resolver.task_id
if task_type is None or task_id is None:
raise ValueError("When `cluster_spec` is given, you must also specify "
"`task_type` and `task_id`.")
self._cluster_spec = cluster_spec
self._task_type = task_type
self._task_id = task_id
self._num_workers = multi_worker_util.worker_count(cluster_spec, task_type)
if not self._num_workers:
raise ValueError("No `worker`, `chief` or `evaluator` tasks can be found "
"in `cluster_spec`.")
self._is_chief = multi_worker_util.is_chief(cluster_spec, task_type,
task_id)
self._worker_device = "/job:%s/task:%d" % (task_type, task_id)
self._host_input_device = numpy_dataset.SingleDevice(self._worker_device)
if (ops.executing_eagerly_outside_functions() and
not getattr(self, "_local_or_standalone_client_mode", False)):
context.context().configure_collective_ops(
collective_leader=multi_worker_util.collective_leader(
cluster_spec, task_type, task_id),
scoped_allocator_enabled_ops=("CollectiveReduce",),
use_nccl_communication=(self._communication == cross_device_ops_lib
.CollectiveCommunication.NCCL),
device_filters=("/job:%s/task:%d" % (task_type, task_id),))
self._collective_ops_configured = True
# Starting a std server in eager mode and in independent worker mode.
if (context.executing_eagerly() and
not getattr(self, "_std_server_started", False) and
not getattr(self, "_local_or_standalone_client_mode", False)):
# Checking _local_or_standalone_client_mode as well because we should not
# create the std server in standalone client mode.
config_proto = config_pb2.ConfigProto()
config_proto = self._update_config_proto(config_proto)
server_def = tensorflow_server_pb2.ServerDef(
cluster=cluster_spec.as_cluster_def(),
default_session_config=config_proto,
job_name=task_type,
task_index=task_id,
protocol=cluster_resolver.rpc_layer or "grpc")
context.context().enable_collective_ops(server_def)
self._std_server_started = True
# The `ensure_initialized` is needed before calling
# `context.context().devices()`.
context.context().ensure_initialized()
logging.info(
"Enabled multi-worker collective ops with available devices: %r",
context.context().devices())
# TODO(yuefengz): The `num_gpus` is only for this particular task. It
# assumes all workers have the same number of GPUs. We should remove this
# assumption by querying all tasks for their numbers of GPUs.
# TODO(b/126786766): TFConfigClusterResolver returns wrong number of GPUs in
# some cases.
if isinstance(cluster_resolver, TFConfigClusterResolver):
num_gpus = context.num_gpus()
else:
num_gpus = cluster_resolver.num_accelerators().get("GPU", 0)
if num_gpus:
local_devices = tuple("%s/device:GPU:%d" % (self._worker_device, i)
for i in range(num_gpus))
else:
local_devices = (self._worker_device,)
self._collective_keys = cross_device_utils.CollectiveKeys()
super(CollectiveAllReduceExtended, self)._initialize_local(local_devices)
self._input_workers = input_lib.InputWorkers(
self._device_map, [(self._worker_device, self.worker_devices)])
self._cross_device_ops = cross_device_ops_lib.CollectiveAllReduce(
num_workers=self._num_workers,
num_gpus_per_worker=num_gpus,
collective_keys=self._collective_keys)
# Add a default device so that ops without specified devices will not end up
# on other workers.
self._default_device = "/job:%s/task:%d" % (task_type, task_id)
# Save the num_gpus_per_worker and rpc_layer for configure method.
self._num_gpus_per_worker = num_gpus
self._rpc_layer = cluster_resolver.rpc_layer
self._warn_nccl_no_gpu()
logging.info(
"Multi-worker CollectiveAllReduceStrategy with cluster_spec = %r, "
"task_type = %r, task_id = %r, num_workers = %r, local_devices = %r, "
"communication = %s", cluster_spec.as_dict(), task_type,
task_id, self._num_workers, local_devices,
self._communication)
def _get_variable_creator_initial_value(self,
replica_id=0,
device=None,
primary_var=None,
**kwargs):
if replica_id == 0: # First replica on each worker.
assert device is not None
assert primary_var is None
def initial_value_fn(): # pylint: disable=g-missing-docstring
# Only the first device participates in the broadcast of initial values.
group_key = self._collective_keys.get_group_key([device])
group_size = self._num_workers
collective_instance_key = (
self._collective_keys.get_variable_instance_key())
with ops.device(device):
initial_value = kwargs["initial_value"]
if callable(initial_value):
initial_value = initial_value()
assert not callable(initial_value)
initial_value = ops.convert_to_tensor(
initial_value, dtype=kwargs.get("dtype", None))
if self._num_workers > 1:
if self._is_chief:
bcast_send = collective_ops.broadcast_send(
initial_value, initial_value.shape, initial_value.dtype,
group_size, group_key, collective_instance_key)
with ops.control_dependencies([bcast_send]):
return array_ops.identity(initial_value)
else:
return collective_ops.broadcast_recv(initial_value.shape,
initial_value.dtype,
group_size, group_key,
collective_instance_key)
return initial_value
return initial_value_fn
else:
return super(CollectiveAllReduceExtended,
self)._get_variable_creator_initial_value(
replica_id=replica_id,
device=device,
primary_var=primary_var,
**kwargs)
def _make_input_context(self):
if self._cluster_spec is None:
input_pipeline_id = 0
else:
input_pipeline_id = multi_worker_util.id_in_cluster(
self._cluster_spec, self._task_type, self._task_id)
input_context = distribute_lib.InputContext(
num_input_pipelines=self._num_workers,
input_pipeline_id=input_pipeline_id,
num_replicas_in_sync=self._num_replicas_in_sync)
return input_context
def _experimental_distribute_dataset(self, dataset):
input_context = self._make_input_context()
return input_lib.get_distributed_dataset(
dataset,
self._input_workers,
self._container_strategy(),
split_batch_by=self._num_replicas_in_sync,
input_context=input_context)
def _make_dataset_iterator(self, dataset):
"""Distributes the dataset to each local GPU."""
input_context = self._make_input_context()
return input_lib.DatasetIterator(
dataset,
self._input_workers,
self._container_strategy(),
split_batch_by=self._num_replicas_in_sync,
input_context=input_context)
def _make_input_fn_iterator(
self,
input_fn,
replication_mode=distribute_lib.InputReplicationMode.PER_WORKER):
"""Distributes the input function to each local GPU."""
input_context = self._make_input_context()
return input_lib.InputFunctionIterator(input_fn, self._input_workers,
[input_context],
self._container_strategy())
def _configure(self,
session_config=None,
cluster_spec=None,
task_type=None,
task_id=None):
"""Configures the object.
Args:
session_config: a `tf.compat.v1.ConfigProto`
cluster_spec: a dict, ClusterDef or ClusterSpec object specifying the
cluster configurations.
task_type: the current task type, such as "worker".
task_id: the current task id.
Raises:
ValueError: if `task_type` is not in the `cluster_spec`.
"""
if cluster_spec:
# Use the num_gpus_per_worker recorded in constructor since _configure
# doesn't take num_gpus.
cluster_resolver = SimpleClusterResolver(
cluster_spec=multi_worker_util.normalize_cluster_spec(cluster_spec),
task_type=task_type,
task_id=task_id,
num_accelerators={"GPU": self._num_gpus_per_worker},
rpc_layer=self._rpc_layer)
self._initialize_multi_worker(cluster_resolver)
assert isinstance(self._get_cross_device_ops(),
cross_device_ops_lib.CollectiveAllReduce)
if session_config:
session_config.CopyFrom(self._update_config_proto(session_config))
def _update_config_proto(self, config_proto):
updated_config = copy.deepcopy(config_proto)
# Enable the scoped allocator optimization for CollectiveOps. This
# optimization converts many small all-reduces into fewer larger
# all-reduces.
rewrite_options = updated_config.graph_options.rewrite_options
rewrite_options.scoped_allocator_optimization = (
rewriter_config_pb2.RewriterConfig.ON)
# We turn on ScopedAllocator only for CollectiveReduce op, i.e. enable_op =
# ["CollectiveReduce"]. Since we can't assign to a repeated proto field, we
# clear and then append.
del rewrite_options.scoped_allocator_opts.enable_op[:]
rewrite_options.scoped_allocator_opts.enable_op.append("CollectiveReduce")
if self._communication == cross_device_ops_lib.CollectiveCommunication.NCCL:
updated_config.experimental.collective_nccl = True
if not self._cluster_spec:
return updated_config
assert self._task_type
assert self._task_id is not None
# Collective group leader is needed for collective ops to coordinate
# workers.
updated_config.experimental.collective_group_leader = (
multi_worker_util.collective_leader(self._cluster_spec, self._task_type,
self._task_id))
# The device filters prevent communication between workers.
del updated_config.device_filters[:]
updated_config.device_filters.append(
"/job:%s/task:%d" % (self._task_type, self._task_id))
return updated_config
def _reduce_to(self, reduce_op, value, destinations):
if (isinstance(value, values.Mirrored) and
reduce_op == reduce_util.ReduceOp.MEAN):
return value
assert not isinstance(value, values.Mirrored)
if (isinstance(value, values.DistributedValues) and
len(self.worker_devices) == 1):
value = value.values[0]
# When there are multiple workers, we need to reduce across workers using
# collective ops.
if (not isinstance(value, values.DistributedValues) and
self._num_workers == 1):
# This function handles reducing values that are not PerReplica or
# Mirrored values. For example, the same value could be present on all
# replicas in which case `value` would be a single value or value could
# be 0.
return cross_device_ops_lib.reduce_non_distributed_value(
reduce_op, self._device_map, value, destinations)
return self._get_cross_device_ops().reduce(
reduce_op, value, destinations=destinations)
def _warn_nccl_no_gpu(self):
if ((self._communication ==
cross_device_ops_lib.CollectiveCommunication.NCCL) and
self._num_gpus_per_worker == 0):
logging.warning("Enabled NCCL communication but no GPUs detected/"
"specified.")
@property
def experimental_between_graph(self):
return True
@property
def experimental_should_init(self):
return True
@property
def should_checkpoint(self):
return self._is_chief
@property
def should_save_summary(self):
return self._is_chief
@property
def _num_replicas_in_sync(self):
return len(self.worker_devices) * self._num_workers
# TODO(priyag): Delete this once all strategies use global batch size.
@property
def _global_batch_size(self):
"""`make_dataset_iterator` and `make_numpy_iterator` use global batch size.
`make_input_fn_iterator` assumes per-replica batching.
Returns:
Boolean.
"""
return True
|
tensorflow-master
|
tensorflow/python/distribute/collective_all_reduce_strategy.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Training utilities for Estimator to use Distribute Coordinator."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import copy
import six
from tensorflow.python.distribute import distribute_coordinator as dc
from tensorflow.python.distribute import distribute_coordinator_context as dc_context
from tensorflow.python.distribute import multi_worker_util
from tensorflow.python.platform import tf_logging as logging
from tensorflow.python.training import server_lib
# pylint: disable=protected-access
CHIEF = dc._TaskType.CHIEF
EVALUATOR = dc._TaskType.EVALUATOR
PS = dc._TaskType.PS
WORKER = dc._TaskType.WORKER
# pylint: enable=protected-access
def _count_ps(cluster_spec):
"""Counts the number of parameter servers in cluster_spec."""
if not cluster_spec:
raise RuntimeError(
'Internal error: `_count_ps` does not expect empty cluster_spec.')
return len(cluster_spec.as_dict().get(PS, []))
def _count_worker(cluster_spec, chief_task_type):
"""Counts the number of workers (including chief) in cluster_spec."""
if not cluster_spec:
raise RuntimeError(
'Internal error: `_count_worker` does not expect empty cluster_spec.')
return (len(cluster_spec.as_dict().get(WORKER, [])) + len(
cluster_spec.as_dict().get(chief_task_type, [])))
def _get_global_id(cluster_spec, task_type, task_id, chief_task_type):
"""Returns the global id of the given task type in a cluster."""
if not task_type:
return 0
# Sort task names in cluster by "chief"/"master", "evaluator", "worker"
# and "ps". More details can be found at the documentation of
# `tf.estimator.RunConfig.global_id_in_cluster`.
task_type_ordered_list = []
if chief_task_type in cluster_spec.jobs:
task_type_ordered_list = [chief_task_type]
task_type_ordered_list.extend([
t for t in sorted(cluster_spec.jobs) if t != chief_task_type and t != PS
])
if PS in cluster_spec.jobs:
task_type_ordered_list.append(PS)
# Find the right global_id for current task.
next_global_id = 0
for t in task_type_ordered_list:
if t == task_type:
return next_global_id + task_id
# `cluster_spec.job_tasks` returns all task addresses of type `t`.
next_global_id += len(cluster_spec.job_tasks(t))
# It is unexpected that it passes through all task_types in
# `task_type_ordered_list`.
raise RuntimeError('Internal Error: `task_type` ({}) is not in '
'cluster_spec ({}).'.format(task_type, cluster_spec))
def _init_run_config_from_worker_context(config, worker_context):
"""Initializes run config from distribute coordinator's worker context."""
# pylint: disable=protected-access
config._service = None
config._cluster_spec = worker_context.cluster_spec
config._task_type = worker_context.task_type
config._task_id = worker_context.task_id
config._evaluation_master = worker_context.master_target
config._master = worker_context.master_target
config._is_chief = worker_context.is_chief
if config._cluster_spec:
# Distributed mode.
if config._task_type != EVALUATOR:
config._num_ps_replicas = _count_ps(config._cluster_spec)
config._num_worker_replicas = _count_worker(
config._cluster_spec, chief_task_type=CHIEF)
config._global_id_in_cluster = _get_global_id(
config._cluster_spec,
config._task_type,
config._task_id,
chief_task_type=CHIEF)
else:
# Evaluator task should not be aware of the other tasks.
config._cluster_spec = server_lib.ClusterSpec({})
config._num_ps_replicas = 0
config._num_worker_replicas = 0
config._global_id_in_cluster = None # undefined
else:
# Local mode.
config._global_id_in_cluster = 0
config._num_ps_replicas = 0
config._num_worker_replicas = 1
def init_run_config(config, tf_config):
"""Initializes RunConfig for distribution strategies."""
# pylint: disable=protected-access
if (config._experimental_distribute and
config._experimental_distribute.train_distribute):
if config._train_distribute:
raise ValueError('Either `train_distribute` or'
'`experimental_distribute.train_distribute` can be set.')
config._train_distribute = config._experimental_distribute.train_distribute
if (config._experimental_distribute and
config._experimental_distribute.eval_distribute):
if config._eval_distribute:
raise ValueError('Either `eval_distribute` or'
'`experimental_distribute.eval_distribute` can be set.')
config._eval_distribute = config._experimental_distribute.eval_distribute
cluster_spec = server_lib.ClusterSpec(tf_config.get('cluster', {}))
config._init_distributed_setting_from_environment_var({})
# Use distribute coordinator with STANDALONE_CLIENT mode if
# `experimental_distribute.remote_cluster` is set.
if (config._train_distribute and config._experimental_distribute and
config._experimental_distribute.remote_cluster):
if cluster_spec:
raise ValueError('Cannot set both "cluster_spec" of TF_CONFIG and '
'`experimental_distribute.remote_cluster`')
config._distribute_coordinator_mode = dc.CoordinatorMode.STANDALONE_CLIENT
config._cluster_spec = config._experimental_distribute.remote_cluster
logging.info('RunConfig initialized for Distribute Coordinator with '
'STANDALONE_CLIENT mode')
return
# Don't use distribute coordinator if it is local training or cluster has a
# MASTER job or `train_distribute` is not specifed.
if (not cluster_spec or 'master' in cluster_spec.jobs or
not config._train_distribute):
config._distribute_coordinator_mode = None
config._init_distributed_setting_from_environment_var(tf_config)
config._maybe_overwrite_session_config_for_distributed_training()
logging.info('Not using Distribute Coordinator.')
return
# Use distribute coordinator with INDEPENDENT_WORKER mode otherwise.
assert tf_config
# Set the cluster_spec only since the distributed setting will come from
# distribute coordinator.
config._cluster_spec = cluster_spec
config._distribute_coordinator_mode = dc.CoordinatorMode.INDEPENDENT_WORKER
logging.info('RunConfig initialized for Distribute Coordinator with '
'INDEPENDENT_WORKER mode')
def should_run_distribute_coordinator(config):
"""Checks the config to see whether to run distribute coordinator."""
# pylint: disable=protected-access
if (not hasattr(config, '_distribute_coordinator_mode') or
config._distribute_coordinator_mode is None):
logging.info('Not using Distribute Coordinator.')
return False
if (not isinstance(config._distribute_coordinator_mode, six.string_types) or
config._distribute_coordinator_mode not in [
dc.CoordinatorMode.STANDALONE_CLIENT,
dc.CoordinatorMode.INDEPENDENT_WORKER
]):
logging.warning('Unexpected distribute_coordinator_mode: %r',
config._distribute_coordinator_mode)
return False
if not config.cluster_spec:
logging.warning('Running `train_and_evaluate` locally, ignoring '
'`experimental_distribute_coordinator_mode`.')
return False
return True
def train_and_evaluate(estimator, train_spec, eval_spec, executor_cls):
"""Run distribute coordinator for Estimator's `train_and_evaluate`.
Args:
estimator: An `Estimator` instance to train and evaluate.
train_spec: A `TrainSpec` instance to specify the training specification.
eval_spec: A `EvalSpec` instance to specify the evaluation and export
specification.
executor_cls: the evaluation executor class of Estimator.
Raises:
ValueError: if `distribute_coordinator_mode` is None in RunConfig.
"""
run_config = estimator.config
if not run_config._distribute_coordinator_mode: # pylint: disable=protected-access
raise ValueError(
'Distribute coordinator mode is not specified in `RunConfig`.')
def _worker_fn(strategy):
"""Function for worker task."""
local_estimator = copy.deepcopy(estimator)
# pylint: disable=protected-access
local_estimator._config._train_distribute = strategy
context = dc_context.get_current_worker_context()
_init_run_config_from_worker_context(local_estimator._config, context)
logging.info('Updated config: %s', str(vars(local_estimator._config)))
local_estimator._train_distribution = strategy
# pylint: enable=protected-access
# In the standalone client, we don't need to run hooks on all threads
# because logging hooks on all threads may be too much on the screen; also
# tensor passed to one hook can only be fetched with the graph where the
# tensor is defined. Other hooks such as checkpointing hooks will added by
# MonitoredTrainingSession.
# TODO(yuefengz): Is there a hook that does need to run on all threads in
# standalone client mode?
if (run_config._distribute_coordinator_mode == # pylint: disable=protected-access
dc.CoordinatorMode.INDEPENDENT_WORKER or context.is_chief):
hooks = list(train_spec.hooks)
else:
hooks = []
# Prevent estimator.train from calling distribute coordinator again. This
# function calls estimator.train which will use distribute coordinator path
# again if `_distribute_coordinator_mode` is set.
local_estimator._config._distribute_coordinator_mode = None # pylint: disable=protected-access
local_estimator.train(
input_fn=train_spec.input_fn,
max_steps=train_spec.max_steps,
hooks=hooks)
def _eval_fn(strategy):
"""Function for evaluator task."""
local_estimator = copy.deepcopy(estimator)
# pylint: disable=protected-access
local_estimator._config._eval_distribute = strategy
_init_run_config_from_worker_context(
local_estimator._config, dc_context.get_current_worker_context())
logging.info('Updated config: %s', str(vars(local_estimator._config)))
local_estimator._eval_distribution = strategy
# Prevent estimator.evaluate from calling distribute coordinator again. This
# function calls estimator.evaluate which will use distribute coordinator
# path again if `_distribute_coordinator_mode` is set.
local_estimator._config._distribute_coordinator_mode = None # pylint: disable=protected-access
executor = executor_cls(local_estimator, train_spec, eval_spec)
executor._start_continuous_evaluation()
# pylint: enable=protected-access
# pylint: disable=protected-access
if (run_config._distribute_coordinator_mode ==
dc.CoordinatorMode.STANDALONE_CLIENT):
cluster_spec = run_config.cluster_spec
assert cluster_spec
else:
# The cluster_spec comes from TF_CONFIG environment variable if it is
# INDEPENDENT_WORKER mode.
cluster_spec = None
dc.run_distribute_coordinator(
_worker_fn,
run_config.train_distribute,
_eval_fn,
run_config.eval_distribute,
mode=run_config._distribute_coordinator_mode,
cluster_spec=cluster_spec,
session_config=run_config.session_config)
# TODO(yuefengz): maybe merge the following two functions?
# pylint: disable=protected-access
def estimator_train(estimator, train_distributed_fn, hooks):
"""Run distribute coordinator for Estimator's `train` method."""
assert estimator._config._distribute_coordinator_mode
run_config = estimator._config
assert estimator._config.cluster_spec
cluster_spec = multi_worker_util.normalize_cluster_spec(
estimator._config.cluster_spec)
assert estimator._config._train_distribute
if 'evaluator' in cluster_spec.jobs:
raise ValueError("'evaluator' job is not supported if you don't use "
'`train_and_evaluate`')
if (estimator._config._distribute_coordinator_mode != # pylint: disable=protected-access
dc.CoordinatorMode.STANDALONE_CLIENT):
raise ValueError('Only `STANDALONE_CLIENT` mode is supported when you call '
'`estimator.train`')
if estimator._config._train_distribute.extended.experimental_between_graph:
# TODO(yuefengz): remove this limitation once we figure out how to merge
# return values from `_worker_fn`s.
raise ValueError('`Estimator.train` API is not supported for %s with '
'`STANDALONE_CLIENT` mode.' %
estimator._config._train_distribute.__class__.__name__)
def _worker_fn(strategy):
"""Function for worker task."""
local_estimator = copy.deepcopy(estimator)
local_estimator._config._train_distribute = strategy
context = dc_context.get_current_worker_context()
_init_run_config_from_worker_context(local_estimator._config, context)
logging.info('Updated config: %s', str(vars(local_estimator._config)))
local_estimator._train_distribution = strategy
if context.is_chief:
chief_hooks = hooks
else:
chief_hooks = []
train_distributed_fn(local_estimator, strategy, chief_hooks)
return local_estimator
return dc.run_distribute_coordinator(
_worker_fn,
estimator._config.train_distribute,
mode=run_config._distribute_coordinator_mode,
cluster_spec=cluster_spec,
session_config=run_config.session_config)
def estimator_evaluate(estimator, evaluate_distributed_fn, hooks):
"""Run distribute coordinator for Estimator's `evaluate` method."""
assert estimator._config._distribute_coordinator_mode
run_config = estimator._config
assert estimator._config.cluster_spec
cluster_spec = multi_worker_util.normalize_cluster_spec(
estimator._config.cluster_spec)
assert estimator._config._eval_distribute
if 'evaluator' in cluster_spec.jobs:
raise ValueError("'evaluator' job is not supported if you don't use "
'`train_and_evaluate`')
if (estimator._config._distribute_coordinator_mode !=
dc.CoordinatorMode.STANDALONE_CLIENT):
raise ValueError('Only `STANDALONE_CLIENT` mode is supported when you call '
'`Estimator.evaluate`')
if estimator._config._eval_distribute.extended.experimental_between_graph:
# TODO(yuefengz): remove this limitation once we figure out how to merge
# return values from `_worker_fn`s.
raise ValueError('`Estimator.evaluate` API is not supported for %s with '
'`STANDALONE_CLIENT` mode.' %
estimator._config._eval_distribute.__class__.__name__)
def _worker_fn(strategy):
"""Function for evaluation."""
local_estimator = copy.deepcopy(estimator)
local_estimator._config._eval_distribute = strategy
context = dc_context.get_current_worker_context()
_init_run_config_from_worker_context(local_estimator._config, context)
logging.info('Updated config: %s', str(vars(local_estimator._config)))
local_estimator._eval_distribution = strategy
if context.is_chief:
chief_hooks = hooks
else:
chief_hooks = []
return evaluate_distributed_fn(local_estimator, strategy, chief_hooks)
return dc.run_distribute_coordinator(
_worker_fn,
estimator._config.eval_distribute,
mode=run_config._distribute_coordinator_mode,
cluster_spec=cluster_spec,
session_config=run_config.session_config)
# pylint: enable=protected-access
|
tensorflow-master
|
tensorflow/python/distribute/estimator_training.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for saving and loading with mixed APIs with distribution strategies.
For saving, Keras's export_saved_model() API is used; and for loading,
saved_model's load() API is used. Keras's export_save_model() when used with
`serving_only` parameter equals to True should be the same as using
tf.saved_model.save().
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.distribute import combinations
from tensorflow.python.distribute import saved_model_test_base as test_base
from tensorflow.python.eager import test
from tensorflow.python.keras.saving import saved_model as keras_saved_model
_DEFAULT_FUNCTION_KEY = 'serving_default'
class SavedModelSaveAndLoadTest(test_base.TestSavedModelBase):
def setUp(self):
self._root_dir = 'saved_model_save_load'
super(SavedModelSaveAndLoadTest, self).setUp()
def _save_model(self, model, saved_dir):
keras_saved_model.export_saved_model(model, saved_dir, serving_only=True)
def _load_and_run_model(self, distribution, saved_dir, predict_dataset,
output_name):
return test_base.load_and_run_with_saved_model_api(distribution, saved_dir,
predict_dataset,
output_name)
@combinations.generate(test_base.simple_models_with_strategies())
def test_save_no_strategy_restore_strategy(self, model_and_input,
distribution):
self.run_test_save_no_strategy_restore_strategy(model_and_input,
distribution)
@combinations.generate(
combinations.times(test_base.simple_models_with_strategies(),
combinations.combine(save_in_scope=[True, False])))
def test_save_strategy_restore_no_strategy(self, model_and_input,
distribution, save_in_scope):
if save_in_scope:
self.skipTest(('Saving model within tf.distribute.Strategy scope is not ',
'supported.'))
self.run_test_save_strategy_restore_no_strategy(model_and_input,
distribution, save_in_scope)
@combinations.generate(
combinations.times(test_base.simple_models_with_strategy_pairs(),
combinations.combine(save_in_scope=[True, False])))
def test_save_strategy_restore_strategy(self, model_and_input,
distribution_for_saving,
distribution_for_restoring,
save_in_scope):
if save_in_scope:
self.skipTest(('Saving model within tf.distribute.Strategy scope is not ',
'supported.'))
self.run_test_save_strategy_restore_strategy(model_and_input,
distribution_for_saving,
distribution_for_restoring,
save_in_scope)
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/saved_model_mixed_api_test.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Base class for testing saving/loading with DS."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
from absl.testing import parameterized
import numpy as np
from tensorflow.python.data.ops import dataset_ops
from tensorflow.python.distribute import combinations
from tensorflow.python.distribute import model_combinations
from tensorflow.python.distribute import strategy_combinations
from tensorflow.python.eager import test
from tensorflow.python.framework import random_seed
from tensorflow.python.ops import array_ops
from tensorflow.python.saved_model import saved_model
_RANDOM_SEED = 1337
_DEFAULT_FUNCTION_KEY = 'serving_default'
_TOLERANCE = 1e-30
PREDICT_STEPS = 1
simple_models = [
model_combinations.simple_functional_model,
model_combinations.simple_sequential_model,
# TODO(b/131715604): figure out why subclass model does not work
# model_combinations.simple_subclass_model,
]
strategies_minus_tpu = [
# TODO(b/132702156): include default strategy
strategy_combinations.one_device_strategy,
strategy_combinations.one_device_strategy_gpu,
strategy_combinations.mirrored_strategy_with_one_cpu,
strategy_combinations.mirrored_strategy_with_one_gpu,
strategy_combinations.mirrored_strategy_with_gpu_and_cpu,
strategy_combinations.mirrored_strategy_with_two_gpus
]
def simple_models_with_strategies():
return combinations.combine(
model_and_input=simple_models,
distribution=strategies_minus_tpu,
mode=['eager'])
def simple_models_with_strategy_pairs():
return combinations.combine(
model_and_input=simple_models,
distribution_for_saving=strategies_minus_tpu,
distribution_for_restoring=strategies_minus_tpu,
mode=['eager'])
def load_and_run_with_saved_model_api(distribution, saved_dir, predict_dataset,
output_name):
"""Loads a saved_model using tf.saved_model API, and runs it."""
func = saved_model.load(saved_dir)
if distribution:
dist_predict_dataset = distribution.experimental_distribute_dataset(
predict_dataset)
per_replica_predict_data = next(iter(dist_predict_dataset))
result = distribution.experimental_run_v2(
func.signatures[_DEFAULT_FUNCTION_KEY],
args=(per_replica_predict_data,))
result = result[output_name]
# Convert the per_replica value to a list, then concatenate them
reduced = distribution.experimental_local_results(result)
concat = array_ops.concat(reduced, 0)
return concat
else:
result = func.signatures[_DEFAULT_FUNCTION_KEY](next(iter(predict_dataset)))
return result[output_name]
class TestSavedModelBase(test.TestCase, parameterized.TestCase):
"""Base class for testing saving/loading with DS."""
def setUp(self):
np.random.seed(_RANDOM_SEED)
random_seed.set_random_seed(_RANDOM_SEED)
self._root_dir = 'base'
super(TestSavedModelBase, self).setUp()
def _save_model(self, model, saved_dir):
"""Save the given model to the given saved_dir.
This method needs to be implemeted by the subclasses.
Args:
model: a keras model object to save.
saved_dir: a string representing the path to save the keras model
"""
raise NotImplementedError('must be implemented in descendants')
def _load_and_run_model(self, distribution, saved_dir, predict_dataset,
output_name):
"""Load the model and run 1 step of predict with it.
This method must be implemented by the subclasses.
Args:
distribution: the distribution strategy used to load the model. None if no
distribution strategy is used
saved_dir: the string representing the path where the model is saved.
predict_dataset: the data used to do the predict on the model for
cross_replica context.
output_name: the string representing the name of the output layer of the
model.
"""
raise NotImplementedError('must be implemented in descendants')
def _train_model(self, model, x_train, y_train, batch_size):
training_dataset = dataset_ops.Dataset.from_tensor_slices(
(x_train, y_train))
training_dataset = training_dataset.repeat()
training_dataset = training_dataset.batch(batch_size)
# Train the model for 1 epoch
model.fit(x=training_dataset, epochs=1, steps_per_epoch=100)
def _get_predict_dataset(self, x_predict, batch_size):
predict_dataset = dataset_ops.Dataset.from_tensor_slices(x_predict)
predict_dataset = predict_dataset.repeat()
predict_dataset = predict_dataset.batch(batch_size)
return predict_dataset
def run_test_save_no_strategy_restore_strategy(self, model_and_input,
distribution):
"""Save a model without DS, and restore it with DS."""
saved_dir = os.path.join(self.get_temp_dir(), self._root_dir,
'test_save_no_dist_restore_dist')
model, output_name = model_and_input.get_model()
x_train, y_train, x_predict = model_and_input.get_data()
batch_size = model_and_input.get_batch_size()
self._train_model(model, x_train, y_train, batch_size)
predict_dataset = self._get_predict_dataset(x_predict, batch_size)
result_before_save = model.predict(predict_dataset, steps=PREDICT_STEPS)
self._save_model(model, saved_dir)
with distribution.scope():
result_after_save = self._load_and_run_model(
distribution=distribution,
saved_dir=saved_dir,
predict_dataset=predict_dataset,
output_name=output_name)
self.assertAllClose(result_before_save, result_after_save, atol=_TOLERANCE)
def run_test_save_strategy_restore_no_strategy(self, model_and_input,
distribution, save_in_scope):
"""Save a model with DS, and restore it without DS."""
saved_dir = os.path.join(self.get_temp_dir(), self._root_dir,
'test_save_no_dist_restore_dist')
with distribution.scope():
model, output_name = model_and_input.get_model()
x_train, y_train, x_predict = model_and_input.get_data()
batch_size = model_and_input.get_batch_size()
self._train_model(model, x_train, y_train, batch_size)
predict_dataset = self._get_predict_dataset(x_predict, batch_size)
result_before_save = model.predict(predict_dataset, steps=PREDICT_STEPS)
if save_in_scope:
with distribution.scope():
self._save_model(model, saved_dir)
else:
self._save_model(model, saved_dir)
load_result = self._load_and_run_model(
distribution=None,
saved_dir=saved_dir,
predict_dataset=predict_dataset,
output_name=output_name)
self.assertAllClose(result_before_save, load_result, atol=_TOLERANCE)
def run_test_save_strategy_restore_strategy(self, model_and_input,
distribution_for_saving,
distribution_for_restoring,
save_in_scope):
"""Save a model with DS, and restore it with potentially different DS."""
saved_dir = os.path.join(self.get_temp_dir(), self._root_dir,
'test_save_dist_restore_dist')
with distribution_for_saving.scope():
model, output_name = model_and_input.get_model()
x_train, y_train, x_predict = model_and_input.get_data()
batch_size = model_and_input.get_batch_size()
self._train_model(model, x_train, y_train, batch_size)
predict_dataset = self._get_predict_dataset(x_predict, batch_size)
result_before_save = model.predict(predict_dataset, steps=PREDICT_STEPS)
if save_in_scope:
with distribution_for_saving.scope():
self._save_model(model, saved_dir)
else:
self._save_model(model, saved_dir)
with distribution_for_restoring.scope():
load_result = self._load_and_run_model(
distribution=distribution_for_restoring,
saved_dir=saved_dir,
predict_dataset=predict_dataset,
output_name=output_name)
self.assertAllClose(result_before_save, load_result, atol=_TOLERANCE)
|
tensorflow-master
|
tensorflow/python/distribute/saved_model_test_base.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""A component for running distributed TensorFlow."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import copy
import json
import os
import threading
import time
from tensorflow.core.protobuf import config_pb2
from tensorflow.python.client import session
from tensorflow.python.distribute import distribute_coordinator_context
from tensorflow.python.distribute import multi_worker_util
from tensorflow.python.platform import tf_logging as logging
from tensorflow.python.training import coordinator
from tensorflow.python.training import monitored_session
from tensorflow.python.training import server_lib
_thread_local = threading.local()
class _TaskType(object):
PS = "ps"
WORKER = "worker"
CHIEF = "chief"
EVALUATOR = "evaluator"
CLIENT = "client"
# TODO(yuefengz): support another mode where the client colocates with one
# worker.
class CoordinatorMode(object):
"""Specify how distribute coordinator runs."""
# The default mode where distribute coordinator will run as a standalone
# client and connects to remote servers for training. Each remote server can
# use the distribute coordinator binary with task_type set correctly which
# will then turn into standard servers.
STANDALONE_CLIENT = "standalone_client"
# The distribute coordinator runs on each worker. It will run a standard
# server on each worker and optionally run the `worker_fn` that is configured
# to talk to its standard server.
INDEPENDENT_WORKER = "independent_worker"
class _Barrier(object):
"""A reusable barrier class for worker synchronization."""
def __init__(self, num_participants):
"""Initializes the barrier object.
Args:
num_participants: an integer which is the expected number of calls of
`wait` pass to through this barrier.
"""
self._num_participants = num_participants
self._counter = 0
self._flag = False
self._local_sense = threading.local()
self._lock = threading.Lock()
self._condition = threading.Condition()
def wait(self):
"""Waits until all other callers reach the same wait call."""
self._local_sense.value = not self._flag
with self._lock:
self._counter += 1
if self._counter == self._num_participants:
self._counter = 0
self._flag = self._local_sense.value
with self._condition:
while self._flag != self._local_sense.value:
self._condition.wait()
self._condition.notify_all()
def _get_num_workers(cluster_spec):
"""Gets number of workers including chief."""
if not cluster_spec:
return 0
return len(cluster_spec.as_dict().get(_TaskType.WORKER, [])) + len(
cluster_spec.as_dict().get(_TaskType.CHIEF, []))
class _WorkerContext(object):
"""The worker context class.
This context object provides configuration information for each task. One
context manager with a worker context object will be created per
invocation to the `worker_fn` where `get_current_worker_context` can be called
to access the worker context object.
"""
def __init__(self,
strategy,
cluster_spec,
task_type,
task_id,
session_config=None,
rpc_layer="grpc",
worker_barrier=None):
"""Initialize the worker context object.
Args:
strategy: a `DistributionStrategy` object.
cluster_spec: a ClusterSpec object. It can be empty or None in the local
training case.
task_type: a string indicating the role of the corresponding task, such as
"worker" or "ps". It can be None if it is local training or in-graph
replicated training.
task_id: an integer indicating id of the corresponding task. It can be
None if it is local training or in-graph replicated training.
session_config: an optional `tf.compat.v1.ConfigProto` object.
rpc_layer: optional string specifying the RPC protocol for communication
with worker masters. If None or empty, hosts in the `cluster_spec` will
be used directly.
worker_barrier: optional, the barrier object for worker synchronization.
"""
self._strategy = strategy
self._cluster_spec = cluster_spec
self._task_type = task_type
self._task_id = task_id
self._session_config = session_config
self._worker_barrier = worker_barrier
self._rpc_layer = rpc_layer
self._master_target = self._get_master_target()
self._num_workers = _get_num_workers(cluster_spec)
self._is_chief_node = self._is_chief()
def _debug_message(self):
if self._cluster_spec:
return "[cluster_spec: %r, task_type: %r, task_id: %r]" % (
self._cluster_spec, self.task_type, self.task_id)
else:
return "[local]"
def __enter__(self):
old_context = distribute_coordinator_context.get_current_worker_context()
if old_context:
raise ValueError(
"You cannot run distribute coordinator in a `worker_fn`.\t" +
self._debug_message())
# pylint: disable=protected-access
distribute_coordinator_context._worker_context.current = self
def __exit__(self, unused_exception_type, unused_exception_value,
unused_traceback):
# pylint: disable=protected-access
distribute_coordinator_context._worker_context.current = None
def _get_master_target(self):
"""Return the master target for a task."""
# If cluster_spec is None or empty, we use local master.
if not self._cluster_spec:
return ""
# If task_type is None, then it is in-graph replicated training. In this
# case we use the chief or first worker's master target.
if not self._task_type:
if _TaskType.CHIEF in self._cluster_spec.jobs:
task_type = _TaskType.CHIEF
task_id = 0
else:
assert _TaskType.WORKER in self._cluster_spec.jobs
task_type = _TaskType.WORKER
task_id = 0
else:
task_type = self._task_type
task_id = self._task_id
prefix = ""
if self._rpc_layer:
prefix = self._rpc_layer + "://"
return prefix + self._cluster_spec.job_tasks(task_type)[task_id or 0]
def _is_chief(self):
"""Return whether the task is the chief worker."""
if (not self._cluster_spec or
self._task_type in [_TaskType.CHIEF, _TaskType.EVALUATOR, None]):
return True
# If not local and chief not in the cluster_spec, use the first worker as
# chief.
if (_TaskType.CHIEF not in self._cluster_spec.jobs and
self._task_type == _TaskType.WORKER and self._task_id == 0):
return True
return False
def wait_for_other_workers(self):
"""Waits for other workers to reach the same call to this method.
Raises:
ValueError: if `worker_barrier` is not passed to the __init__ method.
"""
if not self._worker_barrier:
# TODO(yuefengz): we should throw an error in independent worker mode.
return
self._worker_barrier.wait()
def session_creator(self,
scaffold=None,
config=None,
checkpoint_dir=None,
checkpoint_filename_with_path=None,
max_wait_secs=7200):
"""Returns a session creator.
The returned session creator will be configured with the correct master
target and session configs. It will also run either init ops or ready ops
by querying the `strategy` object when `create_session` is called on it.
Args:
scaffold: A `Scaffold` used for gathering or building supportive ops. If
not specified a default one is created. It's used to finalize the graph.
config: `ConfigProto` proto used to configure the session.
checkpoint_dir: A string. Optional path to a directory where to restore
variables.
checkpoint_filename_with_path: Full file name path to the checkpoint file.
Only one of `checkpoint_dir` or `checkpoint_filename_with_path` can be
specified.
max_wait_secs: Maximum time to wait for the session to become available.
Returns:
a descendant of SessionCreator.
"""
if config:
session_config = copy.deepcopy(config)
session_config.MergeFrom(self._session_config)
else:
session_config = self._session_config
if not self._strategy or self._strategy.extended.experimental_should_init:
logging.info("Creating chief session creator with config: %r", config)
return monitored_session.ChiefSessionCreator(
scaffold,
master=self.master_target,
config=session_config,
checkpoint_dir=checkpoint_dir,
checkpoint_filename_with_path=checkpoint_filename_with_path)
else:
logging.info("Creating worker session creator with config: %r", config)
return monitored_session.WorkerSessionCreator(
scaffold,
master=self.master_target,
config=session_config,
max_wait_secs=max_wait_secs)
@property
def session_config(self):
return copy.deepcopy(self._session_config)
@property
def has_barrier(self):
"""Whether the barrier is set or not."""
return self._worker_barrier is not None
@property
def distributed_mode(self):
"""Whether it is distributed training or not."""
return bool(self._cluster_spec) and self._task_type != _TaskType.EVALUATOR
@property
def cluster_spec(self):
"""Returns a copy of the cluster_spec object."""
return copy.deepcopy(self._cluster_spec)
@property
def task_type(self):
"""Returns the role of the corresponing task."""
return self._task_type
@property
def task_id(self):
"""Returns the id or index of the corresponing task."""
return self._task_id
@property
def master_target(self):
"""Returns the session master for the corresponding task to connect to."""
return self._master_target
@property
def is_chief(self):
"""Returns whether the task is a chief node."""
return self._is_chief_node
@property
def num_workers(self):
"""Returns number of workers in the cluster, including chief."""
return self._num_workers
@property
def experimental_should_init(self):
"""Whether to run init ops."""
return self._strategy.extended.experimental_should_init
@property
def should_checkpoint(self):
"""Whether to save checkpoint."""
return self._strategy.extended.should_checkpoint
@property
def should_save_summary(self):
"""Whether to save summaries."""
return self._strategy.extended.should_save_summary
def _run_single_worker(worker_fn,
strategy,
cluster_spec,
task_type,
task_id,
session_config,
rpc_layer="",
worker_barrier=None,
coord=None):
"""Runs a single worker by calling `worker_fn` under context."""
session_config = copy.deepcopy(session_config)
strategy = copy.deepcopy(strategy)
# If there is an EVALUATOR task, we run single-machine eval on that task.
if task_type == _TaskType.EVALUATOR:
# It is possible to not have a strategy object for EVALUATOR task.
if strategy:
strategy.configure(session_config)
else:
assert strategy
strategy.configure(session_config, cluster_spec, task_type, task_id)
context = _WorkerContext(
strategy,
cluster_spec,
task_type,
task_id,
session_config=session_config,
rpc_layer=rpc_layer,
worker_barrier=worker_barrier)
with context:
if coord:
with coord.stop_on_exception():
return worker_fn(strategy)
else:
return worker_fn(strategy)
def _split_cluster_for_evaluator(cluster_spec, task_type):
"""Split the cluster for evaluator since it needn't talk to other tasks."""
# Splitting the cluster is important to prevent the evaluator from talking to
# other tasks in the cluster. Since we allow evaluator not to use
# distribution strategies and as a result ops in the evalauator task may have
# unspecified devices. Those ops may end up on other tasks if we don't split
# the cluster.
# Note: if you bypass distribute coordinator and bring the cluster yourself,
# you can equivalently set device filters to split clusters. This is already
# done by distribution strategy's `update_config_proto` method.
new_cluster_spec = multi_worker_util.normalize_cluster_spec(
cluster_spec).as_dict()
if task_type == _TaskType.EVALUATOR:
assert _TaskType.EVALUATOR in new_cluster_spec
new_cluster_spec = {
_TaskType.EVALUATOR: new_cluster_spec[_TaskType.EVALUATOR]
}
else:
new_cluster_spec.pop(_TaskType.EVALUATOR, None)
return multi_worker_util.normalize_cluster_spec(new_cluster_spec)
def _run_std_server(cluster_spec=None,
task_type=None,
task_id=None,
session_config=None,
rpc_layer=None,
environment=None):
"""Runs a standard server."""
# Check if the Server is already running. If so, assert that no configuration
# options have changed, and return the existing Server. This allows us to
# call `run_distribute_coordinator` multiple times.
if getattr(_thread_local, "server", None) is not None:
assert _thread_local.cluster_spec == cluster_spec
assert _thread_local.task_type == task_type
assert _thread_local.task_id == task_id
assert _thread_local.session_config_str == repr(session_config)
assert _thread_local.rpc_layer == rpc_layer
assert _thread_local.environment == environment
return _thread_local.server
else:
# This method is not thread-safe.
_thread_local.server_started = True
_thread_local.cluster_spec = cluster_spec
_thread_local.task_type = task_type
_thread_local.task_id = task_id
_thread_local.session_config_str = repr(session_config)
_thread_local.rpc_layer = rpc_layer
_thread_local.environment = environment
assert cluster_spec
target = cluster_spec.task_address(task_type, task_id)
if rpc_layer:
target = rpc_layer + "://" + target
class _FakeServer(object):
"""A fake server that runs a master session."""
def start(self):
# A tensorflow server starts when a remote session is created.
logging.info(
"Creating a remote session to start a TensorFlow server, "
"target = %r, session_config=%r", target, session_config)
session.Session(target=target, config=session_config)
def join(self):
while True:
time.sleep(5)
if environment == "google":
server = _FakeServer()
else:
if session_config:
logging.info(
"Starting standard TensorFlow server, target = %r, session_config= "
"%r", target, session_config)
else:
logging.info("Starting standard TensorFlow server, target = %r", target)
cluster_spec = _split_cluster_for_evaluator(cluster_spec, task_type)
server = server_lib.Server(
cluster_spec,
job_name=task_type,
task_index=task_id,
config=session_config,
protocol=rpc_layer)
server.start()
_thread_local.server = server
return server
def _run_between_graph_client(worker_fn, strategy, eval_fn, eval_strategy,
cluster_spec, session_config, rpc_layer):
"""Runs a standalone client for between-graph replication."""
coord = coordinator.Coordinator()
eval_thread = None
if _TaskType.EVALUATOR in cluster_spec.jobs:
eval_thread = threading.Thread(
target=_run_single_worker,
args=(eval_fn, eval_strategy, cluster_spec, _TaskType.EVALUATOR, 0,
session_config),
kwargs={
"rpc_layer": rpc_layer,
"coord": coord,
})
eval_thread.start()
threads = []
worker_barrier = _Barrier(_get_num_workers(cluster_spec))
for task_type in [_TaskType.CHIEF, _TaskType.WORKER]:
for task_id in range(len(cluster_spec.as_dict().get(task_type, []))):
t = threading.Thread(
target=_run_single_worker,
args=(worker_fn, strategy, cluster_spec, task_type, task_id,
session_config),
kwargs={
"rpc_layer": rpc_layer,
"worker_barrier": worker_barrier,
"coord": coord,
})
t.start()
threads.append(t)
if eval_thread:
# TODO(yuefengz): is it necessary to join eval thread?
threads_to_join = threads + [eval_thread]
else:
threads_to_join = threads
coord.join(threads_to_join)
# TODO(yuefengz): we probably want to return results from all workers?
return None
def _run_in_graph_client(worker_fn, strategy, eval_fn, eval_strategy,
cluster_spec, session_config, rpc_layer):
"""Runs a standalone client for in-graph replication."""
coord = coordinator.Coordinator()
eval_thread = None
if _TaskType.EVALUATOR in cluster_spec.jobs:
eval_thread = threading.Thread(
target=_run_single_worker,
args=(eval_fn, eval_strategy, cluster_spec, _TaskType.EVALUATOR, 0,
session_config),
kwargs={
"rpc_layer": rpc_layer,
"coord": coord,
})
eval_thread.start()
worker_result = _run_single_worker(
worker_fn,
strategy,
cluster_spec,
None,
None,
session_config,
rpc_layer=rpc_layer,
coord=coord)
if eval_thread:
coord.join([eval_thread])
return worker_result
def _configure_session_config_for_std_servers(
strategy, eval_strategy, session_config, cluster_spec, task_type, task_id):
# pylint: disable=g-doc-args
"""Call strategy's `configure` to mutate the session_config.
The session_config is currently needed as default config for a TensorFlow
server. In the future, we should be able to remove this method and only pass
the session config to a client session.
"""
if task_type == _TaskType.EVALUATOR:
if eval_strategy:
eval_strategy.configure(session_config=session_config)
else:
# The strategy may be shared in standalone client mode.
strategy = copy.deepcopy(strategy)
strategy.configure(
session_config=session_config,
cluster_spec=cluster_spec,
task_type=task_type,
task_id=task_id)
# Remove the device filters specific to the strategy, so that the
# TensorFlow server brought up with one strategy can be used by other
# strategies. The device filters can be set in the client side as well.
del session_config.device_filters[:]
def run_standard_tensorflow_server(session_config=None):
"""Starts a standard TensorFlow server.
This method parses configurations from "TF_CONFIG" environment variable and
starts a TensorFlow server. The "TF_CONFIG" is typically a json string and
must have information of the cluster and the role of the server in the
cluster. One example is:
TF_CONFIG='{
"cluster": {
"worker": ["host1:2222", "host2:2222", "host3:2222"],
"ps": ["host4:2222", "host5:2222"]
},
"task": {"type": "worker", "index": 1}
}'
This "TF_CONFIG" specifies there are 3 workers and 2 ps tasks in the cluster
and the current role is worker 1.
Valid task types are "chief", "worker", "ps" and "evaluator" and you can have
at most one "chief" and at most one "evaluator".
An optional key-value can be specified is "rpc_layer". The default value is
"grpc".
Args:
session_config: an optional `tf.compat.v1.ConfigProto` object. Users can
pass in the session config object to configure server-local devices.
Returns:
a `tf.distribute.Server` object which has already been started.
Raises:
ValueError: if the "TF_CONFIG" environment is not complete.
"""
tf_config = json.loads(os.environ.get("TF_CONFIG", "{}"))
if "cluster" not in tf_config:
raise ValueError("\"cluster\" is not found in TF_CONFIG.")
cluster_spec = multi_worker_util.normalize_cluster_spec(tf_config["cluster"])
if "task" not in tf_config:
raise ValueError("\"task\" is not found in TF_CONFIG.")
task_env = tf_config["task"]
if "type" not in task_env:
raise ValueError(
"\"task_type\" is not found in the `task` part of TF_CONFIG.")
task_type = task_env["type"]
task_id = int(task_env.get("index", 0))
rpc_layer = tf_config.get("rpc_layer", "grpc")
session_config = session_config or config_pb2.ConfigProto()
# Set the collective group leader for collective ops to initialize collective
# ops when server starts.
if "chief" in cluster_spec.jobs:
session_config.experimental.collective_group_leader = (
"/job:chief/replica:0/task:0")
else:
if "worker" not in cluster_spec.jobs:
raise ValueError(
"You must have `chief` or `worker` jobs in the `cluster_spec`.")
session_config.experimental.collective_group_leader = (
"/job:worker/replica:0/task:0")
server = _run_std_server(
cluster_spec=cluster_spec,
task_type=task_type,
task_id=task_id,
session_config=session_config,
rpc_layer=rpc_layer)
server.start()
return server
# TODO(yuefengz): propagate cluster_spec in the STANDALONE_CLIENT mode.
# TODO(yuefengz): we may need a smart way to figure out whether the current task
# is the special task when we support cluster_spec propagation.
def run_distribute_coordinator(worker_fn,
strategy,
eval_fn=None,
eval_strategy=None,
mode=CoordinatorMode.STANDALONE_CLIENT,
cluster_spec=None,
task_type=None,
task_id=None,
session_config=None,
rpc_layer="grpc"):
"""Runs the coordinator for distributed TensorFlow.
This function runs a split coordinator for distributed TensorFlow in its
default mode, i.e the STANDALONE_CLIENT mode. Given a `cluster_spec`
specifying server addresses and their roles in a cluster, this coordinator
will figure out how to set them up, give the underlying function the right
targets for master sessions via a scope object and coordinate their training.
The cluster consisting of standard servers needs to be brought up either with
the standard server binary or with a binary running distribute coordinator
with `task_type` set to non-client type which will then turn into standard
servers.
In addition to be the distribute coordinator, this is also the source of
configurations for each job in the distributed training. As there are multiple
ways to configure a distributed TensorFlow cluster, its context object
provides these configurations so that users or higher-level APIs don't have to
figure out the configuration for each job by themselves.
In the between-graph replicated training, this coordinator will create
multiple threads and each calls the `worker_fn` which is supposed to create
its own graph and connect to one worker master given by its context object. In
the in-graph replicated training, it has only one thread calling this
`worker_fn`.
Another mode is the INDEPENDENT_WORKER mode where each server runs a
distribute coordinator which will start a standard server and optionally runs
`worker_fn` depending whether it is between-graph training or in-graph
replicated training.
The `strategy` object is expected to be a DistributionStrategy object which
has implemented methods needed by distributed coordinator such as
`configure(session_config, cluster_spec, task_type, task_id)` which configures
the strategy object for a specific task and `experimental_should_init`
property which instructs the distribute coordinator whether to run init ops
for a task. The distribute coordinator will make a copy of the `strategy`
object, call its `configure` method and pass it to `worker_fn` as an argument.
The `worker_fn` defines the training logic and is called under its own
worker context which can be accessed to via `get_current_worker_context`. A
worker context provides access to configurations for each task, e.g. the
task_type, task_id, master target and so on. Since `worker_fn` will be called
in a thread and possibly multiple times, caller should be careful when it
accesses global data. For example, it is unsafe to define flags in a
`worker_fn` or to define different environment variables for different
`worker_fn`s.
The `worker_fn` for the between-graph replication is defined as if there is
only one worker corresponding to the `worker_fn` and possibly ps jobs. For
example, when training with parameter servers, it assigns variables to
parameter servers and all other operations to that worker. In the in-graph
replication case, the `worker_fn` has to define operations for all worker
jobs. Using a distribution strategy can simplify the `worker_fn` by not having
to worry about the replication and device assignment of variables and
operations.
This method is intended to be invoked by high-level APIs so that users don't
have to explictly call it to run this coordinator. For those who don't use
high-level APIs, to change a program to use this coordinator, wrap everything
in a the program after global data definitions such as commandline flag
definition into the `worker_fn` and get task-specific configurations from
the worker context.
The `cluster_spec` can be either passed by the argument or parsed from the
"TF_CONFIG" environment variable. Example of a TF_CONFIG:
```
cluster = {'chief': ['host0:2222'],
'ps': ['host1:2222', 'host2:2222'],
'worker': ['host3:2222', 'host4:2222', 'host5:2222']}
os.environ['TF_CONFIG'] = json.dumps({'cluster': cluster})
```
If `cluster_spec` is not given in any format, it becomes local training and
this coordinator will connect to a local session.
For evaluation, if "evaluator" exists in the cluster_spec, a separate thread
will be created to call `eval_fn` with its `task_type` set to "evaluator". If
`eval_fn` is not defined, fall back to `worker_fn`. This implies that
evaluation will be done on a single machine if there is an "evaluator" task.
If "evaluator" doesn't exist in the cluster_spec, it entirely depends on the
`worker_fn` for how to do evaluation.
Args:
worker_fn: the function to be called. The function should accept a
`strategy` object and will be given access to a context object via a
context manager scope.
strategy: a DistributionStrategy object specifying whether it should
run between-graph replicated training or not, whether to run init ops,
etc. This object will also be configured given `session_config`,
`cluster_spec`, `task_type` and `task_id`.
eval_fn: optional function for "evaluator" task. If `eval_fn` is not passed
in but a "evaluator" task is found in the `cluster_spec`, the `worker_fn`
will be used for this task.
eval_strategy: optional DistributionStrategy object for "evaluator" task.
mode: in which mode this distribute coordinator runs.
cluster_spec: a dict, ClusterDef or ClusterSpec specifying servers and roles
in a cluster. If not set or empty, fall back to local training.
task_type: the current task type, optional if this is a client.
task_id: the current task id, optional if this is a client.
session_config: an optional `tf.compat.v1.ConfigProto` object which will be
passed to `strategy`'s `configure` method and used to create a session.
rpc_layer: optional string, the protocol for RPC, e.g. "grpc".
Raises:
ValueError: if `cluster_spec` is supplied but not a dict or a ClusterDef or
a ClusterSpec.
Returns:
In the client job, return the value returned by `worker_fn` if
it is in-graph replication or INDEPENDENT_WORKER mode; return None
otherwise.
"""
tf_config = json.loads(os.environ.get("TF_CONFIG", "{}"))
if not cluster_spec:
cluster_spec = tf_config.get("cluster", {})
task_env = tf_config.get("task", {})
if task_env:
task_type = task_env.get("type", task_type)
task_id = int(task_env.get("index", task_id))
if cluster_spec:
cluster_spec = multi_worker_util.normalize_cluster_spec(cluster_spec)
# TODO(yuefengz): validate cluster_spec.
rpc_layer = tf_config.get("rpc_layer", rpc_layer)
environment = tf_config.get("environment", None)
# Setting the session config is necessary for some strategies such as
# CollectiveAllReduceStrategy.
session_config = session_config or config_pb2.ConfigProto(
allow_soft_placement=True)
if cluster_spec:
logging.info(
"Running Distribute Coordinator with mode = %r, cluster_spec = %r, "
"task_type = %r, task_id = %r, environment = %r, rpc_layer = %r", mode,
cluster_spec.as_dict(), task_type, task_id, environment, rpc_layer)
if not cluster_spec:
# `mode` is ignored in the local case.
logging.info("Running local Distribute Coordinator.")
_run_single_worker(worker_fn, strategy, None, None, None, session_config,
rpc_layer)
if eval_fn:
_run_single_worker(eval_fn, eval_strategy, None, None, None,
session_config, rpc_layer)
else:
logging.warning("Skipped evaluation since `eval_fn` is not passed in.")
elif mode == CoordinatorMode.STANDALONE_CLIENT:
if not eval_fn:
logging.warning("`eval_fn` is not passed in. The `worker_fn` will be "
"used if an \"evaluator\" task exists in the cluster.")
eval_fn = eval_fn or worker_fn
if not eval_strategy:
logging.warning("`eval_strategy` is not passed in. No distribution "
"strategy will be used for evaluation.")
# The client must know the cluster but servers in the cluster don't have to
# know the client.
if task_type in [_TaskType.CLIENT, None]:
if strategy.extended.experimental_between_graph:
return _run_between_graph_client(worker_fn, strategy, eval_fn,
eval_strategy, cluster_spec,
session_config, rpc_layer)
else:
return _run_in_graph_client(worker_fn, strategy, eval_fn, eval_strategy,
cluster_spec, session_config, rpc_layer)
else:
# If not a client job, run the standard server.
_configure_session_config_for_std_servers(strategy, eval_strategy,
session_config, cluster_spec,
task_type, task_id)
server = _run_std_server(
cluster_spec=cluster_spec,
task_type=task_type,
task_id=task_id,
session_config=session_config,
rpc_layer=rpc_layer,
environment=environment)
server.join()
else:
if mode != CoordinatorMode.INDEPENDENT_WORKER:
raise ValueError("Unexpected coordinator mode: %r" % mode)
if not eval_fn:
logging.warning("`eval_fn` is not passed in. The `worker_fn` will be "
"used if an \"evaluator\" task exists in the cluster.")
eval_fn = eval_fn or worker_fn
if not eval_strategy:
logging.warning("`eval_strategy` is not passed in. No distribution "
"strategy will be used for evaluation.")
# Every one starts a standard server, get session config from `configure`
# method.
_configure_session_config_for_std_servers(strategy, eval_strategy,
session_config, cluster_spec,
task_type, task_id)
if not getattr(strategy.extended, "_std_server_started", False):
# Right now, with eager mode, context is configured with a std server at
# the very beginning while with graph mode the std server is started when
# distribute coordinator is called. We should consolidate these two paths.
server = _run_std_server(
cluster_spec=cluster_spec,
task_type=task_type,
task_id=task_id,
session_config=session_config,
rpc_layer=rpc_layer,
environment=environment)
if task_type in [_TaskType.CHIEF, _TaskType.WORKER]:
if strategy.extended.experimental_between_graph:
# All jobs run `worker_fn` if between-graph.
return _run_single_worker(worker_fn, strategy, cluster_spec, task_type,
task_id, session_config, rpc_layer)
else:
# Only one node runs `worker_fn` if in-graph.
context = _WorkerContext(strategy, cluster_spec, task_type, task_id)
if context.is_chief:
return _run_single_worker(worker_fn, strategy, cluster_spec, None,
None, session_config, rpc_layer)
else:
server.join()
elif task_type == _TaskType.EVALUATOR:
return _run_single_worker(eval_fn, eval_strategy, cluster_spec, task_type,
task_id, session_config, rpc_layer)
else:
if task_type != _TaskType.PS:
raise ValueError("Unexpected task_type: %r" % task_type)
server.join()
|
tensorflow-master
|
tensorflow/python/distribute/distribute_coordinator.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""A simple network to use in tests and examples."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.data.ops import dataset_ops
from tensorflow.python.distribute import step_fn
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import ops
from tensorflow.python.keras.optimizer_v2 import optimizer_v2
from tensorflow.python.layers import core
from tensorflow.python.layers import normalization
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import math_ops
def single_loss_example(optimizer_fn, distribution, use_bias=False,
iterations_per_step=1):
"""Build a very simple network to use in tests and examples."""
def dataset_fn():
return dataset_ops.Dataset.from_tensors([[1.]]).repeat()
optimizer = optimizer_fn()
layer = core.Dense(1, use_bias=use_bias)
def loss_fn(ctx, x):
del ctx
y = array_ops.reshape(layer(x), []) - constant_op.constant(1.)
return y * y
single_loss_step = step_fn.StandardSingleLossStep(
dataset_fn, loss_fn, optimizer, distribution, iterations_per_step)
# Layer is returned for inspecting the kernels in tests.
return single_loss_step, layer
def minimize_loss_example(optimizer, use_bias=False, use_callable_loss=True):
"""Example of non-distribution-aware legacy code."""
def dataset_fn():
dataset = dataset_ops.Dataset.from_tensors([[1.]]).repeat()
# TODO(isaprykin): batch with drop_remainder causes shapes to be
# fully defined for TPU. Remove this when XLA supports dynamic shapes.
return dataset.batch(1, drop_remainder=True)
layer = core.Dense(1, use_bias=use_bias)
def model_fn(x):
"""A very simple model written by the user."""
def loss_fn():
y = array_ops.reshape(layer(x), []) - constant_op.constant(1.)
return y * y
if isinstance(optimizer, optimizer_v2.OptimizerV2):
return optimizer.minimize(loss_fn, lambda: layer.trainable_variables)
elif use_callable_loss:
return optimizer.minimize(loss_fn)
else:
return optimizer.minimize(loss_fn())
return model_fn, dataset_fn, layer
def batchnorm_example(optimizer_fn,
batch_per_epoch=1,
momentum=0.9,
renorm=False,
update_ops_in_replica_mode=False):
"""Example of non-distribution-aware legacy code with batch normalization."""
def dataset_fn():
# input shape is [16, 8], input values are increasing in both dimensions.
return dataset_ops.Dataset.from_tensor_slices(
[[[float(x * 8 + y + z * 100)
for y in range(8)]
for x in range(16)]
for z in range(batch_per_epoch)]).repeat()
optimizer = optimizer_fn()
batchnorm = normalization.BatchNormalization(
renorm=renorm, momentum=momentum, fused=False)
layer = core.Dense(1, use_bias=False)
def model_fn(x):
"""A model that uses batchnorm."""
def loss_fn():
y = batchnorm(x, training=True)
with ops.control_dependencies(
ops.get_collection(ops.GraphKeys.UPDATE_OPS)
if update_ops_in_replica_mode else []):
loss = math_ops.reduce_mean(
math_ops.reduce_sum(layer(y)) - constant_op.constant(1.))
# `x` and `y` will be fetched by the gradient computation, but not `loss`.
return loss
if isinstance(optimizer, optimizer_v2.OptimizerV2):
return optimizer.minimize(loss_fn, lambda: layer.trainable_variables)
# Callable loss.
return optimizer.minimize(loss_fn)
return model_fn, dataset_fn, batchnorm
|
tensorflow-master
|
tensorflow/python/distribute/single_loss_example.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Test DistributionStrategy, ReplicaContext, and supporting APIs."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.data.ops import dataset_ops
from tensorflow.python.distribute import distribute_lib
from tensorflow.python.distribute import distribution_strategy_context as ds_context
from tensorflow.python.distribute import input_lib
from tensorflow.python.distribute import reduce_util
from tensorflow.python.distribute import values
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.ops import variable_scope
from tensorflow.python.ops import variables
from tensorflow.python.platform import test
from tensorflow.python.util import nest
class _TestReplicaContext(distribute_lib.ReplicaContext):
def merge_call(self, fn, *args, **kwargs):
return kwargs["test_arg"]
def _get_test_variable(name, synchronization, aggregation):
return {
"name": name,
"synchronization": synchronization,
"aggregation": aggregation
}
def _test_input_fn(input_context):
del input_context
return dataset_ops.DatasetV2.from_tensors(1.).repeat()
class _TestStrategy(distribute_lib.Strategy):
def __init__(self):
super(_TestStrategy, self).__init__(_TestExtended(self))
class _TestExtended(distribute_lib.StrategyExtendedV1):
def __init__(self, distribute):
super(_TestExtended, self).__init__(distribute)
device_map = values.ReplicaDeviceMap(["/device:CPU:0"])
worker_device_pairs = [("", ["/device:CPU:0"])]
self._input_workers = input_lib.InputWorkers(device_map,
worker_device_pairs)
def _call_for_each_replica(self, fn, args, kwargs):
with _TestReplicaContext(
self._container_strategy(),
replica_id_in_sync_group=constant_op.constant(0, dtypes.int32)):
return fn(*args, **kwargs)
def _create_variable(self, next_creator, *args, **kwargs):
return _get_test_variable(kwargs["name"], kwargs["synchronization"],
kwargs["aggregation"])
def _make_input_fn_iterator(
self,
input_fn,
replication_mode=distribute_lib.InputReplicationMode.PER_WORKER):
return input_lib.InputFunctionIterator(input_fn, self._input_workers,
[distribute_lib.InputContext()],
self._container_strategy())
def _local_results(self, value):
return (value,)
def _reduce_to(self, reduce_op, value, destinations):
del reduce_op, destinations
return value
def _experimental_make_numpy_dataset(self, numpy_input, session):
del session
return dataset_ops.DatasetV2.from_tensor_slices(numpy_input)
def _experimental_run_steps_on_iterator(self, fn, iterator, iterations,
initial_loop_values=None):
# TODO(tomhennigan) This is missing many things (e.g. ctx.run_op).
ctx = input_lib.MultiStepContext()
for _ in range(iterations):
fn(ctx, iterator.get_next())
return ctx
def _update(self, var, fn, args, kwargs, group):
# The implementations of _update() and _update_non_slot() are identical
# except _update() passes `var` as the first argument to `fn()`.
return self._update_non_slot(var, fn, (var,) + tuple(args), kwargs, group)
def _update_non_slot(self, colocate_with, fn, args, kwargs, group):
del colocate_with
result = fn(*args, **kwargs)
if group:
return result
else:
return nest.map_structure(self._unwrap, result)
def _assert_in_default_state(t):
t.assertIs(ds_context._get_default_replica_context(),
ds_context.get_replica_context())
t.assertIs(None, ds_context.get_cross_replica_context())
t.assertFalse(ds_context.in_cross_replica_context())
t.assertIs(ds_context._get_default_strategy(), ds_context.get_strategy())
t.assertFalse(ds_context.has_strategy())
def _run_in_and_out_of_scope(unbound_test_method):
def wrapper(test_case):
dist = _TestStrategy()
# Running in the default (replica) scope should be supported.
_assert_in_default_state(test_case)
unbound_test_method(test_case, dist)
# As well as running in the strategy scope.
with dist.scope():
unbound_test_method(test_case, dist)
_assert_in_default_state(test_case)
# When run under a different strategy the test method should fail.
another_strategy = _TestStrategy()
msg = "Mixing different .*Strategy objects"
with test_case.assertRaisesRegexp(RuntimeError, msg):
with another_strategy.scope():
unbound_test_method(test_case, dist)
return wrapper
class TestStrategyTest(test.TestCase):
def testCallForEachReplica(self):
_assert_in_default_state(self)
dist = _TestStrategy()
def run_fn():
replica_context = ds_context.get_replica_context()
self.assertTrue(replica_context is not None)
self.assertIs(None, ds_context.get_cross_replica_context())
self.assertFalse(ds_context.in_cross_replica_context())
self.assertTrue(ds_context.has_strategy())
self.assertIs(dist, ds_context.get_strategy())
self.assertEqual("foo", replica_context.merge_call(None, test_arg="foo"))
expected_value = _get_test_variable(
"bar", variable_scope.VariableSynchronization.AUTO,
variable_scope.VariableAggregation.NONE)
self.assertDictEqual(expected_value,
variable_scope.variable(1.0, name="bar"))
dist.extended.call_for_each_replica(run_fn)
with dist.scope():
dist.extended.call_for_each_replica(run_fn)
_assert_in_default_state(self)
def testScope(self):
_assert_in_default_state(self)
dist = _TestStrategy()
with dist.scope():
self.assertIs(None, ds_context.get_replica_context())
self.assertIs(dist, ds_context.get_cross_replica_context())
self.assertTrue(ds_context.in_cross_replica_context())
self.assertTrue(ds_context.has_strategy())
self.assertIs(dist, ds_context.get_strategy())
expected_value = _get_test_variable(
"baz", variable_scope.VariableSynchronization.AUTO,
variable_scope.VariableAggregation.NONE)
self.assertDictEqual(expected_value,
variable_scope.variable(1.0, name="baz"))
_assert_in_default_state(self)
def testScopeDeviceNestingError(self):
_assert_in_default_state(self)
dist = _TestStrategy()
# Open a device scope with dist.scope().
dist.extended._default_device = "/device:GPU:0"
scope = dist.scope()
scope.__enter__()
self.assertIs(dist, ds_context.get_strategy())
with ops.device("/device:CPU:0"):
with self.assertRaisesRegexp(RuntimeError, "Device scope nesting error"):
scope.__exit__(None, None, None)
scope.__exit__(None, None, None)
_assert_in_default_state(self)
def testScopeVarCreatorNestingError(self):
def creator(next_creator, **kwargs):
return next_creator(**kwargs)
_assert_in_default_state(self)
dist = _TestStrategy()
scope = dist.scope()
scope.__enter__()
self.assertIs(dist, ds_context.get_strategy())
with variable_scope.variable_creator_scope(creator):
with self.assertRaisesRegexp(RuntimeError,
"Variable creator scope nesting error"):
scope.__exit__(None, None, None)
scope.__exit__(None, None, None)
_assert_in_default_state(self)
def testScopeVarScopeNestingError(self):
# We create a new graph here to simplify clean-up, since the error
# we are triggering happens in the middle of scope.__exit__() and
# leaves us in a weird state.
with ops.Graph().as_default():
_assert_in_default_state(self)
dist = _TestStrategy()
scope = dist.scope()
scope.__enter__()
self.assertIs(dist, ds_context.get_strategy())
with variable_scope.variable_scope("AA"):
with self.assertRaisesRegexp(RuntimeError,
"Variable scope nesting error"):
scope.__exit__(None, None, None)
_assert_in_default_state(self)
def testSettingSynchronizationAndAggregation(self):
_assert_in_default_state(self)
dist = _TestStrategy()
with dist.scope():
expected_value = _get_test_variable(
"baz", variable_scope.VariableSynchronization.ON_WRITE,
variable_scope.VariableAggregation.MEAN)
self.assertDictEqual(
expected_value,
variable_scope.variable(
1.0,
name="baz",
synchronization=variable_scope.VariableSynchronization.ON_WRITE,
aggregation=variable_scope.VariableAggregation.MEAN))
_assert_in_default_state(self)
def testSetStrategy(self):
_assert_in_default_state(self)
dist = _TestStrategy()
dist2 = _TestStrategy()
ds_context.experimental_set_strategy(dist)
self.assertIs(None, ds_context.get_replica_context())
self.assertIs(dist, ds_context.get_cross_replica_context())
self.assertTrue(ds_context.in_cross_replica_context())
self.assertTrue(ds_context.has_strategy())
self.assertIs(dist, ds_context.get_strategy())
expected_value = _get_test_variable(
"baz", variable_scope.VariableSynchronization.AUTO,
variable_scope.VariableAggregation.NONE)
self.assertDictEqual(expected_value,
variable_scope.variable(1.0, name="baz"))
ds_context.experimental_set_strategy(dist2)
self.assertIs(dist2, ds_context.get_strategy())
ds_context.experimental_set_strategy(None)
_assert_in_default_state(self)
def testSetStrategyInScope(self):
_assert_in_default_state(self)
dist = _TestStrategy()
with dist.scope():
with self.assertRaisesRegexp(
RuntimeError,
"Must not be called inside a `tf.distribute.Strategy` scope"):
ds_context.experimental_set_strategy(_TestStrategy())
with self.assertRaisesRegexp(
RuntimeError,
"Must not be called inside a `tf.distribute.Strategy` scope"):
ds_context.experimental_set_strategy(dist)
with self.assertRaisesRegexp(
RuntimeError,
"Must not be called inside a `tf.distribute.Strategy` scope"):
ds_context.experimental_set_strategy(None)
_assert_in_default_state(self)
def testSameScopeNesting(self):
_assert_in_default_state(self)
dist = _TestStrategy()
scope_a = dist.scope()
with scope_a:
self.assertIs(dist, ds_context.get_strategy())
scope_b = dist.scope()
with scope_b:
self.assertIs(dist, ds_context.get_strategy())
with scope_a:
self.assertIs(dist, ds_context.get_strategy())
self.assertIs(dist, ds_context.get_strategy())
self.assertIs(dist, ds_context.get_strategy())
dist2 = _TestStrategy()
scope2 = dist2.scope()
with self.assertRaisesRegexp(
RuntimeError,
"Mixing different tf.distribute.Strategy objects"):
with scope2:
pass
_assert_in_default_state(self)
with scope_b:
self.assertIs(dist, ds_context.get_strategy())
_assert_in_default_state(self)
@_run_in_and_out_of_scope
def testMakeInputFnIterator(self, dist):
self.assertIsNotNone(dist.make_input_fn_iterator(_test_input_fn))
@_run_in_and_out_of_scope
def testReduce(self, dist):
x = constant_op.constant(1.)
x_r = dist.reduce(reduce_util.ReduceOp.MEAN, x, axis=None)
self.assertEqual(self.evaluate(x), self.evaluate(x_r))
def testReductions_acceptStringOps(self):
dist = _TestStrategy()
for op in ("mean", "MEAN", "sum", "SUM"):
x = constant_op.constant(1.)
y = constant_op.constant(1.)
x_r = dist.reduce(op, x, axis=None)
self.assertEqual(self.evaluate(x), self.evaluate(x_r))
x_r = dist.extended.reduce_to(op, x, "/CPU:0")
self.assertEqual(self.evaluate(x), self.evaluate(x_r))
x_r, y_r = dist.extended.batch_reduce_to(op,
((x, "/CPU:0"), (y, "/CPU:0")))
self.assertEqual(self.evaluate(x), self.evaluate(x_r))
self.assertEqual(self.evaluate(y), self.evaluate(y_r))
@_run_in_and_out_of_scope
def testExperimentalMakeNumpyDataset(self, dist):
numpy_input = np.ones([10], dtype=np.float32)
dataset = dist.experimental_make_numpy_dataset(numpy_input)
self.assertEqual(
self.evaluate(dataset.reduce(0., lambda a, b: a + b)), 10.)
@_run_in_and_out_of_scope
def testExperimentalRunStepsOnIterator(self, dist):
all_inputs = []
dataset = dataset_ops.Dataset.from_tensors(1.).repeat()
dist.extended.experimental_run_steps_on_iterator(
lambda _, inputs: all_inputs.append(self.evaluate(inputs)),
dataset.make_one_shot_iterator())
self.assertEqual(all_inputs, [1.])
@_run_in_and_out_of_scope
def testReduceTo(self, dist):
x = constant_op.constant(1.)
x_r = dist.extended.reduce_to(reduce_util.ReduceOp.MEAN, x, "/CPU:0")
self.assertEqual(self.evaluate(x), self.evaluate(x_r))
@_run_in_and_out_of_scope
def testBatchReduceTo(self, dist):
x = constant_op.constant(1.)
y = constant_op.constant(1.)
x_r, y_r = dist.extended.batch_reduce_to(reduce_util.ReduceOp.MEAN,
((x, "/CPU:0"), (y, "/CPU:0")))
self.assertEqual(self.evaluate(x), self.evaluate(x_r))
self.assertEqual(self.evaluate(y), self.evaluate(y_r))
@_run_in_and_out_of_scope
def testUpdate(self, dist):
with dist.scope():
v = variables.Variable(1.)
t = constant_op.constant(2.)
def assign_fn(vv, tt):
self.assertIs(vv, v)
self.assertIs(tt, t)
dist.extended.update(v, assign_fn, (t,))
@_run_in_and_out_of_scope
def testUpdateNonSlot(self, dist):
t = constant_op.constant(2.)
update_calls = []
dist.extended.update_non_slot(t, lambda: update_calls.append(1))
self.assertEqual(len(update_calls), 1)
# _TestStrategy2 is like _TestStrategy, except it doesn't change variable
# creation.
class _TestStrategy2(distribute_lib.Strategy):
def __init__(self):
super(_TestStrategy2, self).__init__(_TestExtended2(self))
class _TestExtended2(_TestExtended):
def _create_variable(self, next_creator, *args, **kwargs):
return next_creator(*args, **kwargs)
class DefaultDistributionStrategyTest(test.TestCase):
def testMergeCall(self):
_assert_in_default_state(self)
def merge_fn(dist, s):
self.assertIs(ds_context._get_default_strategy(), dist)
self.assertIs(None, ds_context.get_replica_context())
self.assertIs(dist, ds_context.get_cross_replica_context())
self.assertTrue(ds_context.in_cross_replica_context())
self.assertIs(dist, ds_context.get_strategy())
self.assertFalse(ds_context.has_strategy())
return "foo_" + s
replica_ctx = ds_context.get_replica_context()
self.assertIs(ds_context._get_default_replica_context(), replica_ctx)
self.assertEqual("foo_bar", replica_ctx.merge_call(merge_fn, args=("bar",)))
_assert_in_default_state(self)
def testScopeMostlyNoOp(self):
_assert_in_default_state(self)
test_strategy = _TestStrategy2()
with test_strategy.scope():
variable_scope.variable(1.0, name="before")
default_strategy = ds_context._get_default_strategy()
scope = default_strategy.scope()
with scope:
_assert_in_default_state(self)
with test_strategy.scope():
with self.assertRaisesRegexp(
RuntimeError, "Mixing different tf.distribute.Strategy objects"):
variable_scope.variable(1.0, name="error")
with scope:
_assert_in_default_state(self)
with test_strategy.scope():
with self.assertRaisesRegexp(
RuntimeError, "Mixing different tf.distribute.Strategy objects"):
variable_scope.variable(1.0, name="also_error")
_assert_in_default_state(self)
_assert_in_default_state(self)
with test_strategy.scope():
variable_scope.variable(1.0, name="after")
def testExperimentalRunV2(self):
default_strategy = ds_context._get_default_strategy()
dataset = dataset_ops.Dataset.range(10).batch(2)
iterator = default_strategy.extended._make_dataset_iterator(dataset)
next_val = iterator.get_next()
def train_step(input_data):
return input_data
for _ in range(2):
default_strategy.experimental_run_v2(train_step, args=(next_val,))
class InputContextTest(test.TestCase):
def testProperties(self):
input_context = distribute_lib.InputContext(
num_input_pipelines=2, input_pipeline_id=1, num_replicas_in_sync=6)
self.assertEqual(6, input_context.num_replicas_in_sync)
self.assertEqual(1, input_context.input_pipeline_id)
self.assertEqual(2, input_context.num_input_pipelines)
def testPerReplicaBatchSize(self):
input_context = distribute_lib.InputContext(
num_input_pipelines=2, input_pipeline_id=1, num_replicas_in_sync=6)
self.assertEqual(2, input_context.get_per_replica_batch_size(12))
with self.assertRaises(ValueError):
input_context.get_per_replica_batch_size(13)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/distribute_lib_test.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for warm_starting_util with Distribution Strategy.
These tests are located here instead of as part of `WarmStartingUtilTest`
because they need access to distribution strategies which are only present in
contrib right now.
TODO(priyag): Move the tests to core `WarmStartingUtilTest` when distribution
strategy moves out of contrib.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
from absl.testing import parameterized
from tensorflow.python.distribute import combinations
from tensorflow.python.distribute import strategy_combinations
from tensorflow.python.framework import ops
from tensorflow.python.ops import variable_scope
from tensorflow.python.ops import variables
from tensorflow.python.platform import test
from tensorflow.python.training import saver as saver_lib
from tensorflow.python.training import warm_starting_util as ws_util
class WarmStartingUtilWithDistributionStrategyTest(
test.TestCase, parameterized.TestCase):
@combinations.generate(
combinations.combine(
distribution=[
strategy_combinations.default_strategy,
strategy_combinations.one_device_strategy,
strategy_combinations.mirrored_strategy_with_gpu_and_cpu,
strategy_combinations.mirrored_strategy_with_two_gpus,
],
save_with_distribution=[True, False],
restore_with_distribution=[True, False],
mode=["graph"]))
def testWarmStart(self, distribution, save_with_distribution,
restore_with_distribution):
var_name = "v"
original_value = [[1., 2.], [3., 4.]]
# Create variable and save checkpoint from which to warm-start.
def create_var(g):
with self.session(graph=g) as sess:
var = variable_scope.get_variable(var_name, initializer=original_value)
sess.run(variables.global_variables_initializer())
saver = saver_lib.Saver()
ckpt_prefix = os.path.join(self.get_temp_dir(), "model")
saver.save(sess, ckpt_prefix, global_step=0)
return var, sess.run(var)
if save_with_distribution:
with ops.Graph().as_default() as g, distribution.scope():
_, prev_init_val = create_var(g)
else:
with ops.Graph().as_default() as g:
_, prev_init_val = create_var(g)
# Verify we initialized the values correctly.
self.assertAllEqual(original_value, prev_init_val)
def warm_start(g):
with self.session(graph=g) as sess:
# Initialize with zeros.
var = variable_scope.get_variable(
var_name, initializer=[[0., 0.], [0., 0.]])
ws_util.warm_start(self.get_temp_dir())
sess.run(variables.global_variables_initializer())
# Verify weights were correctly warm-started to previous values.
self.assertAllEqual(original_value, self.evaluate(var))
# Warm start in a new graph.
if restore_with_distribution:
with ops.Graph().as_default() as g, distribution.scope():
warm_start(g)
else:
with ops.Graph().as_default() as g:
warm_start(g)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/warm_starting_util_test.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Classes for different algorithms of reduction and broadcasting."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import enum
import six
from tensorflow.python.client import device_lib
from tensorflow.python.distribute import cross_device_utils
from tensorflow.python.distribute import device_util
from tensorflow.python.distribute import reduce_util
from tensorflow.python.distribute import values as value_lib
from tensorflow.python.eager import context
from tensorflow.python.framework import kernels
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import resource_variable_ops
from tensorflow.python.platform import tf_logging as logging
from tensorflow.python.util.tf_export import tf_export
from tensorflow.tools.docs import doc_controls
def check_destinations(destinations):
"""Checks whether `destinations` is not empty.
Args:
destinations: a `DistributedValues`, variable, or string object.
Returns:
Boolean which is True if `destinations` is not empty.
"""
# Calling bool() on a ResourceVariable is not allowed.
if isinstance(destinations, resource_variable_ops.BaseResourceVariable):
return bool(destinations.device)
return bool(destinations)
def validate_destinations(destinations):
if not isinstance(destinations,
(value_lib.DistributedValues,
resource_variable_ops.BaseResourceVariable,
value_lib.AggregatingVariable,
six.string_types,
value_lib.TPUMirroredVariable,
# LogicalDeviceSpec is only used internally, e.g. as a
# broadcast destination, never supplied by a user.
value_lib.LogicalDeviceSpec)):
raise ValueError("destinations must be one of a `DistributedValues` object,"
" a tf.Variable object, or a device string.")
if not check_destinations(destinations):
raise ValueError("destinations can not be empty")
def reduce_non_distributed_value(reduce_op, device_map, value, destinations):
"""Reduce a non-DistributedValue `value` to `destinations`."""
if isinstance(value, value_lib.DistributedValues):
raise ValueError("You are passing a `DistributedValue` to "
"`reduce_non_distributed_value`, which is not allowed.")
# If the same value is present on all replicas then the PerReplica value will
# be a single value. We also handle the case when `value` is a single value
# and equal to 0.
if value == 0:
return 0
# If there is only a single value and the reduce op is MEAN,
# that value should be on all destinations.
if reduce_op == reduce_util.ReduceOp.MEAN:
return value
validate_destinations(destinations)
# We do not support a reduce op of SUM if the value is the same across
# all replicas. We call this as part of assign functions for MirroredVariables
# and summing up identical values across replicas is not clearly defined.
if device_map.num_replicas_in_graph != 1:
raise ValueError("A non-DistributedValues value %s cannot be reduced with "
"the given reduce op %s." % (value, reduce_op))
return simple_broadcast(value, destinations)
def _make_tensor_into_per_replica(input_tensor):
"""Converts a single tensor into a PerReplica object."""
if isinstance(input_tensor, (tuple, list)):
raise ValueError("Cannot convert `input_tensor` to a `PerReplica` object, "
"got %r but expected a object that is not a tuple or list."
% (input_tensor,))
if isinstance(input_tensor, value_lib.PerReplica):
return input_tensor
try:
device = input_tensor.device
except AttributeError:
raise ValueError("Cannot convert `input_tensor` to a `PerReplica` object "
"because it doesn't have device set.")
device_map = value_lib.SingleDeviceMap(device)
return value_lib.PerReplica(device_map, (input_tensor,))
def _normalize_value_destination_pairs(value_destination_pairs):
"""Converts each tensor into a PerReplica object in the input list."""
result = []
value_destination_pairs = list(value_destination_pairs)
if not isinstance(value_destination_pairs, (list, tuple)):
raise ValueError("`value_destination_pairs` should be a list or tuple")
for pair in value_destination_pairs:
if not isinstance(pair, tuple):
raise ValueError(
"Each element of `value_destination_pairs` should be a tuple.")
if len(pair) != 2:
raise ValueError("Each element of `value_destination_pairs` should be a "
"tuple of size 2.")
per_replica = _make_tensor_into_per_replica(pair[0])
result.append((per_replica, pair[1]))
return result
def _validate_value_destination_pairs(value_destination_pairs):
# TODO(yuefengz): raise exceptions instead of returning False.
# pylint: disable=g-missing-docstring
if not value_destination_pairs: return False
if not isinstance(value_destination_pairs, (list, tuple)): return False
if not all(isinstance(pair, tuple) for pair in value_destination_pairs):
return False
if not all(isinstance(v[0], value_lib.PerReplica)
for v in value_destination_pairs):
return False
return True
# TODO(yuefengz): consider calling this function in the caller of
# CrossDeviceOps.
def get_devices_from(destinations):
if isinstance(destinations, value_lib.DistributedValues):
return destinations.devices
elif isinstance(destinations, value_lib.LogicalDeviceSpec):
return destinations.device_map.logical_to_actual_devices(
destinations.logical_device)
elif isinstance(destinations, six.string_types):
return (device_util.resolve(destinations),)
return (destinations.device,)
def get_device_map_from(destinations):
if isinstance(destinations, (value_lib.DistributedValues,
value_lib.LogicalDeviceSpec)):
return destinations.device_map, destinations.logical_device
if isinstance(destinations, six.string_types):
device = device_util.resolve(destinations)
else:
device = destinations.device
return value_lib.SingleDeviceMap(device), 0
def _devices_match(left, right):
return set(get_devices_from(left)) == set(get_devices_from(right))
def _all_devices_match(value_destination_pairs):
if not all(_devices_match(v, d) for v, d in value_destination_pairs):
return False
if not all(_devices_match(v, value_destination_pairs[0][0])
for v, _ in value_destination_pairs[1:]):
return False
return True
def simple_broadcast(value, destinations, always_mirrored=False):
"""Broadcast `value` to `destinations` using simple copies."""
device_map, logical_device = get_device_map_from(destinations)
devices = device_map.logical_to_actual_devices(logical_device)
if len(devices) == 1 and not always_mirrored:
return cross_device_utils.copy_tensor_or_indexed_slices_to_device(
value, devices[0])
else:
value_updates = []
for d in devices:
value_updates.append(
cross_device_utils.copy_tensor_or_indexed_slices_to_device(
value, d))
return value_lib.Mirrored(device_map, value_updates, logical_device)
def _simple_reduce(per_replica_value, reduce_to_device, accumulation_fn,
reduce_op):
# pylint: disable=g-missing-docstring
all_values = per_replica_value.values
if not all_values:
raise ValueError("`per_replica_value` must be non-empty")
count = len(all_values)
if (count == 1 and all_values[0].device == reduce_to_device):
return all_values[0]
with ops.device(reduce_to_device):
with context.device_policy(context.DEVICE_PLACEMENT_SILENT):
reduced = cross_device_utils.aggregate_tensors_or_indexed_slices(
all_values, accumulation_fn)
if reduce_op == reduce_util.ReduceOp.MEAN:
reduced = cross_device_utils.divide_by_n_tensors_or_indexed_slices(
reduced, count)
elif reduce_op != reduce_util.ReduceOp.SUM:
raise ValueError("`reduce_op` must be Reduce.SUM or Reduce.MEAN.")
return reduced
@tf_export("distribute.CrossDeviceOps")
class CrossDeviceOps(object):
"""Base class for cross-device reduction and broadcasting algorithms."""
def __init__(self):
pass
@property
def _num_between_graph_workers(self):
# Returns 1 by default, the value may be overridden by sub classes.
return 1
def reduce(self, reduce_op, per_replica_value, destinations):
"""Reduce `per_replica_value` to `destinations`.
It runs the reduction operation defined by `reduce_op` and put the
result on `destinations`.
Args:
reduce_op: Indicates how per_replica_value will be reduced. Accepted
values are `tf.distribute.ReduceOp.SUM`, `tf.distribute.ReduceOp.MEAN`.
per_replica_value: a PerReplica object or a tensor with device set.
destinations: the reduction destinations.
Returns:
a Mirrored object.
Raises:
ValueError: if per_replica_value can't be converted to a PerReplica
object.
"""
if not isinstance(per_replica_value, value_lib.PerReplica):
per_replica_value = _make_tensor_into_per_replica(per_replica_value)
validate_destinations(destinations)
# Shortcut if `per_replica_value` only contains one value.
if self._num_between_graph_workers == 1 and len(
per_replica_value.values) == 1 and _devices_match(
per_replica_value, destinations):
return value_lib.Mirrored(per_replica_value.device_map,
per_replica_value.values)
return self.reduce_implementation(reduce_op, per_replica_value,
destinations)
def batch_reduce(self, reduce_op, value_destination_pairs):
"""Reduce PerReplica objects in a batch.
Reduce each first element in `value_destination_pairs` to each second
element which indicates the destinations.
Args:
reduce_op: Indicates how per_replica_value will be reduced. Accepted
values are `tf.distribute.ReduceOp.SUM`, `tf.distribute.ReduceOp.MEAN`.
value_destination_pairs: a list or a tuple of tuples of PerReplica objects
(or tensors with device set if there is one device) and destinations.
Returns:
a list of Mirrored objects.
Raises:
ValueError: if `value_destination_pairs` is not a list or a tuple of
tuples of PerReplica objects and destinations
"""
# TODO(yuefengz): if destinations are different, split into several
# `_batch_reduce` invocations.
if not _validate_value_destination_pairs(value_destination_pairs):
# If the first element of each pair is a tensor, we try to turn it into a
# PerReplica object.
value_destination_pairs = _normalize_value_destination_pairs(
value_destination_pairs)
for _, d in value_destination_pairs:
validate_destinations(d)
# Shortcut all PerReplica objects only contain one value.
if self._num_between_graph_workers == 1 and _all_devices_match(
value_destination_pairs) and len(
value_destination_pairs[0][0].values) == 1:
return [
value_lib.Mirrored(v.device_map, v.values)
for v, _ in value_destination_pairs
]
return self.batch_reduce_implementation(reduce_op, value_destination_pairs)
def broadcast(self, tensor, destinations):
"""Broadcast the `tensor` to destinations.
Args:
tensor: the tensor to broadcast.
destinations: the broadcast destinations.
Returns:
a Mirrored object.
"""
validate_destinations(destinations)
return self.broadcast_implementation(tensor, destinations)
@doc_controls.for_subclass_implementers
def reduce_implementation(self, reduce_op, per_replica_value, destinations):
"""The implementation of reduce of `per_replica_value` to `destinations`.
It runs the reduction operation defined by `reduce_op` and put the
result on `destinations`.
Args:
reduce_op: Indicates how per_replica_value will be reduced. Accepted
values are `tf.distribute.ReduceOp.SUM`, `tf.distribute.ReduceOp.MEAN`.
per_replica_value: a PerReplica object or a tensor with device set.
destinations: the reduction destinations.
Returns:
a Mirrored object.
Raises:
ValueError: if per_replica_value can't be converted to a PerReplica
object.
"""
raise NotImplementedError(
"_reduce method must be implemented in descendants.")
@doc_controls.for_subclass_implementers
def batch_reduce_implementation(self, reduce_op, value_destination_pairs):
"""Implementation of reduce PerReplica objects in a batch.
Reduce each first element in `value_destination_pairs` to each second
element which indicates the destinations.
Args:
reduce_op: Indicates how per_replica_value will be reduced. Accepted
values are `tf.distribute.ReduceOp.SUM`, `tf.distribute.ReduceOp.MEAN`.
value_destination_pairs: a list or a tuple of tuples of PerReplica objects
(or tensors with device set if there is one device) and destinations.
Returns:
a list of Mirrored objects.
Raises:
ValueError: if `value_destination_pairs` is not a list or a tuple of
tuples of PerReplica objects and destinations
"""
raise NotImplementedError(
"_batch_reduce method must be implemented in descendants.")
@doc_controls.for_subclass_implementers
def broadcast_implementation(self, tensor, destinations):
"""Implementation of broadcast the `tensor` to destinations.
Args:
tensor: the tensor to broadcast.
destinations: the broadcast destinations.
Returns:
a Mirrored object.
"""
return simple_broadcast(tensor, destinations, always_mirrored=True)
@tf_export("distribute.ReductionToOneDevice")
class ReductionToOneDevice(CrossDeviceOps):
"""Always do reduction to one device first and then do broadcasting.
Batch reduction is done by reduction on each element one by one.
"""
def __init__(self, reduce_to_device=None, accumulation_fn=None):
"""Constructor.
Args:
reduce_to_device: the intermediate device to reduce to. If None, reduce
to the first device in `destinations` of the reduce() method.
accumulation_fn: a function that does accumulation. If None, then
`tf.math.add_n` is used.
"""
self.reduce_to_device = reduce_to_device
self.accumulation_fn = accumulation_fn or math_ops.add_n
super(ReductionToOneDevice, self).__init__()
def reduce_implementation(self, reduce_op, per_replica_value, destinations):
if check_destinations(destinations):
devices = get_devices_from(destinations)
else:
devices = get_devices_from(per_replica_value)
reduce_to_device = self.reduce_to_device or devices[0]
logging.log_first_n(
logging.INFO,
"Reduce to %s then broadcast to %r." % (reduce_to_device, devices), 10)
reduced = _simple_reduce(per_replica_value, reduce_to_device,
self.accumulation_fn, reduce_op)
return self.broadcast(reduced, destinations)
def batch_reduce_implementation(self, reduce_op, value_destination_pairs):
return [
self.reduce_implementation(reduce_op, t, destinations=v)
for t, v in value_destination_pairs
]
def _group_value_by_device(per_replica_values):
"""Group values into sublists by their devices.
This grouping is needed to call the all-reduce library because it expects a
list of the following form:
[[(grad0_gpu0, v0_gpu0), (grad1_gpu0, v1_gpu0), (grad2_gpu0, v2_gpu0) ...],
[(grad0_gpu1, v0_gpu1), (grad1_gpu1, v1_gpu1), (grad2_gpu1, v2_gpu1) ...],
[(grad0_gpu2, v0_gpu2), (grad1_gpu0, v1_gpu2), (grad2_gpu0, v2_gpu2) ...],
...
]
Args:
per_replica_values: a list of PerReplica obejcts.
Returns:
a list of lists, each sublist has components for its corresponding device of
PerReplica objects, paired with a None.
"""
destinations = per_replica_values[0].devices
grouped = [[] for _ in range(len(destinations))]
for per_replica_value in per_replica_values:
# pylint: disable=protected-access
for i, v in enumerate(per_replica_value.values):
assert per_replica_value.devices == destinations
grouped[i].append((v, None))
return grouped
def _ungroup_and_make_mirrored(grouped_reduced,
destinations,
reduce_op,
num_between_graph_workers=1):
"""Ungroup results from all-reduce and make Mirrored objects.
Each all-reduce result will be divided by the number of destinations before
Mirrored objects are created if reduce_op is "mean".
Args:
grouped_reduced: a list of lists, each sublist has components for each
device, paired with a None. It is the result from
cross_device_utils.aggregate_gradients_using*.
destinations: a value to colocate the result with.
reduce_op: Indicates how values will be aggregated. Accepted values
are `tf.distribute.ReduceOp.SUM`, `tf.distribute.ReduceOp.MEAN`.
num_between_graph_workers: number of workers in the between-graph
replication.
Returns:
a list of Mirrored objects.
"""
device_map, logical_device = get_device_map_from(destinations)
num_replicas = device_map.num_replicas_in_graph * num_between_graph_workers
index = [[] for _ in range(len(grouped_reduced[0]))]
for per_replica_reduced in grouped_reduced:
for i, (v, _) in enumerate(per_replica_reduced):
if reduce_op == reduce_util.ReduceOp.MEAN:
index[i].append(v / num_replicas)
else:
index[i].append(v)
return [value_lib.Mirrored(device_map, v, logical_device) for v in index]
class _ConcatAndSplitPacker(object):
"""Concatenate and split tensors for reduction."""
def __init__(self, num_packs=1):
"""Initialize the _ConcatAndSplitPacker object.
Args:
num_packs: specifies the number of split packs that will be
formed.
Raises:
ValueError: if num_packs is not greater than 0.
"""
if num_packs <= 0:
raise ValueError("num_packs must be greater than zero.")
self.num_packs = num_packs
def pack(self, grouped_grads_and_vars):
"""Pack tensors."""
self.grouped_grads_and_vars = grouped_grads_and_vars
self.all_device_shapes = []
self.all_device_sizes = []
device_grad_packs = []
for device_grads_and_vars in grouped_grads_and_vars:
with ops.colocate_with(device_grads_and_vars[0][0]):
# Flatten all the grads.
flat_grads = [
array_ops.reshape(g, [-1]) for g, _ in device_grads_and_vars
]
# Remember the original shape of all the grads.
device_shapes = [array_ops.shape(g) for g, _ in device_grads_and_vars]
# Remember the original sizes of all the grads.
device_sizes = [array_ops.size(g) for g, _ in device_grads_and_vars]
# Concat all the flat grads into a big flat tensor.
concat_grads = array_ops.concat(flat_grads, 0)
# Split the big tensor into num_splits packs. In cases where the
# total size is not divisible num_splits, the last pack gets
# more elements.
# TODO(zhengxq): it is also possible to optimize away all the concat
# as well.
num_splits = self.num_packs
# The array_ops.size function will sometimes remove static shapes. So if
# all gradient shapes are defined, we use another method to get the
# total size.
# TODO(yuefengz): move this logic to array_ops.size.
if all(g.shape.is_fully_defined() for g, _ in device_grads_and_vars):
total_grad_size = sum(
[g.shape.num_elements() for g, _ in device_grads_and_vars])
else:
total_grad_size = array_ops.size(concat_grads)
split_size = total_grad_size // num_splits
split_size_last = total_grad_size - split_size * (num_splits - 1)
split_sizes = [split_size] * (num_splits - 1) + [split_size_last]
grad_packs = array_ops.split(concat_grads, split_sizes)
# Ready to aggregate the repacked gradients, with fake variables.
# TODO(zhengxq): It is hacky to have to use fake variables.
# We should remove the need for variables in
# aggregate_gradients_using*.
device_grad_packs.append(zip(grad_packs, [None] * num_splits))
self.all_device_shapes.append(device_shapes)
self.all_device_sizes.append(device_sizes)
return device_grad_packs
def unpack(self, summed_device_grad_packs):
"""Reverse the pack."""
aggregated_device_grads = []
for (summed_device_grad_packs,
device_grads_and_vars, device_shapes, device_sizes) in zip(
summed_device_grad_packs, self.grouped_grads_and_vars,
self.all_device_shapes, self.all_device_sizes):
# pylint: enable=line-too-long
# Reverse the packing operations in the previous steps. Form the
# summed gradients back into their original shapes.
with ops.colocate_with(summed_device_grad_packs[0][0]):
# Form a list of the summed grad packs.
device_grad_packs = [g for g, _ in summed_device_grad_packs]
# Concat them back into a big flat tensor.
device_grads_concat = array_ops.concat(device_grad_packs, 0)
# Split the tensors back into their original sizes.
grads_with_sizes = array_ops.split(device_grads_concat, device_sizes)
# Reshape the tensors back into their original shapes.
grads_with_shapes = [
array_ops.reshape(grad, shape)
for shape, grad in zip(device_shapes, grads_with_sizes)
]
# Form the list with the original list of variables.
summed_device_grads = [
(g, v) for g, (_, v) in zip(grads_with_shapes,
device_grads_and_vars)
]
aggregated_device_grads.append(summed_device_grads)
return aggregated_device_grads
class _AggregateSmallTensorPacker(object):
"""Concatenate small gradient tensors together for reduction."""
def __init__(self,
agg_small_grads_max_bytes=1048576,
agg_small_grads_max_group=16):
"""Initialize the _AggregateSmallTensorPacker object.
Args:
agg_small_grads_max_bytes: largest tensor eligible for aggregation,
in number of bytes.
agg_small_grads_max_group: largest permitted aggregation of small
tensors.
Raises:
ValueError: if `agg_small_grads_max_bytes` or `agg_small_grads_max_group`
is not greater than 0.
"""
if agg_small_grads_max_bytes <= 0 or agg_small_grads_max_group <= 0:
raise ValueError("agg_small_grads_max_bytes and agg_small_grads_max_group"
" should both be greater than zero.")
self.agg_small_grads_max_bytes = agg_small_grads_max_bytes
self.agg_small_grads_max_group = agg_small_grads_max_group
def pack(self, grouped_grads_and_vars):
"""Aggregate small tensors."""
if (self.agg_small_grads_max_bytes > 0 and
self.agg_small_grads_max_group > 0):
device_grads, self.packing = cross_device_utils.pack_small_tensors(
grouped_grads_and_vars,
max_bytes=self.agg_small_grads_max_bytes,
max_group=self.agg_small_grads_max_group)
return device_grads
def unpack(self, summed_device_grad_packs):
"""Reverse the aggregation process."""
return cross_device_utils.unpack_small_tensors(summed_device_grad_packs,
self.packing)
def _pack_tensors(device_grads,
num_packs=0,
agg_small_grads_max_bytes=0,
agg_small_grads_max_group=0):
"""Pack tensors if specified."""
if num_packs > 0:
tensor_packer = _ConcatAndSplitPacker(num_packs)
device_grad_packs = tensor_packer.pack(device_grads)
elif agg_small_grads_max_bytes > 0 and agg_small_grads_max_group > 0:
tensor_packer = _AggregateSmallTensorPacker(agg_small_grads_max_bytes,
agg_small_grads_max_group)
device_grad_packs = tensor_packer.pack(device_grads)
else:
tensor_packer = None
device_grad_packs = device_grads
return device_grad_packs, tensor_packer
def _unpack_tensors(reduced, tensor_packer=None):
"""Unpack tensors if they are packed before all-reduce."""
if tensor_packer:
return tensor_packer.unpack(reduced)
return reduced
class AllReduceCrossDeviceOps(CrossDeviceOps):
"""Reduction using all-reduce."""
def __init__(self,
all_reduce_alg="nccl",
num_packs=1,
agg_small_grads_max_bytes=0,
agg_small_grads_max_group=10):
"""All-reduce implementation of CrossDeviceOps.
Before performing all-reduce, tensors will be repacked or aggregated for
more efficient cross-device transportation:
1) If `num_packs` is non-zero, pack values into
`num_packs` splits.
2) Otherwise, if `agg_small_grads_max_bytes` > 0 and
`agg_small_grads_max_group` > 0, aggregate values smaller than
`agg_small_grads_max_bytes` into groups with at most
`agg_small_grads_max_group` values.
3) Otherwise, no repacking or grouping will happen.
Args:
all_reduce_alg: the all-reduce algorithm to use, currently only "nccl" or
"hierarchical_copy" are supported.
num_packs: see above.
agg_small_grads_max_bytes: see above.
agg_small_grads_max_group: see above.
"""
self._all_reduce_alg = all_reduce_alg
self._num_packs = num_packs
self._agg_small_grads_max_bytes = agg_small_grads_max_bytes
self._agg_small_grads_max_group = agg_small_grads_max_group
self._simple_cross_replica_ops = ReductionToOneDevice()
super(AllReduceCrossDeviceOps, self).__init__()
def reduce_implementation(self, reduce_op, per_replica_value, destinations):
if _devices_match(per_replica_value, destinations):
return self._batch_all_reduce(reduce_op, [per_replica_value])[0]
else:
return self._simple_cross_replica_ops.reduce(reduce_op, per_replica_value,
destinations)
def batch_reduce_implementation(self, reduce_op, value_destination_pairs):
if _all_devices_match(value_destination_pairs):
return self._batch_all_reduce(reduce_op,
[v[0] for v in value_destination_pairs])
else:
return [
self.reduce_implementation(reduce_op, t, destinations=v)
for t, v in value_destination_pairs
]
def _batch_all_reduce(self, reduce_op, per_replica_values):
"""All-reduce algorithm in a batch."""
dense_values, dense_indices, sparse_values, sparse_indices = (
cross_device_utils.split_by_sparsity(per_replica_values))
if dense_values:
dense_results = self._do_batch_all_reduce(reduce_op, dense_values)
else:
dense_results = []
if sparse_values:
sparse_results = self._do_batch_all_reduce_sparse(reduce_op,
sparse_values)
else:
sparse_results = []
return cross_device_utils.stitch_values(((dense_results, dense_indices),
(sparse_results, sparse_indices)))
def _do_batch_all_reduce(self, reduce_op, dense_values):
"""Run batch all-reduces."""
logging.log_first_n(
logging.INFO, "batch_all_reduce: %d all-reduces with algorithm = %s, "
"num_packs = %d, agg_small_grads_max_bytes = %d and "
"agg_small_grads_max_group = %d" %
(len(dense_values), self._all_reduce_alg, self._num_packs,
self._agg_small_grads_max_bytes, self._agg_small_grads_max_group), 10)
destinations = dense_values[0].devices
grouped = _group_value_by_device(dense_values)
device_grad_packs, tensor_packer = _pack_tensors(
grouped, self._num_packs, self._agg_small_grads_max_bytes,
self._agg_small_grads_max_group)
# The actual aggregation of the repacked gradients. Note that they are
# sharded among different aggregation trees. So it is important to strike
# the balance on num_splits.
if self._all_reduce_alg == "nccl":
# TODO(yuefengz): merge this into the all-reduce library.
reduced = cross_device_utils.aggregate_gradients_using_nccl(
device_grad_packs)
else:
# TODO(yuefengz): check that gpu ids in `destinations` are in ascending
# order.
reduced = (
cross_device_utils.aggregate_gradients_using_hierarchical_copy(
destinations, device_grad_packs))
reduced = _unpack_tensors(reduced, tensor_packer)
return _ungroup_and_make_mirrored(reduced, dense_values[0], reduce_op)
def _do_batch_all_reduce_sparse(self, reduce_op, sparse_values):
"""Run batch all-reduce for sparse values."""
logging.log_first_n(
logging.WARN,
"Efficient allreduce is not supported for %d IndexedSlices" %
len(sparse_values), 10)
# Use `sparse_values` as destinations to do all-reduces. It is effectively
# an allgather under the hood but not an efficient one.
return self._simple_cross_replica_ops.batch_reduce(
reduce_op, zip(sparse_values, sparse_values))
# For compatibility with code using the old name of `AllReduceCrossDeviceOps`.
AllReduceCrossTowerOps = AllReduceCrossDeviceOps
AllReduceSpecTuple = collections.namedtuple("AllReduceSpecTuple",
"alg shards limit")
@tf_export("distribute.NcclAllReduce")
class NcclAllReduce(AllReduceCrossDeviceOps):
"""Reduction using NCCL all-reduce."""
def __init__(self, num_packs=1):
"""NCCL all-reduce implementation of CrossDeviceOps.
Before performing all-reduce, tensors will be repacked or aggregated for
more efficient cross-device transportation.
Args:
num_packs: values will be packed in this many splits. `num_packs` should
be greater than 0.
"""
assert num_packs > 0, (
"NCLL all-reduce requires num_packs > 0, but {} is specified".format(
num_packs))
super(NcclAllReduce, self).__init__(
all_reduce_alg="nccl", num_packs=num_packs)
@tf_export("distribute.HierarchicalCopyAllReduce")
class HierarchicalCopyAllReduce(AllReduceCrossDeviceOps):
"""Reduction using hierarchical copy all-reduce.
This is a good reduction for configurations like Nvidia DGX-1.
"""
def __init__(self, num_packs=1):
"""Hierarchical copy all-reduce implementation of CrossDeviceOps.
Before performing all-reduce, tensors will be repacked or aggregated for
more efficient cross-device transportation.
Args:
num_packs: values will be packed in this many splits. `num_packs` should
be greater than 0.
"""
super(HierarchicalCopyAllReduce, self).__init__(
all_reduce_alg="hierarchical_copy",
num_packs=num_packs)
class MultiWorkerAllReduce(AllReduceCrossDeviceOps):
"""All-reduce algorithms for distributed TensorFlow."""
def __init__(self,
worker_devices,
num_gpus_per_worker,
all_reduce_spec=("pscpu/pscpu", 2, -1),
num_packs=0,
agg_small_grads_max_bytes=0,
agg_small_grads_max_group=10):
"""Initialize the all-reduce algorithm.
Args:
worker_devices: a list of device strings for workers participating in
all-reduce.
num_gpus_per_worker: number of GPU devices per worker.
all_reduce_spec: a tuple or a named tuple or a list of tuples specifying
the all-reduce algorithm.
1. The first element of a tuple is the name of the all-reduce algorithm.
Valid algorithm names are: "nccl", "nccl/xring", "nccl/rechd",
"nccl/pscpu", "xring", "pscpu", "psgpu", "pscpu/pscpu". Algorithms with
a "/" are hierarchical, so two all-reduces are executed, the first one
aggregates tensors within a worker and the second aggregates across
workers.
2. The second element of a tuple is the number of shards when doing
all-reduce. Let's say its values is M, each tensor after packing will be
split into M shards and then M parallel all-reduces would be performed
before finally they are concatenated backed into a complete tensor.
3. The third element is the maximum size of tensors that will be
applicable for the algorithm specified by the first element. For
example, if all_reduce_spec=[("nccl", 2, 1024), ("pscpu/pscpu", 2, -1)],
tensors with size not larger than 1024 bytes will be applied a 2-shard
"nccl" all-reduce and other tensors will be applied a 2-shard
"pscpu/pscpu" algorithm. The third elements should be in increasing
order across tuples and end with -1 which indicates infinity.
num_packs: see AllReduceCrossDeviceOps.
agg_small_grads_max_bytes: see AllReduceCrossDeviceOps.
agg_small_grads_max_group: see AllReduceCrossDeviceOps.
"""
self._worker_devices = worker_devices
self._num_gpus_per_worker = num_gpus_per_worker
super(MultiWorkerAllReduce, self).__init__(
num_packs=num_packs,
agg_small_grads_max_bytes=agg_small_grads_max_bytes,
agg_small_grads_max_group=agg_small_grads_max_group)
def validate_and_complete_spec(spec):
"""Validate and complete the all-reduce spec."""
# TODO(yuefengz): support namedtuple.
if not isinstance(spec, tuple):
raise ValueError(
"A tuple is expected for all-reduce spec: %r" % all_reduce_spec)
if not spec or len(spec) > 3:
raise ValueError(
"Too many elements in the all-reduce spec tuple: %r" % spec)
if len(spec) == 1:
return AllReduceSpecTuple(spec[0], 1, -1)
elif len(spec) == 2:
return AllReduceSpecTuple(spec[0], spec[1], -1)
else:
return AllReduceSpecTuple(*spec)
self._all_reduce_spec = []
if isinstance(all_reduce_spec, six.string_types):
self._all_reduce_spec.append(AllReduceSpecTuple(all_reduce_spec, 1, -1))
elif isinstance(all_reduce_spec, tuple):
self._all_reduce_spec.append(validate_and_complete_spec(all_reduce_spec))
elif isinstance(all_reduce_spec, list):
self._all_reduce_spec = [
validate_and_complete_spec(spec) for spec in all_reduce_spec
]
def _batch_all_reduce(self, reduce_op, per_replica_values):
"""All-reduce algorithm in a batch."""
logging.log_first_n(
logging.INFO,
"Distributed batch_all_reduce: %d all-reduces with "
"allreduce_spec = %r, num_packs = %d, agg_small_grads_max_bytes = %d, "
"and agg_small_grads_max_group = %d" %
(len(per_replica_values), self._all_reduce_spec, self._num_packs,
self._agg_small_grads_max_bytes, self._agg_small_grads_max_group), 10)
device_grads = _group_value_by_device(per_replica_values)
# The all-reduce library requires fully defined shapes.
# TODO(yuefengz): when tensor sharding is not needed, static shapes are not
# required as well.
for device_grad in device_grads:
for grad, _ in device_grad:
if not grad.shape.is_fully_defined():
raise ValueError("Shape is unknown for node %r" % grad)
remaining_grads = device_grads
aggregated_grads = []
for spec_tuple in self._all_reduce_spec:
if spec_tuple.limit < 0:
this_grads = remaining_grads
remaining_grads = []
else:
(this_grads, remaining_grads) = cross_device_utils.split_grads_by_size(
spec_tuple.limit, remaining_grads)
if this_grads:
device_grad_packs, tensor_packer = _pack_tensors(
this_grads, self._num_packs, self._agg_small_grads_max_bytes,
self._agg_small_grads_max_group)
range_agg_grads = cross_device_utils.sum_gradients_all_reduce(
self._worker_devices, device_grad_packs, len(self._worker_devices),
spec_tuple.alg, spec_tuple.shards, range(self._num_gpus_per_worker))
range_agg_grads = _unpack_tensors(range_agg_grads, tensor_packer)
if not aggregated_grads:
aggregated_grads = range_agg_grads
else:
assert len(aggregated_grads) == len(range_agg_grads)
for i in range(len(aggregated_grads)):
aggregated_grads[i] += range_agg_grads[i]
assert not remaining_grads
return _ungroup_and_make_mirrored(aggregated_grads, per_replica_values[0],
reduce_op)
@tf_export("distribute.experimental.CollectiveCommunication")
class CollectiveCommunication(enum.Enum):
"""Communication choices for CollectiveOps.
* `AUTO`: Default to runtime's automatic choices.
* `RING`: TensorFlow's ring algorithms for all-reduce and
all-gather.
* `NCCL`: Use ncclAllReduce for all-reduce, and ring algorithms for
all-gather. TODO(ayushd): add ncclAllGather implementation.
"""
AUTO = "AUTO"
RING = "RING"
NCCL = "NCCL"
# TODO(yuefengz): support in-graph collective all-reduce.
class CollectiveAllReduce(CrossDeviceOps):
"""All-reduce cross device ops using collective ops.
In the between-graph replicated training, it will still do all-reduces across
all workers and then put results on the right destinations.
"""
def __init__(self,
num_workers=1,
num_gpus_per_worker=0,
all_reduce_merge_scope=32,
collective_keys=None):
"""Initializes the object.
Args:
num_workers: number of workers in the between-graph replicated training.
num_gpus_per_worker: number of GPUs per worker.
all_reduce_merge_scope: size of groups into which to partition consecutive
gradients grouped under a common 'allreduce' name scope. This is useful
for some optimization of collective ops.
collective_keys: an optional CollectiveKey object.
"""
self._num_workers = num_workers
self._num_gpus_per_worker = num_gpus_per_worker
self._all_reduce_merge_scope = all_reduce_merge_scope
self._collective_keys = (collective_keys or
cross_device_utils.CollectiveKeys())
super(CollectiveAllReduce, self).__init__()
@property
def _num_between_graph_workers(self):
return self._num_workers
def reduce_implementation(self, reduce_op, per_replica_value, destinations):
all_reduced = self._batch_all_reduce(reduce_op, [per_replica_value])[0]
device_map, logical_device = get_device_map_from(destinations)
if (all_reduced.device_map is device_map and
all_reduced.logical_device == logical_device):
return all_reduced
devices = device_map.logical_to_actual_devices(logical_device)
index = []
with ops.control_dependencies(all_reduced.values):
for d in devices:
with ops.device(d):
if d in all_reduced.devices:
index.append(array_ops.identity(all_reduced.get(d)))
else:
# TODO(josh11b): Once we add support for model parallelism, get the
# copy from the corresponding replica instead of the primary.
index.append(array_ops.identity(all_reduced.primary))
return value_lib.Mirrored(device_map, index, logical_device)
def batch_reduce_implementation(self, reduce_op, value_destination_pairs):
all_devices_match = _all_devices_match(value_destination_pairs)
if all_devices_match:
return self._batch_all_reduce(reduce_op,
[v[0] for v in value_destination_pairs])
else:
if not all_devices_match:
logging.log_first_n(
logging.WARN, "Efficient batch_reduce is not supported if "
"destinations are different.", 10)
return [
self.reduce_implementation(reduce_op, t, destinations=v)
for t, v in value_destination_pairs
]
def _make_gradient_chunks(self, per_replica_values, all_reduce_merge_scope):
"""Make `per_replica_values` into chunks."""
grouped_by_device = _group_value_by_device(per_replica_values)
grouped_by_var = list(zip(*grouped_by_device))
# grouped_by_var is grouped by variables and takes the following format:
# [((grad0_gpu0, v0_gpu0), (grad0_gpu1, v0_gpu1), (grad0_gpu2, v0_gpu2) ..),
# ((grad1_gpu0, v1_gpu0), (grad1_gpu1, v1_gpu1), (grad1_gpu0, v1_gpu2) ..),
# ((grad2_gpu0, v2_gpu0), (grad2_gpu1, v2_gpu1), (grad2_gpu0, v2_gpu2) ..),
# ...
# ]
chunked_gv = [
grouped_by_var[x:x + all_reduce_merge_scope]
for x in range(0, len(grouped_by_var), all_reduce_merge_scope)
]
return chunked_gv
def _batch_all_reduce(self, reduce_op, per_replica_values):
"""All reduce algorithm in a batch."""
dense_values, dense_indices, sparse_values, sparse_indices = (
cross_device_utils.split_by_sparsity(per_replica_values))
if dense_values:
dense_results = self._do_batch_all_reduce_dense(reduce_op, dense_values)
else:
dense_results = []
if sparse_values:
sparse_results = self._do_batch_all_reduce_sparse(reduce_op,
sparse_values)
else:
sparse_results = []
return cross_device_utils.stitch_values(((dense_results, dense_indices),
(sparse_results, sparse_indices)))
def _do_batch_all_reduce_dense(self, reduce_op, per_replica_values):
"""All-reduce across all workers in a batch."""
logging.log_first_n(
logging.INFO, "Collective batch_all_reduce: %d all-reduces, "
"num_workers = %d" % (len(per_replica_values), self._num_workers), 10)
chunked_gv = self._make_gradient_chunks(per_replica_values,
self._all_reduce_merge_scope)
reduced_gv_list = []
for chunk in chunked_gv:
with ops.name_scope("allreduce"):
for grad_and_vars in chunk:
# Gradients for the same variable but from different devices.
scaled_grads = [g for g, _ in grad_and_vars]
collective_reduced = cross_device_utils.build_collective_reduce(
scaled_grads, self._num_workers, self._collective_keys, "Add",
"Id")
result = []
for (_, v), g in zip(grad_and_vars, collective_reduced):
result.append([g, v])
reduced_gv_list.append(result)
new_device_grads = [list(x) for x in zip(*reduced_gv_list)]
return _ungroup_and_make_mirrored(
new_device_grads,
per_replica_values[0],
reduce_op,
num_between_graph_workers=self._num_workers)
def _do_batch_all_reduce_sparse(self, reduce_op, per_replica_values):
"""All-reduce IndexedSlices across all workers in a batch."""
logging.log_first_n(
logging.INFO, "Collective batch_all_reduce for IndexedSlices: "
"%d all-reduces, num_workers = %d" %
(len(per_replica_values), self._num_workers), 10)
chunked_gv = self._make_gradient_chunks(per_replica_values,
self._all_reduce_merge_scope)
reduced_gv_list = []
for chunk in chunked_gv:
with ops.name_scope("allreduce"):
for grad_and_vars in chunk:
# Gradients for the same variable but from different devices.
scaled_grads = [g for g, _ in grad_and_vars]
values = [g.values for g in scaled_grads]
indices = [g.indices for g in scaled_grads]
assert len(values) == len(indices)
# Build two separate allgathers, one for values, the other one for
# indices.
gathered_values = cross_device_utils.build_collective_gather(
values, self._num_workers, self._collective_keys)
gathered_indices = cross_device_utils.build_collective_gather(
indices, self._num_workers, self._collective_keys)
assert len(gathered_values) == len(gathered_indices)
collective_reduced = []
for i in range(len(values)):
reduced = ops.IndexedSlices(
gathered_values[i],
gathered_indices[i],
dense_shape=scaled_grads[i].dense_shape)
collective_reduced.append(reduced)
result = []
for (_, v), g in zip(grad_and_vars, collective_reduced):
result.append([g, v])
reduced_gv_list.append(result)
new_device_grads = [list(x) for x in zip(*reduced_gv_list)]
return _ungroup_and_make_mirrored(
new_device_grads,
per_replica_values[0],
reduce_op,
num_between_graph_workers=self._num_workers)
def choose_the_best(devices, session_config=None):
"""Find the best CrossDeviceOps locally given a `tf.compat.v1.ConfigProto`.
Args:
devices: a list of devices passed to `tf.distribute.Strategy`.
session_config: a `tf.compat.v1.ConfigProto` or `None`. If `None`, it will
make decision based on all local devices.
Returns:
A subclass of `CrossDeviceOps`.
"""
requested_devices = set([device_util.canonicalize(d) for d in devices])
machine_devices = device_lib.list_local_devices(session_config=session_config)
using_devices = set()
for d in machine_devices:
if device_util.canonicalize(d.name) in requested_devices:
using_devices.add(d.name)
if len(using_devices) != len(requested_devices):
logging.warning(
"Some requested devices in `tf.distribute.Strategy` are not visible "
"to TensorFlow: %s", ",".join(list(requested_devices - using_devices)))
return ReductionToOneDevice()
if any("gpu" not in d.lower() for d in using_devices):
logging.warning("There is non-GPU devices in `tf.distribute.Strategy`, not "
"using nccl allreduce.")
return ReductionToOneDevice()
if kernels.get_registered_kernels_for_op("NcclAllReduce"):
return NcclAllReduce(num_packs=1)
else:
logging.warning("Nccl kernel is not found, not using nccl allreduce.")
return ReductionToOneDevice()
|
tensorflow-master
|
tensorflow/python/distribute/cross_device_ops.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for saving and loading using keras experimental APIs with DS."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.distribute import combinations
from tensorflow.python.distribute import saved_model_test_base as test_base
from tensorflow.python.eager import test
from tensorflow.python.keras.saving import saved_model
class KerasExperimentalSaveLoadTest(test_base.TestSavedModelBase):
def setUp(self):
self._root_dir = 'keras_experimental_save_load'
super(KerasExperimentalSaveLoadTest, self).setUp()
def _save_model(self, model, saved_dir):
saved_model.export_saved_model(model, saved_dir)
def _load_and_run_model(self, distribution, saved_dir, predict_dataset,
output_name):
restored_keras_model = saved_model.load_from_saved_model(saved_dir)
return restored_keras_model.predict(
predict_dataset, steps=test_base.PREDICT_STEPS)
@combinations.generate(test_base.simple_models_with_strategies())
def test_save_no_strategy_restore_strategy(self, model_and_input,
distribution):
self.run_test_save_no_strategy_restore_strategy(model_and_input,
distribution)
@combinations.generate(
combinations.times(test_base.simple_models_with_strategies(),
combinations.combine(save_in_scope=[True, False])))
def test_save_strategy_restore_no_strategy(self, model_and_input,
distribution, save_in_scope):
self.run_test_save_strategy_restore_no_strategy(model_and_input,
distribution, save_in_scope)
@combinations.generate(
combinations.times(test_base.simple_models_with_strategy_pairs(),
combinations.combine(save_in_scope=[True, False])))
def test_save_strategy_restore_strategy(self, model_and_input,
distribution_for_saving,
distribution_for_restoring,
save_in_scope):
self.run_test_save_strategy_restore_strategy(model_and_input,
distribution_for_saving,
distribution_for_restoring,
save_in_scope)
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/keras_experimental_saved_model_test.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for multi_worker_util."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.core.protobuf import cluster_pb2
from tensorflow.python.distribute import multi_worker_util
from tensorflow.python.eager import test
from tensorflow.python.training import server_lib
class NormalizeClusterSpecTest(test.TestCase):
def assert_same_cluster(self, lhs, rhs):
self.assertEqual(
server_lib.ClusterSpec(lhs).as_dict(),
server_lib.ClusterSpec(rhs).as_dict())
def testDictAsInput(self):
cluster_spec = {
"chief": ["127.0.0.1:1234"],
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"ps": ["127.0.0.1:1926", "127.0.0.1:3141"]
}
self.assert_same_cluster(
cluster_spec, multi_worker_util.normalize_cluster_spec(cluster_spec))
def testClusterDefAsInput(self):
cluster_def = cluster_pb2.ClusterDef()
job = cluster_def.job.add()
job.name = "chief"
job.tasks[0] = "127.0.0.1:1234"
job = cluster_def.job.add()
job.name = "worker"
job.tasks[0] = "127.0.0.1:8964"
job.tasks[1] = "127.0.0.1:2333"
job = cluster_def.job.add()
job.name = "ps"
job.tasks[0] = "127.0.0.1:1926"
job.tasks[1] = "127.0.0.1:3141"
self.assert_same_cluster(
cluster_def, multi_worker_util.normalize_cluster_spec(cluster_def))
def testClusterSpecAsInput(self):
cluster_spec = server_lib.ClusterSpec({
"chief": ["127.0.0.1:1234"],
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"ps": ["127.0.0.1:1926", "127.0.0.1:3141"]
})
self.assert_same_cluster(
cluster_spec, multi_worker_util.normalize_cluster_spec(cluster_spec))
def testUnexpectedInput(self):
cluster_spec = ["127.0.0.1:8964", "127.0.0.1:2333"]
with self.assertRaisesRegexp(
ValueError,
"`cluster_spec' should be dict or a `tf.train.ClusterSpec` or a "
"`tf.train.ClusterDef` object"):
multi_worker_util.normalize_cluster_spec(cluster_spec)
class IsChiefTest(test.TestCase):
def testClusterWithChief(self):
cluster_spec = {
"chief": ["127.0.0.1:1234"],
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"ps": ["127.0.0.1:1926", "127.0.0.1:3141"]
}
self.assertTrue(multi_worker_util.is_chief(cluster_spec, "chief", 0))
self.assertFalse(multi_worker_util.is_chief(cluster_spec, "worker", 0))
def testClusterWithoutChief(self):
cluster_spec = {"worker": ["127.0.0.1:8964", "127.0.0.1:2333"]}
self.assertTrue(multi_worker_util.is_chief(cluster_spec, "worker", 0))
self.assertFalse(multi_worker_util.is_chief(cluster_spec, "worker", 1))
with self.assertRaisesRegexp(
ValueError, "`task_type` 'chief' not found in cluster_spec."):
multi_worker_util.is_chief(cluster_spec, "chief", 0)
with self.assertRaisesRegexp(
ValueError, "The `task_id` 2 exceeds the maximum id of worker."):
multi_worker_util.is_chief(cluster_spec, "worker", 2)
def testEvaluatorIsChief(self):
cluster_spec = {
"chief": ["127.0.0.1:1234"],
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"evaluator": ["127.0.0.1:2019"]
}
self.assertTrue(multi_worker_util.is_chief(cluster_spec, "evaluator", 0))
class NumWorkersTest(test.TestCase):
def testCountWorker(self):
cluster_spec = {
"chief": ["127.0.0.1:1234"],
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"ps": ["127.0.0.1:1926", "127.0.0.1:3141"]
}
self.assertEqual(
multi_worker_util.worker_count(cluster_spec, task_type="chief"), 3)
self.assertEqual(
multi_worker_util.worker_count(cluster_spec, task_type="worker"), 3)
def testCountEvaluator(self):
cluster_spec = {
"chief": ["127.0.0.1:1234"],
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"evaluator": ["127.0.0.1:7566"]
}
self.assertEqual(
multi_worker_util.worker_count(cluster_spec, task_type="evaluator"), 1)
def testTaskTypeNotFound(self):
cluster_spec = {}
with self.assertRaisesRegexp(
ValueError, "`task_type` 'worker' not found in cluster_spec."):
multi_worker_util.worker_count(cluster_spec, task_type="worker")
def testCountPs(self):
cluster_spec = {
"chief": ["127.0.0.1:1234"],
"ps": ["127.0.0.1:1926", "127.0.0.1:3141"]
}
# A "ps" job shouldn't call this method.
with self.assertRaisesRegexp(ValueError, "Unexpected `task_type` 'ps'"):
multi_worker_util.worker_count(cluster_spec, task_type="ps")
class IdInClusterTest(test.TestCase):
def testChiefId(self):
cluster_spec = {
"chief": ["127.0.0.1:1234"],
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"ps": ["127.0.0.1:1926", "127.0.0.1:3141"]
}
self.assertEqual(
multi_worker_util.id_in_cluster(cluster_spec, "chief", 0), 0)
def testWorkerId(self):
cluster_spec = {
"chief": ["127.0.0.1:1234"],
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"ps": ["127.0.0.1:1926", "127.0.0.1:3141"]
}
self.assertEqual(
multi_worker_util.id_in_cluster(cluster_spec, "worker", 1), 2)
cluster_spec = {
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"ps": ["127.0.0.1:1926", "127.0.0.1:3141"]
}
self.assertEqual(
multi_worker_util.id_in_cluster(cluster_spec, "worker", 1), 1)
def testEvaluatorId(self):
cluster_spec = {
"chief": ["127.0.0.1:1234"],
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"evaluator": ["127.0.0.1:7566"]
}
self.assertEqual(
multi_worker_util.id_in_cluster(cluster_spec, "evaluator", 0), 0)
def testPsId(self):
cluster_spec = {"chief": ["127.0.0.1:1234"], "ps": ["127.0.0.1:7566"]}
with self.assertRaisesRegexp(ValueError,
"There is no id for task_type 'ps'"):
multi_worker_util.id_in_cluster(cluster_spec, "ps", 0)
def testMultipleChiefs(self):
cluster_spec = {
"chief": ["127.0.0.1:8258", "127.0.0.1:7566"],
}
with self.assertRaisesRegexp(ValueError,
"There must be at most one 'chief' job."):
multi_worker_util.id_in_cluster(cluster_spec, "chief", 0)
class CollectiveLeaderTest(test.TestCase):
def testChiefAsLeader(self):
cluster_spec = {
"chief": ["127.0.0.1:1234"],
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"ps": ["127.0.0.1:1926", "127.0.0.1:3141"]
}
self.assertEqual(
multi_worker_util.collective_leader(cluster_spec, "worker", 0),
"/job:chief/replica:0/task:0")
def testWorkerAsLeader(self):
cluster_spec = {
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"ps": ["127.0.0.1:1926", "127.0.0.1:3141"]
}
self.assertEqual(
multi_worker_util.collective_leader(cluster_spec, "worker", 1),
"/job:worker/replica:0/task:0")
def testLeaderForEvaluator(self):
cluster_spec = {
"chief": ["127.0.0.1:1234"],
"worker": ["127.0.0.1:8964", "127.0.0.1:2333"],
"ps": ["127.0.0.1:1926", "127.0.0.1:3141"],
"evaluator": ["127.0.0.1:2019"]
}
self.assertEqual(
multi_worker_util.collective_leader(cluster_spec, "evaluator", 0), "")
def testLocalLeader(self):
cluster_spec = {}
self.assertEqual(
multi_worker_util.collective_leader(cluster_spec, None, 0), "")
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/multi_worker_util_test.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Custom Training Loop correctness test.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from absl.testing import parameterized
import numpy as np
from tensorflow.python import keras
from tensorflow.python.compat import v2_compat
from tensorflow.python.data.ops import dataset_ops
from tensorflow.python.distribute import combinations
from tensorflow.python.distribute import reduce_util
from tensorflow.python.distribute import strategy_combinations
from tensorflow.python.eager import backprop
from tensorflow.python.eager import def_function
from tensorflow.python.eager import test
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import random_seed
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import nn
_NUM_SAMPLES = 64
_BATCH_SIZE = 32
_RANDOM_SEED = 1337
_NUM_EPOCHS = 2
_STEPS_PER_EPOCH = 2
class MaybeStrategyScope(object):
"""Provides a context allowing no distribution strategy."""
def __init__(self, strategy):
self._strategy = strategy
self._scope = None
def __enter__(self):
if self._strategy:
self._scope = self._strategy.scope()
self._scope.__enter__()
def __exit__(self, exc_type, value, traceback):
if self._strategy:
self._scope.__exit__(exc_type, value, traceback)
self._scope = None
def get_model():
model = keras.Sequential()
model.add(keras.layers.Dense(10, activation='relu', input_shape=(1,)))
model.add(keras.layers.Dense(
10, activation='relu',
kernel_regularizer=keras.regularizers.l2(1e-4)))
model.add(keras.layers.Dense(10, activation='relu'))
model.add(keras.layers.Dense(1))
return model
def get_data():
x_train = np.random.rand(_NUM_SAMPLES, 1)
y_train = 3 * x_train
x_train = x_train.astype('float32')
y_train = y_train.astype('float32')
train_dataset = dataset_ops.DatasetV2.from_tensor_slices((x_train, y_train))
train_dataset = train_dataset.batch(_BATCH_SIZE)
return train_dataset
def compute_loss(labels, logits, reg_losses):
pred_loss = keras.losses.mean_squared_error(labels, logits)
scaled_loss = nn.compute_average_loss(
pred_loss, global_batch_size=_BATCH_SIZE)
l2_loss = nn.scale_regularization_loss(reg_losses)
return scaled_loss + l2_loss
def iteration_inside_func(initial_weights, dataset, optimizer_fn,
iteration_type, strategy=None):
"""Helper function to test iterating over data inside a tf.function."""
with MaybeStrategyScope(strategy):
model = get_model()
model.set_weights(initial_weights)
optimizer = optimizer_fn()
training_accuracy = keras.metrics.CategoricalAccuracy(
'training_accuracy', dtype=dtypes.float32)
@def_function.function
def train_epoch(dist_input):
"""Training StepFn."""
def step_fn(inputs):
samples, labels = inputs
with backprop.GradientTape() as tape:
logits = model(samples)
loss = compute_loss(labels, logits, model.losses)
grads = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients(zip(grads, model.trainable_variables))
training_accuracy.update_state(labels, logits)
return loss
total_loss = 0.0
num_batches = 0
if iteration_type == 'dataset':
for x in dist_input:
if strategy:
per_replica_losses = strategy.experimental_run_v2(step_fn,
args=(x,))
total_loss += strategy.reduce(reduce_util.ReduceOp.SUM,
per_replica_losses,
axis=None)
else:
total_loss += step_fn(x)
num_batches += 1
else:
iterator = iter(dist_input)
for _ in range(_STEPS_PER_EPOCH):
if strategy:
per_replica_losses = strategy.experimental_run_v2(
step_fn, args=(next(iterator),))
total_loss += strategy.reduce(reduce_util.ReduceOp.SUM,
per_replica_losses,
axis=None)
else:
total_loss += step_fn(next(iterator))
num_batches += 1
return total_loss / math_ops.cast(num_batches, dtype=dtypes.float32)
if strategy:
dataset = strategy.experimental_distribute_dataset(dataset)
for _ in range(_NUM_EPOCHS):
loss = train_epoch(dataset)
return (model.get_weights(),
loss,
training_accuracy.result())
def iteration_outside_func(initial_weights, dataset, optimizer_fn,
iteration_type, strategy=None):
"""Helper function to test iterating over data outside a tf.function."""
with MaybeStrategyScope(strategy):
model = get_model()
model.set_weights(initial_weights)
optimizer = optimizer_fn()
training_accuracy = keras.metrics.CategoricalAccuracy(
'training_accuracy', dtype=dtypes.float32)
@def_function.function
def train_step(dist_inputs):
"""Training StepFn."""
def step_fn(inputs):
samples, labels = inputs
with backprop.GradientTape() as tape:
logits = model(samples)
loss = compute_loss(labels, logits, model.losses)
grads = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients(zip(grads, model.trainable_variables))
training_accuracy.update_state(labels, logits)
return loss
if strategy:
per_replica_losses = strategy.experimental_run_v2(
step_fn, args=(dist_inputs,))
return strategy.reduce(reduce_util.ReduceOp.SUM,
per_replica_losses,
axis=None)
else:
return step_fn(dist_inputs)
if strategy:
dataset = strategy.experimental_distribute_dataset(dataset)
total_loss = 0.0
num_batches = 0
if iteration_type == 'dataset':
for _ in range(_NUM_EPOCHS):
for x in dataset:
total_loss += train_step(x)
num_batches += 1
else:
for _ in range(_NUM_EPOCHS):
iterator = iter(dataset)
for _ in range(_STEPS_PER_EPOCH):
total_loss += train_step(next(iterator))
num_batches += 1
return (model.get_weights(),
total_loss / math_ops.cast(num_batches, dtype=dtypes.float32),
training_accuracy.result())
class TestDistributionStrategyDnnCorrectness(test.TestCase,
parameterized.TestCase):
"""Test custom training loop correctness with a simple DNN model."""
def setUp(self):
super(TestDistributionStrategyDnnCorrectness, self).setUp()
v2_compat.enable_v2_behavior()
np.random.seed(_RANDOM_SEED)
random_seed.set_random_seed(_RANDOM_SEED)
@combinations.generate(
combinations.combine(
distribution=strategy_combinations.strategies_minus_tpu,
optimizer_fn=strategy_combinations.optimizers_v1_and_v2,
mode=['eager'],
iteration_type=['iterator', 'dataset'],
inside_func=[False, True]
))
def test_dnn_correctness_minus_tpus(self, distribution, optimizer_fn,
iteration_type, inside_func):
self.dnn_correctness(distribution, optimizer_fn, iteration_type,
inside_func)
# TODO(b/133325470): Enable this test for all optimizers once we understand
# the root cause of flakiness.
@combinations.generate(
combinations.combine(
distribution=[strategy_combinations.tpu_strategy_one_step],
optimizer_fn=[strategy_combinations.adagrad_optimizer_keras_v2_fn],
mode=['eager'],
iteration_type=['iterator', 'dataset'],
inside_func=[False, True]
))
def test_dnn_correctness_tpus(self, distribution, optimizer_fn,
iteration_type, inside_func):
self.dnn_correctness(distribution, optimizer_fn, iteration_type,
inside_func)
def dnn_correctness(self, distribution, optimizer_fn, iteration_type,
inside_func):
model = get_model()
initial_weights = model.get_weights()
dataset = get_data()
if inside_func:
iteration_func = iteration_inside_func
else:
iteration_func = iteration_outside_func
wts_with_ds, loss_with_ds, acc_with_ds = iteration_func(
initial_weights, dataset, optimizer_fn, iteration_type,
strategy=distribution)
wts, loss, acc = iteration_func(initial_weights, dataset, optimizer_fn,
iteration_type)
self.assertAllClose(wts, wts_with_ds, atol=1e-3, rtol=1e-3)
self.assertAllClose(loss, loss_with_ds, atol=1e-3, rtol=1e-3)
self.assertAllClose(acc, acc_with_ds, atol=1e-3, rtol=1e-3)
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/ctl_correctness_test.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Code for creating a dataset out of a NumPy array."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.data.ops import dataset_ops
from tensorflow.python.eager import context
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import variable_scope
from tensorflow.python.util import nest
def init_var_from_numpy(input_var, numpy_input, session):
"""Initialize `input_var` to `numpy_input` using `session` in graph mode."""
with ops.init_scope():
if context.executing_eagerly():
input_var.assign(numpy_input)
return
assert session is not None
session.run(input_var.initializer)
start_placeholder = array_ops.placeholder(dtypes.int64, ())
end_placeholder = array_ops.placeholder(dtypes.int64, ())
slice_placeholder = array_ops.placeholder(input_var.dtype)
assign_slice_op = input_var[start_placeholder:end_placeholder].assign(
slice_placeholder)
# If each batch element is > 64 MB, then we copy each batch element
# individually. Otherwise, the slices will be < 128 MB. There might be
# padding which might mean that the slices are 128 MB even if the size of
# the tensor allocated is less than 128 MB. This formula gives slices with
# size: ceil(64 MB / byte size per batch element) bytes. Using ceil()
# guarantees we get a number >= 1.
# Calculate the size of each batch element.
byte_size_per_batch_element = (
np.prod(numpy_input.shape[1:]) * input_var.dtype.size)
# Calculate number of elements we want to copy per slice.
batch_size_per_slice = int(
np.ceil((64 << 20) / byte_size_per_batch_element))
# Copy slices of the above size starting at 0, except the last slice will be
# smaller.
start = 0
limit = numpy_input.shape[0]
while start < limit:
end = min(start + batch_size_per_slice, limit)
session.run(assign_slice_op, feed_dict={
start_placeholder: start,
end_placeholder: end,
slice_placeholder: numpy_input[start:end]})
start = end
def one_host_numpy_dataset(numpy_input, colocate_with, session):
"""Create a dataset on `colocate_with` from `numpy_input`."""
def create_colocated_variable(next_creator, *args, **kwargs):
kwargs["colocate_with"] = colocate_with
return next_creator(*args, **kwargs)
numpy_flat = nest.flatten(numpy_input)
with variable_scope.variable_creator_scope(create_colocated_variable):
vars_flat = tuple(variable_scope.variable(array_ops.zeros(i.shape, i.dtype),
trainable=False)
for i in numpy_flat)
for v, i in zip(vars_flat, numpy_flat):
init_var_from_numpy(v, i, session)
vars_nested = nest.pack_sequence_as(numpy_input, vars_flat)
return dataset_ops.Dataset.from_tensor_slices(vars_nested)
class SingleDevice(object):
"""Used with `colocate_with` to create a non-mirrored variable."""
def __init__(self, device):
self.device = device
|
tensorflow-master
|
tensorflow/python/distribute/numpy_dataset.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for Distribute Coordinator."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import contextlib
import copy
import json
import os
import sys
import threading
import time
import six
_portpicker_import_error = None
try:
import portpicker # pylint: disable=g-import-not-at-top
except ImportError as _error: # pylint: disable=invalid-name
_portpicker_import_error = _error
portpicker = None
# pylint: disable=g-import-not-at-top
from tensorflow.core.protobuf import config_pb2
from tensorflow.python.client import session
from tensorflow.python.distribute import distribute_coordinator
from tensorflow.python.distribute import distribute_coordinator_context
from tensorflow.python.framework import errors
from tensorflow.python.framework import ops
from tensorflow.python.framework import test_util
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import variable_scope
from tensorflow.python.ops import variables
from tensorflow.python.platform import test
from tensorflow.python.training import coordinator
from tensorflow.python.training import monitored_session
from tensorflow.python.training import session_manager
CHIEF = distribute_coordinator._TaskType.CHIEF
WORKER = distribute_coordinator._TaskType.WORKER
PS = distribute_coordinator._TaskType.PS
EVALUATOR = distribute_coordinator._TaskType.EVALUATOR
STANDALONE_CLIENT = distribute_coordinator.CoordinatorMode.STANDALONE_CLIENT
INDEPENDENT_WORKER = distribute_coordinator.CoordinatorMode.INDEPENDENT_WORKER
NUM_WORKERS = 3
NUM_PS = 2
original_sys_exit = sys.exit
def _bytes_to_str(maybe_bytes):
if isinstance(maybe_bytes, six.string_types):
return maybe_bytes
else:
return str(maybe_bytes, "utf-8")
def _strip_protocol(target):
# cluster_spec expects "host:port" strings.
if "//" in target:
return target.split("//")[1]
else:
return target
class MockExtended(object):
def __init__(self,
between_graph=False,
should_init=None,
should_checkpoint=None,
should_save_summary=None):
self.experimental_between_graph = between_graph
self.experimental_should_init = should_init
self.should_checkpoint = should_checkpoint
self.should_save_summary = should_save_summary
class MockStrategy(object):
def __init__(self,
between_graph=False,
should_init=None,
should_checkpoint=None,
should_save_summary=None):
self.extended = MockExtended(between_graph, should_init, should_checkpoint,
should_save_summary)
def configure(self,
session_config=None,
cluster_spec=None,
task_type=None,
task_id=None):
if self.extended.experimental_should_init is None:
if task_id == 0:
self.extended.experimental_should_init = True
else:
self.extended.experimental_should_init = False
if self.extended.should_checkpoint is None:
if task_id == 0:
self.extended.should_checkpoint = True
else:
self.extended.should_checkpoint = False
if self.extended.should_save_summary is None:
if task_id == 0:
self.extended.should_save_summary = True
else:
self.extended.should_save_summary = False
if session_config:
if (cluster_spec and task_type and task_id is not None and
self.extended.experimental_between_graph):
session_config.intra_op_parallelism_threads += 1
if task_type in ["chief", "worker"]:
session_config.device_filters.extend(
["/job:%s/task:%d" % (task_type, task_id), "/job:ps"])
else:
session_config.inter_op_parallelism_threads += 1
session_config.device_filters.append("/job:somejob")
class MockServer(object):
def __init__(self):
self._joined = False
self._started = False
def start(self):
self._started = True
def join(self):
assert not self._joined
self._joined = True
@property
def joined(self):
return self._joined
@property
def started(self):
return self._started
class DistributeCoordinatorTestBase(test.TestCase):
@classmethod
def setUpClass(cls):
# We have to create a global in-process cluster because once an in-process
# tensorflow server is created, there is no way to terminate it. Please see
# multi_worker_test_base.py for more details.
# TODO(yuefengz): use the utitliy from multi_worker_test_base.
cls._workers, cls._ps = test_util.create_local_cluster(
NUM_WORKERS, num_ps=NUM_PS)
cls._cluster_spec = {
WORKER: [
_strip_protocol(_bytes_to_str(w.target)) for w in cls._workers
],
PS: [_strip_protocol(_bytes_to_str(ps.target)) for ps in cls._ps]
}
def setUp(self):
self._result_correct = 0
self._lock = threading.Lock()
self._worker_context = {}
self._strategy_property = {}
self._std_servers = {}
self._barrier = distribute_coordinator._Barrier(NUM_WORKERS)
self._coord = coordinator.Coordinator()
@contextlib.contextmanager
def _test_session(self, target):
config = config_pb2.ConfigProto(allow_soft_placement=True)
config.graph_options.optimizer_options.opt_level = -1
with session.Session(graph=None, config=config, target=target) as sess:
yield sess
# TODO(yuefengz): use the utitliy from multi_worker_test_base.
def _create_cluster_spec(self,
has_chief=False,
num_workers=1,
num_ps=0,
has_eval=False):
if _portpicker_import_error:
raise _portpicker_import_error # pylint: disable=raising-bad-type
cluster_spec = {}
if has_chief:
cluster_spec[CHIEF] = ["localhost:%s" % portpicker.pick_unused_port()]
if num_workers:
cluster_spec[WORKER] = [
"localhost:%s" % portpicker.pick_unused_port()
for _ in range(num_workers)
]
if num_ps:
cluster_spec[PS] = [
"localhost:%s" % portpicker.pick_unused_port() for _ in range(num_ps)
]
if has_eval:
cluster_spec[EVALUATOR] = ["localhost:%s" % portpicker.pick_unused_port()]
return cluster_spec
def _in_graph_worker_fn(self, strategy):
context = distribute_coordinator_context.get_current_worker_context()
self.assertTrue(context is not None)
with self._test_session(target=context.master_target) as sess:
xs = []
expected = 0.0
for i in range(context.num_workers):
with ops.device("/job:worker/task:%d" % i):
x = variable_scope.get_variable("x_%d" % i, initializer=10.0)
x_add = x.assign_add(float(i))
xs.append(x_add)
expected += i + 10.0
with ops.device("/job:worker/task:0"):
result = math_ops.add_n(xs)
self.evaluate(variables.global_variables_initializer())
result_value = sess.run(result)
self.assertEqual(result_value, expected)
if result_value == expected:
self._result_correct += 1
def _wrapped_worker_fn(self, worker_fn):
def wrapped(*args, **kwargs):
with self._coord.stop_on_exception():
return worker_fn(*args, **kwargs)
return wrapped
def _run_coordinator_in_thread(self, worker_fn, strategy, **kwargs):
t = threading.Thread(
target=distribute_coordinator.run_distribute_coordinator,
args=(self._wrapped_worker_fn(worker_fn), strategy),
kwargs=kwargs)
t.start()
return t
def _run_multiple_coordinator_in_threads(self, worker_fn, strategy,
cluster_spec, **kwargs):
threads = {}
for task_type in cluster_spec.keys():
threads[task_type] = []
for task_id in range(len(cluster_spec[task_type])):
t = self._run_coordinator_in_thread(
worker_fn,
strategy,
cluster_spec=cluster_spec,
task_type=task_type,
task_id=task_id,
**kwargs)
threads[task_type].append(t)
return threads
def _join_threads(self, threads):
try:
self._coord.join(threads)
except errors.UnknownError as e:
if "Could not start gRPC server" in e.message:
self.skipTest("Cannot start std servers.")
else:
raise
def _between_graph_worker_fn(self, strategy):
context = distribute_coordinator_context.get_current_worker_context()
self.assertTrue(context is not None)
with self._test_session(target=context.master_target) as sess:
with ops.device("/job:ps/task:0"):
# TODO(yuefengz): investigate why not using resource variable will make
# the test flaky.
x = variable_scope.get_variable(
"x", initializer=10.0, use_resource=True)
with ops.device("/job:ps/task:1"):
y = variable_scope.get_variable(
"y", initializer=20.0, use_resource=True)
x_add = x.assign_add(2.0)
y_sub = y.assign_sub(2.0)
train_op = control_flow_ops.group([x_add, y_sub])
if context.is_chief:
self.evaluate(variables.global_variables_initializer())
# Synchronize workers after initializaton.
if context.has_barrier:
context.wait_for_other_workers()
else:
while True:
uninit_vars = sess.run(variables.report_uninitialized_variables())
# pylint: disable=g-explicit-length-test
if len(uninit_vars) == 0:
break
sess.run(train_op)
# Synchronize workers after one step to make sure they all have finished
# training.
if context.has_barrier:
context.wait_for_other_workers()
else:
self._barrier.wait()
x_val, y_val = sess.run([x, y])
self.assertEqual(x_val, 16.0)
self.assertEqual(y_val, 14.0)
if x_val == 16.0 and y_val == 14.0:
with self._lock:
self._result_correct += 1
def _between_graph_with_monitored_session(self, strategy):
context = distribute_coordinator_context.get_current_worker_context()
self.assertTrue(context is not None)
with ops.device("/job:ps/task:0"):
# TODO(yuefengz): investigate why not using resource variable will make
# the test flaky.
x = variable_scope.get_variable("xx", initializer=10.0, use_resource=True)
with ops.device("/job:ps/task:1"):
y = variable_scope.get_variable("yy", initializer=20.0, use_resource=True)
x_add = x.assign_add(2.0)
y_sub = y.assign_sub(2.0)
train_op = control_flow_ops.group([x_add, y_sub])
# The monitored session will run init or ready ops.
with monitored_session.MonitoredSession() as sess:
sess.run(train_op)
# Synchronize workers after one step to make sure they all have finished
# training.
if context.has_barrier:
context.wait_for_other_workers()
else:
self._barrier.wait()
x_val, y_val = sess.run([x, y])
self.assertEqual(x_val, 16.0)
self.assertEqual(y_val, 14.0)
if x_val == 16.0 and y_val == 14.0:
with self._lock:
self._result_correct += 1
def _dump_worker_context(self, strategy):
"""Dumps the propoerties of each worker context.
It dumps the context properties to a dict mapping from task_type to a list
of tuples of master_target, num_workers, is_chief and distribute_mode, where
the list is indexed by the task_id.
Args:
strategy: a `DistributionStrategy` object.
"""
context = distribute_coordinator_context.get_current_worker_context()
self.assertTrue(context is not None)
task_type = str(context.task_type)
task_id = context.task_id or 0
with self._lock:
if task_type not in self._worker_context:
self._worker_context[task_type] = []
while len(self._worker_context[task_type]) <= task_id:
self._worker_context[task_type].append(None)
self._worker_context[task_type][task_id] = (context.master_target,
context.num_workers,
context.is_chief,
context.distributed_mode)
def _dump_strategy_property(self, strategy):
context = distribute_coordinator_context.get_current_worker_context()
self.assertTrue(context is not None)
self.assertEqual(context._strategy.extended.experimental_should_init,
strategy.extended.experimental_should_init)
self.assertEqual(context.should_checkpoint,
strategy.extended.should_checkpoint)
self.assertEqual(context.should_save_summary,
strategy.extended.should_save_summary)
task_type = str(context.task_type)
task_id = context.task_id or 0
with self._lock:
if task_type not in self._strategy_property:
self._strategy_property[task_type] = []
while len(self._strategy_property[task_type]) <= task_id:
self._strategy_property[task_type].append(None)
self._strategy_property[task_type][task_id] = (
context._strategy.extended.experimental_should_init,
context.should_checkpoint,
context.should_save_summary)
def _run_mock_std_server(self,
session_config=None,
cluster_spec=None,
task_type=None,
task_id=None,
rpc_layer=None,
environment=None):
task_type = str(task_type)
task_id = task_id or 0
with self._lock:
if task_type not in self._std_servers:
self._std_servers[task_type] = []
while len(self._std_servers[task_type]) <= task_id:
self._std_servers[task_type].append(None)
server = MockServer()
self._std_servers[task_type][task_id] = server
return server
class DistributeCoordinatorTestStandaloneMode(DistributeCoordinatorTestBase):
def testInGraphStandaloneMode(self):
"""Test it runs in-graph replication in standalone client mode."""
distribute_coordinator.run_distribute_coordinator(
self._in_graph_worker_fn,
MockStrategy(between_graph=False),
cluster_spec=self._cluster_spec)
self.assertEqual(self._result_correct, 1)
def testBetweenGraph(self):
"""Test it runs between-graph replication in standalone client mode."""
distribute_coordinator.run_distribute_coordinator(
self._between_graph_worker_fn,
MockStrategy(between_graph=True),
cluster_spec=self._cluster_spec)
# Each finished worker will increment self._result_correct.
self.assertEqual(self._result_correct, NUM_WORKERS)
@test_util.run_v1_only("MonitoredSession removed from v2")
def testBetweenGraphWithMonitoredSession(self):
"""Test monitored session in standalone client mode."""
distribute_coordinator.run_distribute_coordinator(
self._between_graph_with_monitored_session,
MockStrategy(between_graph=True),
cluster_spec=self._cluster_spec)
# Each finished worker will increment self._result_correct.
self.assertEqual(self._result_correct, NUM_WORKERS)
def testBetweenGraphContext(self):
# Dumps the task contexts to the self._worker_context dict.
distribute_coordinator.run_distribute_coordinator(
self._dump_worker_context,
MockStrategy(between_graph=True),
cluster_spec=self._cluster_spec)
# There is only one type of task and there three such tasks.
self.assertEqual(len(self._worker_context), 1)
self.assertTrue(WORKER in self._worker_context)
self.assertEqual(len(self._worker_context[WORKER]), NUM_WORKERS)
# Check whether each task has the right master_target, num_workers, is_chief
# and distributed_mode.
self.assertEqual(
self._worker_context[WORKER][0],
(_bytes_to_str(self._workers[0].target), NUM_WORKERS, True, True))
self.assertEqual(
self._worker_context[WORKER][1],
(_bytes_to_str(self._workers[1].target), NUM_WORKERS, False, True))
self.assertEqual(
self._worker_context[WORKER][2],
(_bytes_to_str(self._workers[2].target), NUM_WORKERS, False, True))
def testBetweenGraphStrategyProperties(self):
# Dumps properties of the strategy objects.
distribute_coordinator.run_distribute_coordinator(
self._dump_strategy_property,
MockStrategy(between_graph=True, should_init=True),
cluster_spec=self._cluster_spec)
# There is only one type of task and there three such tasks.
self.assertEqual(len(self._strategy_property), 1)
self.assertTrue(WORKER in self._strategy_property)
self.assertEqual(len(self._strategy_property[WORKER]), NUM_WORKERS)
# Check whether each task has the right properties of should_init,
# should_checkpoint and should_save_summary.
self.assertEqual(self._strategy_property[WORKER][0], (True, True, True))
self.assertEqual(self._strategy_property[WORKER][1], (True, False, False))
self.assertEqual(self._strategy_property[WORKER][2], (True, False, False))
def testInGraphContext(self):
# Dumps the task contexts to the self._worker_context dict.
distribute_coordinator.run_distribute_coordinator(
self._dump_worker_context,
MockStrategy(between_graph=False),
cluster_spec=self._cluster_spec)
# There is only a "None" task in the dumped task context.
self.assertEqual(len(self._worker_context), 1)
self.assertTrue("None" in self._worker_context)
self.assertEqual(len(self._worker_context["None"]), 1)
# Check whether each task has the right master_target, num_workers, is_chief
# and distributed_mode.
self.assertEqual(
self._worker_context["None"][0],
(_bytes_to_str(self._workers[0].target), NUM_WORKERS, True, True))
def testLocalContext(self):
# Dumps the task contexts to the self._worker_context dict.
distribute_coordinator.run_distribute_coordinator(
self._dump_worker_context,
MockStrategy(between_graph=False),
cluster_spec=None)
# There is only a "None" task.
self.assertEqual(len(self._worker_context), 1)
self.assertTrue("None" in self._worker_context)
self.assertEqual(len(self._worker_context["None"]), 1)
# Check whether each task has the right master_target, num_workers, is_chief
# and distributed_mode.
self.assertEqual(self._worker_context["None"][0], ("", 0, True, False))
def testBetweenGraphContextWithChief(self):
# Adds a chief node, so there are NUM_WORKERS + 1 workers in total.
cluster_spec = copy.deepcopy(self._cluster_spec)
cluster_spec[CHIEF] = ["fake_chief"]
# Dumps the task contexts to the self._worker_context dict.
distribute_coordinator.run_distribute_coordinator(
self._dump_worker_context,
MockStrategy(between_graph=True),
cluster_spec=cluster_spec,
rpc_layer="grpc")
# There are one CHIEF and three workers.
self.assertEqual(len(self._worker_context), 2)
self.assertTrue(CHIEF in self._worker_context)
self.assertTrue(WORKER in self._worker_context)
self.assertEqual(len(self._worker_context[CHIEF]), 1)
self.assertEqual(len(self._worker_context[WORKER]), NUM_WORKERS)
# Check whether each task has the right master_target, num_workers, is_chief
# and distributed_mode.
self.assertEqual(self._worker_context[CHIEF][0],
("grpc://fake_chief", 4, True, True))
self.assertEqual(
self._worker_context[WORKER][0],
(_bytes_to_str(self._workers[0].target), NUM_WORKERS + 1, False, True))
self.assertEqual(
self._worker_context[WORKER][1],
(_bytes_to_str(self._workers[1].target), NUM_WORKERS + 1, False, True))
self.assertEqual(
self._worker_context[WORKER][2],
(_bytes_to_str(self._workers[2].target), NUM_WORKERS + 1, False, True))
def testInGraphContextWithEval(self):
# Adds a EVALUATOR job.
cluster_spec = copy.deepcopy(self._cluster_spec)
cluster_spec[EVALUATOR] = ["fake_evaluator"]
# Dumps the task contexts to the self._worker_context dict.
distribute_coordinator.run_distribute_coordinator(
self._dump_worker_context,
MockStrategy(between_graph=False),
cluster_spec=cluster_spec,
rpc_layer=None)
# There are one "None" task and one EVALUATOR task.
self.assertEqual(len(self._worker_context), 2)
self.assertTrue("None" in self._worker_context)
self.assertTrue(EVALUATOR in self._worker_context)
self.assertEqual(len(self._worker_context["None"]), 1)
self.assertEqual(len(self._worker_context[EVALUATOR]), 1)
# Check whether each task has the right master_target, num_workers, is_chief
# and distributed_mode.
self.assertEqual(self._worker_context["None"][0], (_strip_protocol(
_bytes_to_str(self._workers[0].target)), 3, True, True))
self.assertEqual(self._worker_context[EVALUATOR][0],
("fake_evaluator", 3, True, False))
class DistributeCoordinatorTestInpendentWorkerMode(
DistributeCoordinatorTestBase):
def testInGraph(self):
cluster_spec = self._create_cluster_spec(num_workers=NUM_WORKERS)
threads = self._run_multiple_coordinator_in_threads(
self._in_graph_worker_fn,
MockStrategy(between_graph=False),
cluster_spec,
mode=INDEPENDENT_WORKER)
self._join_threads([threads[WORKER][0]])
self.assertEqual(self._result_correct, 1)
def testBetweenGraph(self):
cluster_spec = self._create_cluster_spec(
num_workers=NUM_WORKERS, num_ps=NUM_PS)
threads = self._run_multiple_coordinator_in_threads(
self._between_graph_worker_fn,
MockStrategy(between_graph=True),
cluster_spec,
mode=INDEPENDENT_WORKER)
self._join_threads(threads[WORKER])
# Each finished worker will increment self._result_correct.
self.assertEqual(self._result_correct, NUM_WORKERS)
@test_util.run_v1_only("MonitoredSession removed from v2")
def testBetweenGraphWithMonitoredSession(self):
cluster_spec = self._create_cluster_spec(
num_workers=NUM_WORKERS, num_ps=NUM_PS)
threads = self._run_multiple_coordinator_in_threads(
self._between_graph_with_monitored_session,
MockStrategy(between_graph=True),
cluster_spec,
mode=INDEPENDENT_WORKER)
self._join_threads(threads[WORKER])
# Each finished worker will increment self._result_correct.
self.assertEqual(self._result_correct, NUM_WORKERS)
def testBetweenGraphContext(self):
cluster_spec = self._create_cluster_spec(num_workers=NUM_WORKERS)
# Dumps the task contexts and std server arguments.
with test.mock.patch.object(distribute_coordinator, "_run_std_server",
self._run_mock_std_server):
threads = self._run_multiple_coordinator_in_threads(
self._dump_worker_context,
MockStrategy(between_graph=True),
cluster_spec,
mode=INDEPENDENT_WORKER,
rpc_layer=None)
self._join_threads(threads[WORKER])
# There is only one type of task and three such tasks.
self.assertEqual(len(self._worker_context), 1)
self.assertTrue(WORKER in self._worker_context)
self.assertEqual(len(self._worker_context[WORKER]), NUM_WORKERS)
# Check whether each task has the right master_target, num_workers, is_chief
# and distributed_mode.
self.assertEqual(
self._worker_context[WORKER][0],
(_bytes_to_str(cluster_spec[WORKER][0]), NUM_WORKERS, True, True))
self.assertEqual(
self._worker_context[WORKER][1],
(_bytes_to_str(cluster_spec[WORKER][1]), NUM_WORKERS, False, True))
self.assertEqual(
self._worker_context[WORKER][2],
(_bytes_to_str(cluster_spec[WORKER][2]), NUM_WORKERS, False, True))
# Make sure each worker runs a std server.
self.assertEqual(len(self._std_servers), 1)
self.assertTrue(WORKER in self._std_servers)
self.assertEqual(len(self._std_servers[WORKER]), 3)
self.assertFalse(self._std_servers[WORKER][0].joined)
self.assertFalse(self._std_servers[WORKER][1].joined)
self.assertFalse(self._std_servers[WORKER][2].joined)
def testBetweenGraphStrategyProperties(self):
cluster_spec = self._create_cluster_spec(num_workers=NUM_WORKERS)
# Dumps properties of the strategy objects.
with test.mock.patch.object(distribute_coordinator, "_run_std_server",
self._run_mock_std_server):
threads = self._run_multiple_coordinator_in_threads(
self._dump_strategy_property,
MockStrategy(between_graph=True, should_init=True),
cluster_spec,
mode=INDEPENDENT_WORKER,
rpc_layer=None)
self._join_threads(threads[WORKER])
# There is only one type of task and there three such tasks.
self.assertEqual(len(self._strategy_property), 1)
self.assertTrue(WORKER in self._strategy_property)
self.assertEqual(len(self._strategy_property[WORKER]), NUM_WORKERS)
# Check whether each task has the right properties of should_init,
# should_checkpoint and should_save_summary.
self.assertEqual(self._strategy_property[WORKER][0], (True, True, True))
self.assertEqual(self._strategy_property[WORKER][1], (True, False, False))
self.assertEqual(self._strategy_property[WORKER][2], (True, False, False))
def testInGraphContext(self):
cluster_spec = self._create_cluster_spec(num_workers=NUM_WORKERS)
# Dumps the task contexts and std server arguments.
with test.mock.patch.object(distribute_coordinator, "_run_std_server",
self._run_mock_std_server):
threads = self._run_multiple_coordinator_in_threads(
self._dump_worker_context,
MockStrategy(between_graph=False),
cluster_spec,
mode=INDEPENDENT_WORKER,
rpc_layer=None)
self._join_threads(threads[WORKER])
# There is only a "None" task in the dumped task context.
self.assertEqual(len(self._worker_context), 1)
self.assertTrue("None" in self._worker_context)
self.assertEqual(len(self._worker_context["None"]), 1)
# Check whether each task has the right master_target, num_workers, is_chief
# and distributed_mode.
self.assertEqual(
self._worker_context["None"][0],
(_bytes_to_str(cluster_spec[WORKER][0]), NUM_WORKERS, True, True))
# Make sure each worker runs a std server.
self.assertEqual(len(self._std_servers), 1)
self.assertTrue(WORKER in self._std_servers)
self.assertEqual(len(self._std_servers[WORKER]), 3)
self.assertFalse(self._std_servers[WORKER][0].joined)
self.assertTrue(self._std_servers[WORKER][1].joined)
self.assertTrue(self._std_servers[WORKER][2].joined)
def testInGraphContextWithEval(self):
# Adds a EVALUATOR job.
cluster_spec = self._create_cluster_spec(
num_workers=NUM_WORKERS, has_eval=True)
# Dumps the task contexts and std server arguments.
with test.mock.patch.object(distribute_coordinator, "_run_std_server",
self._run_mock_std_server):
threads = self._run_multiple_coordinator_in_threads(
self._dump_worker_context,
MockStrategy(between_graph=False),
cluster_spec,
mode=INDEPENDENT_WORKER,
rpc_layer=None)
self._join_threads(threads[WORKER])
self._join_threads([threads[EVALUATOR][0]])
# There are one "None" task and one EVALUATOR task.
self.assertEqual(len(self._worker_context), 2)
self.assertTrue("None" in self._worker_context)
self.assertTrue(EVALUATOR in self._worker_context)
self.assertEqual(len(self._worker_context["None"]), 1)
self.assertEqual(len(self._worker_context[EVALUATOR]), 1)
# Check whether each task has the right master_target, num_workers, is_chief
# and distributed_mode.
self.assertEqual(self._worker_context["None"][0],
(_bytes_to_str(cluster_spec[WORKER][0]), 3, True, True))
self.assertEqual(self._worker_context[EVALUATOR][0],
(cluster_spec[EVALUATOR][0], 3, True, False))
# Make sure each worker runs a std server.
self.assertEqual(len(self._std_servers), 2)
self.assertTrue(WORKER in self._std_servers)
self.assertTrue(EVALUATOR in self._std_servers)
self.assertEqual(len(self._std_servers[WORKER]), 3)
self.assertEqual(len(self._std_servers[EVALUATOR]), 1)
self.assertFalse(self._std_servers[WORKER][0].joined)
self.assertTrue(self._std_servers[WORKER][1].joined)
self.assertTrue(self._std_servers[WORKER][2].joined)
self.assertFalse(self._std_servers[EVALUATOR][0].joined)
def testRunStdServerInGoogleEnvironment(self):
cluster_spec = {"worker": ["fake_worker"], "ps": ["localhost:0"]}
tf_config = {"cluster": cluster_spec, "environment": "google"}
joined = [False]
def _fake_sleep(_):
joined[0] = True
original_sys_exit(0)
def _thread_fn(cluster_spec):
distribute_coordinator.run_distribute_coordinator(
None,
MockStrategy(between_graph=True),
mode=INDEPENDENT_WORKER,
cluster_spec=cluster_spec,
task_type="ps",
task_id=0)
with test.mock.patch.dict(
"os.environ",
{"TF_CONFIG": json.dumps(tf_config)}), test.mock.patch.object(
time, "sleep", _fake_sleep):
t = threading.Thread(target=_thread_fn, args=(cluster_spec,))
t.start()
t.join()
self.assertTrue(joined[0])
def testRpcLayerEnvironmentVariable(self):
cluster_spec = {"worker": ["fake_worker"], "ps": ["fake_ps"]}
tf_config = {"cluster": cluster_spec, "rpc_layer": "cake"}
rpc_layer_from_coordinator = [None]
def _run_mock_server(cluster_spec=None,
task_type=None,
task_id=None,
session_config=None,
rpc_layer=None,
environment=None):
del cluster_spec, task_type, task_id, session_config, environment
rpc_layer_from_coordinator[0] = rpc_layer
return MockServer()
with test.mock.patch.dict(
"os.environ",
{"TF_CONFIG": json.dumps(tf_config)}), test.mock.patch.object(
distribute_coordinator, "_run_std_server", _run_mock_server):
distribute_coordinator.run_distribute_coordinator(
None,
MockStrategy(between_graph=True),
mode=INDEPENDENT_WORKER,
cluster_spec=cluster_spec,
task_type="ps",
task_id=0)
self.assertEqual(rpc_layer_from_coordinator[0], "cake")
class StrategyConfigureTest(test.TestCase):
def setUp(self):
self._device_filters = []
self._intra_op_parallelism_threads = None
self._inter_op_parallelism_threads = None
super(StrategyConfigureTest, self).setUp()
def _dump_device_filters(self, *args, **kwargs):
session_config = kwargs.get("session_config", None)
self._device_filters.extend(session_config.device_filters)
self._intra_op_parallelism_threads = (
session_config.intra_op_parallelism_threads)
self._inter_op_parallelism_threads = (
session_config.inter_op_parallelism_threads)
return MockServer()
def _worker_fn(self, strategy):
worker_context = distribute_coordinator_context.get_current_worker_context()
session_config = worker_context._session_config
self._device_filters.extend(session_config.device_filters)
self._intra_op_parallelism_threads = (
session_config.intra_op_parallelism_threads)
self._inter_op_parallelism_threads = (
session_config.inter_op_parallelism_threads)
return MockServer()
def test_session_config_in_std_server(self):
cluster_spec = {"worker": ["fake_worker"], "ps": ["fake_ps"]}
tf_config = {"cluster": cluster_spec}
with test.mock.patch.dict(
"os.environ",
{"TF_CONFIG": json.dumps(tf_config)}), test.mock.patch.object(
distribute_coordinator, "_run_std_server",
self._dump_device_filters):
distribute_coordinator.run_distribute_coordinator(
lambda _: None,
MockStrategy(between_graph=True),
mode=INDEPENDENT_WORKER,
cluster_spec=cluster_spec,
task_type="worker",
task_id=0)
self.assertEqual(self._intra_op_parallelism_threads, 1)
self.assertEqual(self._inter_op_parallelism_threads, 0)
def test_session_config_in_session_creator(self):
cluster_spec = {"worker": ["localhost:0"]}
tf_config = {"cluster": cluster_spec}
# Reset the saved Server state.
distribute_coordinator._thread_local = threading.local() # pylint: disable=protected-access
with test.mock.patch.dict("os.environ",
{"TF_CONFIG": json.dumps(tf_config)}):
distribute_coordinator.run_distribute_coordinator(
self._worker_fn,
MockStrategy(between_graph=True),
mode=INDEPENDENT_WORKER,
cluster_spec=cluster_spec,
task_type="worker",
task_id=0)
self.assertEqual(self._device_filters, ["/job:worker/task:0", "/job:ps"])
self.assertEqual(self._intra_op_parallelism_threads, 2)
self.assertEqual(self._inter_op_parallelism_threads, 0)
def test_eval_strategy_configure(self):
cluster_spec = {"evaluator": ["localhost:0"]}
tf_config = {"cluster": cluster_spec}
with test.mock.patch.dict("os.environ",
{"TF_CONFIG": json.dumps(tf_config)}):
distribute_coordinator.run_distribute_coordinator(
lambda _: None,
MockStrategy(between_graph=False),
eval_fn=self._worker_fn,
eval_strategy=MockStrategy(between_graph=True),
mode=INDEPENDENT_WORKER,
cluster_spec=cluster_spec,
task_type="evaluator",
task_id=0)
self.assertEqual(self._device_filters, ["/job:somejob"])
self.assertEqual(self._intra_op_parallelism_threads, 0)
self.assertEqual(self._inter_op_parallelism_threads, 2)
class RunStandardTensorflowServerTest(test.TestCase):
def test_std_server_arguments(self):
cs = {"worker": ["fake_worker"], "ps": ["fake_ps"]}
tf_config = {"cluster": cs, "task": {"type": "ps", "id": 0}}
def _mock_run_std_server(cluster_spec=None,
task_type=None,
task_id=None,
session_config=None,
rpc_layer=None):
self.assertEqual(cluster_spec.as_dict(), cs)
self.assertEqual(task_type, "ps")
self.assertEqual(task_id, 0)
self.assertEqual(session_config.experimental.collective_group_leader,
"/job:worker/replica:0/task:0")
self.assertEqual(session_config.intra_op_parallelism_threads, 1)
self.assertEqual(rpc_layer, "grpc")
return MockServer()
with test.mock.patch.dict(
"os.environ",
{"TF_CONFIG": json.dumps(tf_config)}), test.mock.patch.object(
distribute_coordinator, "_run_std_server", _mock_run_std_server):
session_config = config_pb2.ConfigProto()
session_config.intra_op_parallelism_threads = 1
mock_server = distribute_coordinator.run_standard_tensorflow_server(
session_config)
self.assertTrue(mock_server.started)
if __name__ == "__main__":
# TODO(yuefengz): find a smart way to terminite std server threads.
with test.mock.patch.object(sys, "exit", os._exit):
# Reduce `recovery_wait_secs` from 30 seconds so the test completes quickly.
orig_init = session_manager.SessionManager.__init__
def new_init(*args, **kwargs):
kwargs.pop("recovery_wait_secs", None)
kwargs["recovery_wait_secs"] = 0.5
orig_init(*args, **kwargs)
session_manager.SessionManager.__init__ = new_init
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/distribute_coordinator_test.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Library for testing DistributionStrategy descendants."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import tempfile
import numpy as np
from tensorflow.core.protobuf import config_pb2
from tensorflow.core.util import event_pb2
from tensorflow.python.client import session as session_lib
from tensorflow.python.data.ops import dataset_ops
from tensorflow.python.distribute import distribution_strategy_context as ds_context
from tensorflow.python.distribute import reduce_util
from tensorflow.python.distribute import values
from tensorflow.python.eager import backprop
from tensorflow.python.eager import context
from tensorflow.python.eager import def_function
from tensorflow.python.eager import test
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import errors
from tensorflow.python.framework import ops
from tensorflow.python.framework import test_util
from tensorflow.python.layers import core
from tensorflow.python.lib.io import tf_record
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gradients_impl
from tensorflow.python.ops import init_ops
from tensorflow.python.ops import summary_ops_v2 as summary_ops
from tensorflow.python.ops import variable_scope
from tensorflow.python.ops import variables
from tensorflow.python.platform import gfile
from tensorflow.python.training import optimizer
from tensorflow.python.training import training_util
from tensorflow.python.util import nest
class _TestException(Exception):
pass
# May be the argument to either distribution.extended.call_for_each_replica() or
# get_replica_context().merge_call()
def _raise_exception_fn(_=None):
raise _TestException()
# Must be the argument to a distribution.extended.call_for_each_replica() call,
# calls a get_replica_context().merge_call() that raises an exception.
def _merge_raises_fn():
ds_context.get_replica_context().merge_call(_raise_exception_fn)
# Must be the argument to a get_replica_context().merge_call() call, calls
# dist.extended.call_for_each_replica() with a function that raises an
# exception.
def _call_raises_fn(dist):
dist.extended.call_for_each_replica(_raise_exception_fn)
# Must be the argument to a distribution.extended.call_for_each_replica() call,
# calls a get_replica_context().merge_call() that calls a
# call_for_each_replica() that raises an exception.
def _merge_call_raises_fn():
ds_context.get_replica_context().merge_call(_call_raises_fn)
# Must be the argument to a get_replica_context().merge_call() call, calls
# dist.extended.call_for_each_replica() with a function that calls a
# get_replica_context().merge_call() that raises an exception.
def _call_merge_raises_fn(dist):
dist.extended.call_for_each_replica(_merge_raises_fn)
# Must be the argument to a distribution.extended.call_for_each_replica() call,
# calls a get_replica_context().merge_call() that calls a
# call_for_each_replica() that calls a get_replica_context().merge_call() that
# raises an exception.
def _merge_call_merge_raises_fn():
ds_context.get_replica_context().merge_call(_call_merge_raises_fn)
def _events_from_logdir(test_case, logdir):
"""Reads summary events from log directory."""
test_case.assertTrue(gfile.Exists(logdir))
files = gfile.ListDirectory(logdir)
test_case.assertLen(files, 1)
records = list(tf_record.tf_record_iterator(os.path.join(logdir, files[0])))
result = []
for r in records:
event = event_pb2.Event()
event.ParseFromString(r)
result.append(event)
return result
class DistributionTestBase(test.TestCase):
"""Some tests that should work with any DistributionStrategy."""
def _test_minimize_loss_eager(self, d):
with d.scope():
l = core.Dense(1, use_bias=False)
def loss(x):
y = array_ops.reshape(l(x), []) - constant_op.constant(1.)
return y * y
# TODO(isaprykin): Extract implicit_grad+get_filtered_grad_fn into a
# common `implicit_grad` function and put it in DistributionStrategy.
grad_fn = backprop.implicit_grad(loss)
grad_fn = optimizer.get_filtered_grad_fn(grad_fn)
def update(v, g):
return v.assign_sub(0.2 * g)
one = constant_op.constant([[1.]])
def step():
"""Perform one optimization step."""
# Run forward & backward to get gradients, variables list.
g_v = d.extended.call_for_each_replica(grad_fn, args=(one,))
# Update the variables using the gradients and the update() function.
before_list = []
after_list = []
for g, v in g_v:
fetched = d.extended.read_var(v)
before_list.append(fetched)
# control_dependencies irrelevant but harmless in eager execution
with ops.control_dependencies([fetched]):
g = d.extended.reduce_to(
reduce_util.ReduceOp.SUM, g, destinations=v)
with ops.control_dependencies(
d.extended.update(v, update, args=(g,), group=False)):
after_list.append(d.extended.read_var(v))
return before_list, after_list
for i in range(10):
b, a = step()
if i == 0:
before, = b # pylint: disable=unbalanced-tuple-unpacking
after, = a # pylint: disable=unbalanced-tuple-unpacking
error_before = abs(before.numpy() - 1)
error_after = abs(after.numpy() - 1)
# Error should go down
self.assertLess(error_after, error_before)
def _test_minimize_loss_graph(self,
d,
soft_placement=False,
learning_rate=0.2):
config = config_pb2.ConfigProto()
config.allow_soft_placement = soft_placement
config.gpu_options.per_process_gpu_memory_fraction = 0.3
with context.graph_mode(), \
ops.Graph().as_default(), \
self.cached_session(config=config) as sess, \
d.scope():
l = core.Dense(1, use_bias=False)
def loss(x):
y = array_ops.reshape(l(x), []) - constant_op.constant(1.)
return y * y
grad_fn = backprop.implicit_grad(loss)
def update(v, g):
return v.assign_sub(learning_rate * g)
one = constant_op.constant([[1.]])
def step():
"""Perform one optimization step."""
# Run forward & backward to get gradients, variables list.
g_v = d.extended.call_for_each_replica(grad_fn, args=(one,))
# Update the variables using the gradients and the update() function.
before_list = []
after_list = []
for g, v in g_v:
fetched = d.extended.read_var(v)
before_list.append(fetched)
with ops.control_dependencies([fetched]):
g = d.extended.reduce_to(
reduce_util.ReduceOp.SUM, g, destinations=v)
with ops.control_dependencies(
d.extended.update(v, update, args=(g,), group=False)):
after_list.append(d.extended.read_var(v))
return before_list, after_list
before_out, after_out = step()
variables.global_variables_initializer().run()
for i in range(10):
b, a = sess.run((before_out, after_out))
if i == 0:
before, = b
after, = a
error_before = abs(before - 1)
error_after = abs(after - 1)
# Error should go down
self.assertLess(error_after, error_before)
def _test_summary_for_replica_zero_only(self, d):
logdir = tempfile.mkdtemp()
def run_fn():
"""Function executed for each replica."""
with summary_writer.as_default():
replica_id = ds_context.get_replica_context().replica_id_in_sync_group
return summary_ops.write("a", replica_id)
with self.cached_session() as sess, d.scope(), \
summary_ops.always_record_summaries():
# We need global_step because summary writing op *always* has global_step
# as input, even when we always record summary or never record summary.
global_step = training_util.get_or_create_global_step()
if not context.executing_eagerly():
# When executing eagerly, variables are initialized immediately after
# creation, and its initializer will be None.
global_step.initializer.run()
summary_ops.set_step(0)
summary_writer = summary_ops.create_file_writer(logdir)
output = d.extended.call_for_each_replica(run_fn)
unwrapped = d.unwrap(output)
if not context.executing_eagerly():
sess.run(summary_writer.init())
sess.run(unwrapped)
sess.run(summary_writer.close())
events = _events_from_logdir(self, logdir)
# There will be 2 entries: 1 summary file header entry, and 1 entry
# written by replica 0.
self.assertLen(events, 2)
self.assertEqual(events[1].summary.value[0].tag, "a")
self.assertEqual(events[1].summary.value[0].simple_value, 0.0)
def _test_replica_id(self, d):
with d.scope():
expected_devices = [False] * len(d.extended.worker_devices)
def mark_devices_fn():
replica_id = self.evaluate(
ds_context.get_replica_context().replica_id_in_sync_group)
self.assertLess(replica_id, len(d.extended.worker_devices))
self.assertFalse(expected_devices[replica_id])
expected_devices[replica_id] = True
d.extended.call_for_each_replica(mark_devices_fn)
self.assertAllEqual(expected_devices,
[True] * len(d.extended.worker_devices))
def _test_call_and_merge_exceptions(self, dist):
with dist.scope():
with self.assertRaises(_TestException):
dist.extended.call_for_each_replica(_raise_exception_fn)
with self.assertRaises(_TestException):
dist.extended.call_for_each_replica(_merge_raises_fn)
with self.assertRaises(_TestException):
dist.extended.call_for_each_replica(_merge_call_raises_fn)
with self.assertRaises(_TestException):
dist.extended.call_for_each_replica(_merge_call_merge_raises_fn)
def _input_fn_to_test_input_context(self, dataset_or_callable_fn,
expected_num_replicas_in_sync,
expected_num_input_pipelines,
expected_input_pipeline_id):
# Use a list of one element as counter so that it can be captured by the
# `_input_fn`. This counter is incremented by 1 each time an input_fn is
# called. We use this counter to check whether the `input_pipeline_id`
# matches the counter in the in-graph replication.
worker_id_counter = [0]
def _input_fn(input_context):
"""Input fn for testing."""
self.assertIsNotNone(input_context)
self.assertEqual(expected_num_replicas_in_sync,
input_context.num_replicas_in_sync)
self.assertEqual(expected_num_input_pipelines,
input_context.num_input_pipelines)
if expected_input_pipeline_id is not None:
self.assertEqual(expected_input_pipeline_id,
input_context.input_pipeline_id)
else:
self.assertEqual(worker_id_counter[0], input_context.input_pipeline_id)
worker_id_counter[0] += 1
return dataset_or_callable_fn()
return _input_fn
def _test_input_fn_iterable(
self, strategy, input_fn, expected_values, ignore_order=False):
if not context.executing_eagerly():
self.skipTest("Only supported with eager execution.")
assert_same = self.assertCountEqual if ignore_order else self.assertEqual
iterable = strategy.experimental_distribute_datasets_from_function(input_fn)
iterator = iter(iterable)
for expected_value in expected_values:
computed_value = self.evaluate(
list(strategy.experimental_local_results(next(iterator))))
assert_same(expected_value, computed_value)
with self.assertRaises(StopIteration):
self.evaluate(strategy.experimental_local_results(next(iterator)))
# After re-initializing the iterator, should be able to iterate again.
iterator = iter(iterable)
for expected_value in expected_values:
computed_value = self.evaluate(
list(strategy.experimental_local_results(next(iterator))))
assert_same(expected_value, computed_value)
def _test_input_fn_iterator(self,
iterator,
devices,
expected_values,
sess=None,
test_reinitialize=True,
ignore_order=False):
evaluate = lambda x: sess.run(x) if sess else self.evaluate(x)
evaluate(iterator.initialize())
for expected_value in expected_values:
next_element = iterator.get_next()
computed_value = evaluate(
[values.select_replica(r, next_element) for r in range(len(devices))])
if ignore_order:
self.assertCountEqual(expected_value, computed_value)
else:
self.assertEqual(expected_value, computed_value)
with self.assertRaises(errors.OutOfRangeError):
next_element = iterator.get_next()
evaluate(
[values.select_replica(r, next_element) for r in range(len(devices))])
# After re-initializing the iterator, should be able to iterate again.
if test_reinitialize:
evaluate(iterator.initialize())
for expected_value in expected_values:
next_element = iterator.get_next()
computed_value = evaluate([
values.select_replica(r, next_element) for r in range(len(devices))
])
if ignore_order:
self.assertCountEqual(expected_value, computed_value)
else:
self.assertEqual(expected_value, computed_value)
def _test_global_step_update(self, strategy):
with strategy.scope():
global_step = variable_scope.get_variable(
"global_step",
shape=[],
dtype=dtypes.int64,
initializer=init_ops.zeros_initializer(),
trainable=False,
aggregation=variables.VariableAggregation.ONLY_FIRST_REPLICA)
self.evaluate(variables.global_variables_initializer())
def model_fn():
train_op = global_step.assign_add(1)
value = global_step.read_value()
return train_op, value
train_ops, value = strategy.extended.call_for_each_replica(model_fn)
self.evaluate(strategy.group(train_ops))
global_step_tensors = strategy.experimental_local_results(value)
global_step_values = self.evaluate(global_step_tensors)
self.assertEqual((1,) * len(global_step_tensors), global_step_values)
def _test_numpy_dataset(self, strategy):
with strategy.scope(), self.cached_session() as sess:
x = np.asarray([[1, 2], [6, 12], [2, 4], [5, 10], [3, 6], [4, 8]])
y = np.asarray([5, 4, 3, 2, 1, 0])
batch_size = 6
if not strategy.extended._global_batch_size: # pylint: disable=protected-access
batch_size = batch_size // strategy.num_replicas_in_sync
ds = strategy.extended.experimental_make_numpy_dataset((x, y),
session=sess)
ds = ds.repeat(2) # 2 epochs
# We need to use the drop_remainder argument to get a known static
# input shape which is required for TPUs.
drop_remainder = strategy.extended.experimental_require_static_shapes
ds = ds.batch(batch_size, drop_remainder=drop_remainder)
i = strategy.make_dataset_iterator(ds)
self.evaluate(i.initialize())
def run_and_concatenate(strategy, i):
x, y = strategy.experimental_run(lambda z: z, i)
x, y = self.evaluate((strategy.experimental_local_results(x),
strategy.experimental_local_results(y)))
return np.concatenate(x), np.concatenate(y)
x_1, y_1 = run_and_concatenate(strategy, i)
self.assertAllEqual(x, x_1)
self.assertAllEqual(y, y_1)
x_2, y_2 = run_and_concatenate(strategy, i)
self.assertAllEqual(x, x_2)
self.assertAllEqual(y, y_2)
with self.assertRaises(errors.OutOfRangeError):
run_and_concatenate(strategy, i)
def _test_trainable_variable(self, strategy):
for cls in [variables.VariableV1, variables.Variable]:
with strategy.scope():
v1 = cls(1.0)
self.assertEqual(True, v1.trainable)
v2 = cls(1.0, synchronization=variables.VariableSynchronization.ON_READ)
self.assertEqual(False, v2.trainable)
v3 = cls(1.0, synchronization=variables.VariableSynchronization.ON_READ,
trainable=True)
self.assertEqual(True, v3.trainable)
v4 = cls(1.0, synchronization=variables.VariableSynchronization.ON_READ,
trainable=False)
self.assertEqual(False, v4.trainable)
class OneDeviceDistributionTestBase(test.TestCase):
"""Some tests that should work with any one-device DistributionStrategy."""
def _test_run(self, strategy):
out1 = strategy.experimental_run_v2(lambda: constant_op.constant(4.))
self.assertAllEqual([4.], self.evaluate(strategy.unwrap(out1)))
out2 = strategy.experimental_run_v2(
lambda x: {"a": x * 2, "b": x * x}, args=(out1,))
out2_vals = self.evaluate(nest.map_structure(strategy.unwrap, out2))
self.assertAllEqual([8.], out2_vals["a"])
self.assertAllEqual([16.], out2_vals["b"])
out3 = strategy.experimental_run_v2(lambda b, a: a + 2 * b + 2, kwargs=out2)
self.assertAllEqual([42.], self.evaluate(strategy.unwrap(out3)))
def _test_all_reduce_sum(self, strategy):
self._test_collective_comms(
strategy, _all_sum, inputs=(4., [42., 43.]), expected=(4., [42., 43.]))
def _test_all_reduce_sum_gradients(self, strategy):
self._test_collective_comms_gradients(
strategy, _all_sum, inputs=[4.], expected_grads=[4.])
def _test_all_reduce_sum_gradient_tape(self, strategy):
self._test_collective_comms_gradient_tape(
strategy, _all_sum, inputs=[4.], expected_grads=[4.])
def _test_all_reduce_mean(self, strategy):
self._test_collective_comms(
strategy, _all_mean, inputs=(2., [21., 22.]), expected=(2., [21., 22.]))
def _test_all_reduce_mean_gradients(self, strategy):
self._test_collective_comms_gradients(
strategy, _all_mean, inputs=[5.], expected_grads=[5.])
def _test_all_reduce_mean_gradient_tape(self, strategy):
self._test_collective_comms_gradient_tape(
strategy, _all_mean, inputs=[5.], expected_grads=[5.])
def _test_collective_comms(self, strategy, comm_fn, inputs, expected):
inputs = strategy.make_input_fn_iterator(
lambda _: dataset_ops.Dataset.from_tensors(inputs))
self.evaluate(inputs.initialize())
outputs = self.evaluate(
list(
map(strategy.experimental_local_results,
strategy.experimental_run(comm_fn, inputs))))
self.assertAllEqual([expected[0]], outputs[0])
self.assertAllEqual([expected[1]], outputs[1])
def _test_collective_comms_gradients(self, strategy, comm_fn, inputs,
expected_grads):
if context.executing_eagerly():
self.skipTest("`tf.gradients` is not supported with eager execution.")
def step(c):
x = constant_op.constant(42.)
y = comm_fn(x) * c
return gradients_impl.gradients(y, [x])[0]
inputs = strategy.make_input_fn_iterator(
lambda _: dataset_ops.Dataset.from_tensors(inputs))
self.evaluate(inputs.initialize())
self.assertAllEqual(
expected_grads,
self.evaluate(
strategy.experimental_local_results(
strategy.experimental_run(step, inputs))))
def _test_collective_comms_gradient_tape(self, strategy, comm_fn, inputs,
expected_grads):
def step(c):
x = constant_op.constant(42.)
with backprop.GradientTape() as tape:
tape.watch(x)
y = comm_fn(x) * c
return tape.gradient(y, x)
inputs = strategy.make_input_fn_iterator(
lambda _: dataset_ops.Dataset.from_tensors(inputs))
self.evaluate(inputs.initialize())
self.assertAllEqual(
expected_grads,
self.evaluate(
strategy.experimental_local_results(
strategy.experimental_run(step, inputs))))
def _test_device_and_input_device_are_colocated(self, strategy):
if context.executing_eagerly():
self.skipTest(
"cross-device tests are not supported with eager execution.")
workers, _ = test_util.create_local_cluster(2, 0)
inputs = strategy.make_input_fn_iterator(
lambda _: dataset_ops.Dataset.range(5))
comm_fn = lambda x: x + 1
run_op = strategy.experimental_run(comm_fn, inputs)
with session_lib.Session(target=workers[1].target) as sess:
sess.run(inputs.initialize())
sess.run(run_op)
def _test_device_and_input_device_are_colocated_with_function(self, strategy):
if context.executing_eagerly():
self.skipTest(
"cross-device tests are not supported with eager execution.")
workers, _ = test_util.create_local_cluster(2, 0)
inputs = strategy.make_input_fn_iterator(
lambda _: dataset_ops.Dataset.range(5))
comm_fn = lambda x: x + 1
experimental_run = def_function.function()(strategy.experimental_run)
with ops.device("/job:worker/replica:0/task:1/device:CPU:0"):
# The tf.function must be defined on the right device as well.
run_op = experimental_run(comm_fn, inputs)
with session_lib.Session(target=workers[1].target) as sess:
sess.run(inputs.initialize())
sess.run(run_op)
class TwoDeviceDistributionTestBase(test.TestCase):
"""Some tests that should work with any two-device DistributionStrategy."""
def _test_run(self, strategy):
out1 = strategy.experimental_run_v2(
lambda: ds_context.get_replica_context().replica_id_in_sync_group + 1)
self.assertAllEqual([1, 2], self.evaluate(strategy.unwrap(out1)))
out2 = strategy.experimental_run_v2(
lambda x: {"a": x * 2, "b": x * x}, args=(out1,))
out2_vals = self.evaluate(nest.map_structure(strategy.unwrap, out2))
self.assertAllEqual([2, 4], out2_vals["a"])
self.assertAllEqual([1, 4], out2_vals["b"])
out3 = strategy.experimental_run_v2(lambda b, a: a + 2 * b + 2, kwargs=out2)
self.assertAllEqual([6, 14], self.evaluate(strategy.unwrap(out3)))
def _test_all_reduce_sum(self, strategy):
self._test_collective_comms(
strategy,
_all_sum,
inputs=([1., 3.], [[39., 2.], [3., 41.]]),
expected=(4., [42., 43.]))
def _test_all_reduce_sum_gradients(self, strategy):
self._test_collective_comms_gradients(
strategy, _all_sum, inputs=[1., 3.], expected_grads=[4., 4.])
def _test_all_reduce_sum_gradient_tape(self, strategy):
self._test_collective_comms_gradient_tape(
strategy, _all_sum, inputs=[1., 3.], expected_grads=[4., 4.])
def _test_all_reduce_mean(self, strategy):
self._test_collective_comms(
strategy,
_all_mean,
inputs=([1., 3.], [[39., 2.], [3., 41.]]),
expected=(2., [21., 21.5]))
def _test_all_reduce_mean_gradients(self, strategy):
self._test_collective_comms_gradients(
strategy, _all_mean, inputs=[1., 3.], expected_grads=[2., 2.])
def _test_all_reduce_mean_gradient_tape(self, strategy):
self._test_collective_comms_gradient_tape(
strategy, _all_mean, inputs=[1., 3.], expected_grads=[2., 2.])
def _test_collective_comms(self, strategy, comm_fn, inputs, expected):
inputs = strategy.make_input_fn_iterator(
lambda _: dataset_ops.Dataset.from_tensor_slices(inputs))
self.evaluate(inputs.initialize())
outputs = self.evaluate(
list(
map(strategy.experimental_local_results,
strategy.experimental_run(comm_fn, inputs))))
self.assertAllEqual([expected[0], expected[0]], outputs[0])
self.assertAllEqual([expected[1], expected[1]], outputs[1])
def _test_collective_comms_gradients(self, strategy, comm_fn, inputs,
expected_grads):
if context.executing_eagerly():
self.skipTest("`tf.gradients` is not supported with eager execution.")
def step(c):
x = constant_op.constant(42.)
y = comm_fn(x) * c
return gradients_impl.gradients(y, [x])[0]
inputs = strategy.make_input_fn_iterator(
lambda _: dataset_ops.Dataset.from_tensor_slices(inputs))
self.evaluate(inputs.initialize())
self.assertAllEqual(
expected_grads,
self.evaluate(
strategy.experimental_local_results(
strategy.experimental_run(step, inputs))))
def _test_collective_comms_gradient_tape(self, strategy, comm_fn, inputs,
expected_grads):
def step(c):
x = constant_op.constant(42.)
with backprop.GradientTape() as tape:
tape.watch(x)
y = comm_fn(x) * c
return tape.gradient(y, x)
inputs = strategy.make_input_fn_iterator(
lambda _: dataset_ops.Dataset.from_tensor_slices(inputs))
self.evaluate(inputs.initialize())
self.assertAllEqual(
expected_grads,
self.evaluate(
strategy.experimental_local_results(
strategy.experimental_run(step, inputs))))
def _all_sum(value):
ctx = ds_context.get_replica_context()
return ctx.all_reduce(reduce_util.ReduceOp.SUM, value)
def _all_mean(value):
ctx = ds_context.get_replica_context()
return ctx.all_reduce(reduce_util.ReduceOp.MEAN, value)
|
tensorflow-master
|
tensorflow/python/distribute/strategy_test_lib.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""The context retrieval method for distribute coordinator."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import threading
_worker_context = threading.local()
def get_current_worker_context():
"""Returns the current task context."""
try:
return _worker_context.current
except AttributeError:
return None
|
tensorflow-master
|
tensorflow/python/distribute/distribute_coordinator_context.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""This module customizes `test_combinations` for Tensorflow.
Additionally it provides `generate()`, `combine()` and `times()` with Tensorflow
customizations as a default.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import functools
import sys
from tensorflow.python.distribute import test_combinations
from tensorflow.python.eager import context
from tensorflow.python.framework import ops
# TODO(rchao): Rename `distribution` parameter to `strategy` or
# `distribute_strategy` in all tests.
class DistributionParameter(test_combinations.ParameterModifier):
"""Transforms arguments of type `NamedDistribution`.
Convert all arguments of type `NamedDistribution` to the value of their
`strategy` property.
"""
def modified_arguments(self, kwargs, requested_parameters):
del requested_parameters
distribution_arguments = {}
for k, v in kwargs.items():
if isinstance(v, NamedDistribution):
distribution_arguments[k] = v.strategy
return distribution_arguments
class NamedGPUCombination(test_combinations.TestCombination):
"""Enable tests to request GPU hardware and skip non-GPU combinations.
This class expects test_combinations to be genarated with `NamedDistribution`
wrapping instances of `tf.distribute.Strategy`.
Optionally, the `required_gpus` argument is supported. GPU hardware is
required, if its value is `True` or > 0.
Attributes:
GPU_TEST: The environment is considered to have GPU hardware available if
the name of the program contains "test_gpu".
"""
GPU_TEST = "test_gpu" in sys.argv[0]
def should_execute_combination(self, kwargs):
distributions = [
v for v in kwargs.values() if isinstance(v, NamedDistribution)
]
required_gpus = kwargs.get("required_gpus", None)
if distributions and required_gpus:
raise ValueError("Do not use `required_gpus` and arguments of type "
"NamedDistribution together.")
number_of_required_gpus = max([required_gpus or 0] +
[d.required_gpus or 0 for d in distributions])
if not number_of_required_gpus and GPUCombination.GPU_TEST:
return (False, "Test that doesn't require GPUs.")
elif context.num_gpus() < number_of_required_gpus:
return (False, ("Only {} of {} required GPUs are available.".format(
context.num_gpus(), number_of_required_gpus)))
else:
return (True, None)
def parameter_modifiers(self):
return [test_combinations.OptionalParameter("required_gpus")]
class GPUCombination(NamedGPUCombination):
"""NamedGPUCombination that passes `tf.distribute.Strategy` to the tests."""
def parameter_modifiers(self):
return [DistributionParameter()
] + NamedGPUCombination.parameter_modifiers(self)
class NamedTPUCombination(test_combinations.TestCombination):
"""Allow to request TPU hardware and skip non-TPU combinations.
This class expects test_combinations to be genarated with `NamedDistribution`
wrapping instances of `tf.distribute.Strategy`.
Optionally, the `required_tpus` parameter is supported. TPU hardware is
required, if its argument is `True` or > 0.
Attributes:
TPU_TEST: The environment is considered to have GPU hardware available if
the name of the program contains "test_gpu".
"""
TPU_TEST = "test_tpu" in sys.argv[0]
def should_execute_combination(self, kwargs):
distributions = [
v for v in kwargs.values() if isinstance(v, NamedDistribution)
]
# TODO(isaprykin): Migrate all tests away from using 'required_tpu' in favor
# of 'required_tpus'.
if "required_tpus" in kwargs and "required_tpu" in kwargs:
raise ValueError("Do not use `required_tpu`. Both `required_tpus` and "
"`required_tpu` were specified.")
required_tpus = kwargs.get("required_tpus", None) or kwargs.get(
"required_tpu", None)
if distributions and required_tpus:
raise ValueError("Do not use `required_tpus` and arguments of type "
"NamedDistribution together.")
# TODO(isaprykin): Add support for a particular number of TPUs. Right now
# it's binary.
number_of_required_tpus = max([required_tpus or 0] +
[d.required_tpu or 0 for d in distributions])
if not number_of_required_tpus and TPUCombination.TPU_TEST:
return (False, "Test that doesn't require TPUs.")
elif number_of_required_tpus and not TPUCombination.TPU_TEST:
return (False, "Test requires a TPU, but it's not available.")
else:
return (True, None)
def parameter_modifiers(self):
return [
test_combinations.OptionalParameter("required_tpus"),
test_combinations.OptionalParameter("required_tpu")
]
class TPUCombination(NamedTPUCombination):
"""NamedTPUCombination that passes `tf.distribute.Strategy` to the tests."""
def parameter_modifiers(self):
return [DistributionParameter()
] + NamedTPUCombination.parameter_modifiers(self)
class EagerGraphCombination(test_combinations.TestCombination):
"""Run the test in Graph or Eager mode. Graph is the default.
The optional `mode` parameter controls the test's execution mode. Its
accepted values are "graph" or "eager" literals.
"""
def context_managers(self, kwargs):
# TODO(isaprykin): Switch the default to eager.
mode = kwargs.pop("mode", "graph")
if mode == "eager":
return [context.eager_mode()]
elif mode == "graph":
return [ops.Graph().as_default(), context.graph_mode()]
else:
raise ValueError(
"'mode' has to be either 'eager' or 'graph' and not {}".format(mode))
def parameter_modifiers(self):
return [test_combinations.OptionalParameter("mode")]
class NamedDistribution(object):
"""Wraps a `tf.distribute.Strategy` and adds a name for test titles."""
def __init__(self, name, distribution_fn, required_gpus=None,
required_tpu=False):
object.__init__(self)
self._name = name
self._distribution_fn = distribution_fn
self._required_gpus = required_gpus
self._required_tpu = required_tpu
@property
def strategy(self):
return self._distribution_fn()
@property
def required_gpus(self):
return self._required_gpus
@property
def required_tpu(self):
return self._required_tpu
def __repr__(self):
return self._name
generate = functools.partial(
test_combinations.generate,
test_combinations=(EagerGraphCombination(), GPUCombination(),
TPUCombination()))
combine = test_combinations.combine
times = test_combinations.times
NamedObject = test_combinations.NamedObject
|
tensorflow-master
|
tensorflow/python/distribute/combinations.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""A configure tuple for high-level APIs for running distribution strategies."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
class DistributeConfig(
collections.namedtuple(
'DistributeConfig',
['train_distribute', 'eval_distribute', 'remote_cluster'])):
"""A config tuple for distribution strategies.
Attributes:
train_distribute: a `DistributionStrategy` object for training.
eval_distribute: an optional `DistributionStrategy` object for
evaluation.
remote_cluster: a dict, `ClusterDef` or `ClusterSpec` object specifying
the cluster configurations. If this is given, the `train_and_evaluate`
method will be running as a standalone client which connects to the
cluster for training.
"""
def __new__(cls,
train_distribute=None,
eval_distribute=None,
remote_cluster=None):
return super(DistributeConfig, cls).__new__(cls, train_distribute,
eval_distribute, remote_cluster)
|
tensorflow-master
|
tensorflow/python/distribute/distribute_config.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for custom training loops."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from absl.testing import parameterized
from tensorflow.python import tf2
from tensorflow.python.data.ops import dataset_ops
from tensorflow.python.distribute import combinations
from tensorflow.python.distribute import strategy_combinations
from tensorflow.python.distribute import tpu_strategy
from tensorflow.python.eager import backprop
from tensorflow.python.eager import def_function
from tensorflow.python.eager import test
from tensorflow.python.framework import constant_op
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import variables
class InputIterationTest(test.TestCase, parameterized.TestCase):
@combinations.generate(
combinations.combine(
distribution=strategy_combinations.strategies_minus_tpu,
mode=["eager"]
))
def testFullEager(self, distribution):
dataset = self._get_dataset()
def train_step(data):
return data
dist_dataset = distribution.experimental_distribute_dataset(dataset)
results = []
for x in dist_dataset:
output = distribution.experimental_local_results(
distribution.experimental_run_v2(train_step, args=(x,)))
results.append(output)
self._validate_outputs(results)
@combinations.generate(
combinations.combine(
distribution=strategy_combinations.strategies_minus_tpu,
mode=["eager"]
))
def testStepInFunction(self, distribution):
dataset = self._get_dataset()
@def_function.function
def train_step(data):
return data
dist_dataset = distribution.experimental_distribute_dataset(dataset)
results = []
for x in dist_dataset:
output = distribution.experimental_local_results(
distribution.experimental_run_v2(train_step, args=(x,)))
results.append(output)
self._validate_outputs(results)
@combinations.generate(
combinations.combine(
distribution=strategy_combinations.strategies_minus_tpu +
[strategy_combinations.tpu_strategy_one_step],
mode=["eager"]
))
def testRunInFunction(self, distribution):
dataset = self._get_dataset()
def train_step(data):
return data
@def_function.function
def f_train_step(input_data):
return distribution.experimental_local_results(
distribution.experimental_run_v2(train_step, args=(input_data,)))
dist_dataset = distribution.experimental_distribute_dataset(dataset)
results = []
for x in dist_dataset:
output = f_train_step(x)
results.append(output)
self._validate_outputs(results)
@combinations.generate(
combinations.combine(
distribution=strategy_combinations.strategies_minus_tpu +
[strategy_combinations.tpu_strategy_one_step],
mode=["eager"]
))
def testRunInFunctionAutoGraphApplication(self, distribution):
dataset = self._get_dataset()
def train_step(data):
if math_ops.reduce_sum(data) < 0:
return -data
return data
@def_function.function
def f_train_step(input_data):
return distribution.experimental_local_results(
distribution.experimental_run_v2(train_step, args=(input_data,)))
dist_dataset = distribution.experimental_distribute_dataset(dataset)
results = []
for x in dist_dataset:
output = f_train_step(x)
results.append(output)
self._validate_outputs(results)
@combinations.generate(
combinations.combine(
distribution=strategy_combinations.strategies_minus_tpu,
mode=["eager"]
))
def testDatasetIterationInFunction(self, distribution):
with distribution.scope():
a = variables.Variable(
1.0, aggregation=variables.VariableAggregation.ONLY_FIRST_REPLICA)
def train_step(_):
a.assign_add(1.0)
@def_function.function
def f_train_step(dist_dataset):
number_of_steps = constant_op.constant(0.0)
product_of_means = constant_op.constant(2.0)
for x in dist_dataset: # loop with values modified each iteration
number_of_steps += 1
product_of_means *= math_ops.cast(
distribution.reduce("MEAN", x, axis=0), product_of_means.dtype)
for y in dist_dataset: # loop with no intermediate state
distribution.experimental_run_v2(train_step, args=(y,))
return number_of_steps, product_of_means
dataset = self._get_dataset()
dist_dataset = distribution.experimental_distribute_dataset(dataset)
number_of_steps, product_of_means = f_train_step(dist_dataset)
self.assertEqual(5, number_of_steps.numpy())
# 2.0 * (0+1)/2 * (2+3)/2 * (4+5)/2 * (6+7)/2 * (8+9)/2
# = (5 * 9 * 13 * 17) / 16
self.assertNear((5 * 9 * 13 * 17) / 16, product_of_means.numpy(), 1e-3)
# We set the initial value of `a` to 1 and iterate through the dataset 5
# times(10/2 where 10 is the number of dataset elements and 2 is the batch
# size). Hence the final result is 6.
self.assertEqual(6.0, (a.numpy()))
@combinations.generate(
combinations.combine(
distribution=strategy_combinations.strategies_minus_tpu +
[strategy_combinations.tpu_strategy_one_step],
mode=["eager"]
))
def testIterationInsideFunction(self, distribution):
def step_fn(data):
return data
@def_function.function
def train(dataset):
results = []
iterator = iter(dataset)
# we iterate through the loop 5 times since we have 10 elements and a
# global batch of 2.
for _ in range(5):
elem = next(iterator)
output = distribution.experimental_local_results(
distribution.experimental_run_v2(step_fn, args=(elem,)))
results.append(output)
return results
dataset = self._get_dataset()
dist_dataset = distribution.experimental_distribute_dataset(dataset)
results = train(dist_dataset)
self._validate_outputs(results)
@combinations.generate(
combinations.combine(
distribution=strategy_combinations.strategies_minus_tpu +
[strategy_combinations.tpu_strategy_one_step],
mode=["eager"]
))
def testIterationOutsideFunction(self, distribution):
def train_step(data):
return data
@def_function.function
def f_train_step(input_data):
return distribution.experimental_local_results(
distribution.experimental_run_v2(train_step, args=(input_data,)))
dataset = self._get_dataset()
dist_dataset = distribution.experimental_distribute_dataset(dataset)
iterator = iter(dist_dataset)
results = []
# we iterate through the loop 5 times since we have 10 elements and a
# global batch of 2.
for _ in range(5):
output = f_train_step(next(iterator))
results.append(output)
self._validate_outputs(results)
def _get_dataset(self):
if tf2.enabled():
return dataset_ops.DatasetV2.range(10).batch(2)
else:
return dataset_ops.Dataset.range(10).batch(2)
def _validate_outputs(self, actual_results):
expected_results = [[i, i+1] for i in range(0, 10, 2)]
self.assertEqual(len(expected_results), len(actual_results))
for i, expected_result in enumerate(expected_results):
final_result = []
actual_result = actual_results[i]
for val in actual_result:
final_result.extend(val.numpy())
self.assertAllEqual(expected_result, final_result)
class GradientTapeTest(test.TestCase, parameterized.TestCase):
@combinations.generate(
combinations.combine(
distribution=strategy_combinations.strategies_minus_tpu +
[strategy_combinations.tpu_strategy_one_step],
mode=["eager"],
model_in_tf_function=[True, False]
))
def test1(self, distribution, model_in_tf_function):
# b/134975331
if model_in_tf_function and isinstance(
distribution, (tpu_strategy.TPUStrategy, tpu_strategy.TPUStrategyV1)):
self.skipTest("model inside tf.function doesn't work with TPUStrategy")
def model(x):
return x * x
if model_in_tf_function:
model = def_function.function(model)
with distribution.scope():
x = variables.Variable(1.0)
@def_function.function
def train_step():
def replica_step():
with backprop.GradientTape() as tape:
y = model(x)
return tape.gradient(y, x)
return distribution.experimental_run_v2(replica_step)
grads = distribution.experimental_local_results(train_step())
self.assertLen(grads, distribution.num_replicas_in_sync)
self.assertTrue(all(g is not None for g in grads))
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/custom_training_loop_test.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for saving and loading using keras save/load APIs with DS."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.distribute import combinations
from tensorflow.python.distribute import saved_model_test_base as test_base
from tensorflow.python.eager import test
from tensorflow.python.keras.saving import save
class KerasSaveLoadTest(test_base.TestSavedModelBase):
def setUp(self):
self._root_dir = 'keras_save_load'
super(KerasSaveLoadTest, self).setUp()
def _save_model(self, model, saved_dir):
model.save(saved_dir, save_format='tf')
def _load_and_run_model(self, distribution, saved_dir, predict_dataset,
output_name):
restored_keras_model = save.load_model(saved_dir)
return restored_keras_model.predict(
predict_dataset, steps=test_base.PREDICT_STEPS)
@combinations.generate(test_base.simple_models_with_strategies())
def test_save_no_strategy_restore_strategy(self, model_and_input,
distribution):
self.run_test_save_no_strategy_restore_strategy(model_and_input,
distribution)
@combinations.generate(
combinations.times(test_base.simple_models_with_strategies(),
combinations.combine(save_in_scope=[True, False])))
def test_save_strategy_restore_no_strategy(self, model_and_input,
distribution, save_in_scope):
if save_in_scope:
self.skipTest(('b/134703272 - Saving model in tf.distribute.Strategy ',
'scope is not supported.'))
self.run_test_save_strategy_restore_no_strategy(model_and_input,
distribution, save_in_scope)
@combinations.generate(
combinations.times(test_base.simple_models_with_strategy_pairs(),
combinations.combine(save_in_scope=[True, False])))
def test_save_strategy_restore_strategy(self, model_and_input,
distribution_for_saving,
distribution_for_restoring,
save_in_scope):
if save_in_scope:
self.skipTest(('b/134703272 - Saving model in tf.distribute.Strategy ',
'scope is not supported.'))
self.run_test_save_strategy_restore_strategy(model_and_input,
distribution_for_saving,
distribution_for_restoring,
save_in_scope)
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/keras_save_load_test.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""TPU Strategy."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import contextlib
import copy
import weakref
import numpy as np
from tensorflow.python.autograph.core import ag_ctx
from tensorflow.python.autograph.impl import api as autograph
from tensorflow.python.distribute import cross_device_ops as cross_device_ops_lib
from tensorflow.python.distribute import device_util
from tensorflow.python.distribute import distribute_lib
from tensorflow.python.distribute import input_lib
from tensorflow.python.distribute import numpy_dataset
from tensorflow.python.distribute import reduce_util
from tensorflow.python.distribute import values
from tensorflow.python.distribute.cluster_resolver import TPUClusterResolver
from tensorflow.python.eager import context
from tensorflow.python.eager import def_function
from tensorflow.python.eager import tape
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import tensor_shape
from tensorflow.python.framework import tensor_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import resource_variable_ops
from tensorflow.python.ops import variable_scope as vs
from tensorflow.python.tpu import device_assignment as device_assignment_lib
from tensorflow.python.tpu import tpu
from tensorflow.python.tpu import tpu_strategy_util
from tensorflow.python.tpu import tpu_system_metadata as tpu_system_metadata_lib
from tensorflow.python.tpu import training_loop
from tensorflow.python.tpu.ops import tpu_ops
from tensorflow.python.util import nest
from tensorflow.python.util.tf_export import tf_export
def get_tpu_system_metadata(tpu_cluster_resolver):
"""Retrieves TPU system metadata given a TPUClusterResolver."""
master = tpu_cluster_resolver.master()
# pylint: disable=protected-access
cluster_spec = tpu_cluster_resolver.cluster_spec()
cluster_def = cluster_spec.as_cluster_def() if cluster_spec else None
tpu_system_metadata = (
tpu_system_metadata_lib._query_tpu_system_metadata(
master,
cluster_def=cluster_def,
query_topology=False))
return tpu_system_metadata
@contextlib.contextmanager
def maybe_init_scope():
if ops.executing_eagerly_outside_functions():
yield
else:
with ops.init_scope():
yield
# TODO(jhseu): Deduplicate with MirroredStrategy?
def _create_tpu_mirrored_variable( # pylint: disable=missing-docstring
strategy, device_map, logical_device, real_mirrored_creator,
*args, **kwargs):
# Figure out what collections this variable should be added to.
# We'll add the TPUMirroredVariable to those collections instead.
var_collections = kwargs.pop("collections", None)
if var_collections is None:
var_collections = [ops.GraphKeys.GLOBAL_VARIABLES]
kwargs["collections"] = []
# TODO(jhseu): Should we have different behavior for different
# synchronization settings?
# Get aggregation value
# TODO(jhseu): Support aggregation in a replica context.
aggregation = kwargs.pop("aggregation", vs.VariableAggregation.NONE)
if aggregation not in [
vs.VariableAggregation.NONE,
vs.VariableAggregation.SUM,
vs.VariableAggregation.MEAN,
vs.VariableAggregation.ONLY_FIRST_REPLICA,
]:
raise ValueError("Invalid variable aggregation mode: {} for variable: {}"
.format(aggregation, kwargs["name"]))
# Ignore user-specified caching device, not needed for mirrored variables.
kwargs.pop("caching_device", None)
# TODO(josh11b,apassos): It would be better if variable initialization
# was never recorded on the tape instead of having to do this manually
# here.
with tape.stop_recording():
devices = device_map.logical_to_actual_devices(logical_device)
value_list = real_mirrored_creator(devices, *args, **kwargs)
result = values.TPUMirroredVariable(
strategy, device_map, value_list, aggregation,
logical_device=logical_device)
if not (context.executing_eagerly() or ops.inside_function()):
g = ops.get_default_graph()
# If "trainable" is True, next_creator() will add the member variables
# to the TRAINABLE_VARIABLES collection, so we manually remove
# them and replace with the MirroredVariable. We can't set
# "trainable" to False for next_creator() since that causes functions
# like implicit_gradients to skip those variables.
if kwargs.get("trainable", True):
var_collections.append(ops.GraphKeys.TRAINABLE_VARIABLES)
l = g.get_collection_ref(ops.GraphKeys.TRAINABLE_VARIABLES)
for v in value_list:
l.remove(v)
g.add_to_collections(var_collections, result)
return result
@tf_export("distribute.experimental.TPUStrategy", v1=[])
class TPUStrategy(distribute_lib.Strategy):
"""TPU distribution strategy implementation."""
def __init__(self,
tpu_cluster_resolver=None,
device_assignment=None):
"""Initializes the TPUStrategy object.
Args:
tpu_cluster_resolver: A tf.distribute.cluster_resolver.TPUClusterResolver,
which provides information about the TPU cluster.
device_assignment: Optional `tf.tpu.experimental.DeviceAssignment` to
specify the placement of replicas on the TPU cluster. Currently only
supports the usecase of using a single core within a TPU cluster.
"""
super(TPUStrategy, self).__init__(TPUExtended(
self, tpu_cluster_resolver, device_assignment=device_assignment))
# TODO(cjfj): Modify `_call_for_each_replica` in `TPUExtended` such that this
# can use the default implementation.
# This implementation runs a single step. It does not use infeed or outfeed.
def experimental_run_v2(self, fn, args=(), kwargs=None):
"""See base class."""
fn = autograph.tf_convert(fn, ag_ctx.control_status_ctx())
return self.extended.tpu_run(fn, args, kwargs)
@tf_export(v1=["distribute.experimental.TPUStrategy"])
class TPUStrategyV1(distribute_lib.StrategyV1):
"""TPU distribution strategy implementation."""
def __init__(self,
tpu_cluster_resolver=None,
steps_per_run=None,
device_assignment=None):
"""Initializes the TPUStrategy object.
Args:
tpu_cluster_resolver: A tf.distribute.cluster_resolver.TPUClusterResolver,
which provides information about the TPU cluster.
steps_per_run: Number of steps to run on device before returning to the
host. Note that this can have side-effects on performance, hooks,
metrics, summaries etc.
This parameter is only used when Distribution Strategy is used with
estimator or keras.
device_assignment: Optional `tf.tpu.experimental.DeviceAssignment` to
specify the placement of replicas on the TPU cluster. Currently only
supports the usecase of using a single core within a TPU cluster.
"""
super(TPUStrategyV1, self).__init__(TPUExtended(
self, tpu_cluster_resolver, steps_per_run, device_assignment))
@property
def steps_per_run(self):
"""DEPRECATED: use .extended.steps_per_run instead."""
return self._extended.steps_per_run
# TODO(cjfj): Modify `_call_for_each_replica` in `TPUExtended` such that this
# can use the default implementation.
# This implementation runs a single step. It does not use infeed or outfeed.
def experimental_run_v2(self, fn, args=(), kwargs=None):
"""See base class."""
fn = autograph.tf_convert(fn, ag_ctx.control_status_ctx())
return self.extended.tpu_run(fn, args, kwargs)
# TODO(josh11b): Switch to V2 when we no longer need to support tf.compat.v1.
class TPUExtended(distribute_lib.StrategyExtendedV1):
"""Implementation of TPUStrategy."""
def __init__(self,
container_strategy,
tpu_cluster_resolver=None,
steps_per_run=None,
device_assignment=None):
super(TPUExtended, self).__init__(container_strategy)
if tpu_cluster_resolver is None:
tpu_cluster_resolver = TPUClusterResolver("")
if steps_per_run is None:
# TODO(frankchn): Warn when we are being used by DS/Keras and this is
# not specified.
steps_per_run = 1
self._tpu_function_cache = weakref.WeakKeyDictionary()
self._tpu_cluster_resolver = tpu_cluster_resolver
self._tpu_metadata = get_tpu_system_metadata(self._tpu_cluster_resolver)
self._device_assignment = device_assignment
# Device assignment is currently only supported for 1 core case.
if self._device_assignment:
assert isinstance(self._device_assignment,
device_assignment_lib.DeviceAssignment)
if self._device_assignment.num_replicas != 1:
raise ValueError("Device assignment is only supported for a single "
"core single replica case currently.")
if self._device_assignment.num_cores_per_replica != 1:
raise ValueError("Device assignment is only supported for a single "
"core single replica case currently.")
if not all(self._device_assignment.core_assignment[0][0] == [0, 0, 0]):
raise ValueError("Device assignment is only supported for a single "
"core single replica case currently.")
# TODO(jhseu): Switch to DeviceAssignment to support pods and model
# parallelism.
self._tpu_devices = [d.name for d in self._tpu_metadata.devices
if "device:TPU:" in d.name]
self._host_device = device_util.get_host_for_device(self._tpu_devices[0])
# Only create variables for the number of replicas we're running.
self._tpu_devices = self._tpu_devices[:self._num_replicas_in_sync]
self._device_map = values.ReplicaDeviceMap(self._tpu_devices)
# Preload the data onto the TPUs.
input_worker_devices = collections.OrderedDict()
for tpu_device in self._tpu_devices:
host_device = device_util.get_host_for_device(tpu_device)
input_worker_devices.setdefault(host_device, [])
input_worker_devices[host_device].append(tpu_device)
self._input_workers = input_lib.InputWorkers(
self._device_map, tuple(input_worker_devices.items()))
# TODO(sourabhbajaj): Remove this once performance of running one step
# at a time is comparable to multiple steps.
self.steps_per_run = steps_per_run
self._require_static_shapes = True
self.experimental_enable_get_next_as_optional = True
def _validate_colocate_with_variable(self, colocate_with_variable):
values.validate_colocate_tpu_variable(colocate_with_variable, self)
def _make_dataset_iterator(self, dataset):
"""Make iterators for each of the TPU hosts."""
return input_lib.DatasetIterator(
dataset,
self._input_workers,
self._container_strategy(),
split_batch_by=self._num_replicas_in_sync)
def _make_input_fn_iterator(
self,
input_fn,
replication_mode=distribute_lib.InputReplicationMode.PER_WORKER):
input_contexts = []
num_workers = self._input_workers.num_workers
for i in range(num_workers):
input_contexts.append(distribute_lib.InputContext(
num_input_pipelines=num_workers,
input_pipeline_id=i,
num_replicas_in_sync=self._num_replicas_in_sync))
return input_lib.InputFunctionIterator(
input_fn,
self._input_workers,
input_contexts,
self._container_strategy())
def _experimental_make_numpy_dataset(self, numpy_input, session):
return numpy_dataset.one_host_numpy_dataset(
numpy_input, numpy_dataset.SingleDevice(self._host_device),
session)
def _experimental_distribute_dataset(self, dataset):
return input_lib.get_distributed_dataset(
dataset,
self._input_workers,
self._container_strategy(),
split_batch_by=self._num_replicas_in_sync)
def _experimental_distribute_datasets_from_function(self, dataset_fn):
input_contexts = []
num_workers = self._input_workers.num_workers
for i in range(num_workers):
input_contexts.append(distribute_lib.InputContext(
num_input_pipelines=num_workers,
input_pipeline_id=i,
num_replicas_in_sync=self._num_replicas_in_sync))
return input_lib.DistributedDatasetsFromFunction(
dataset_fn,
self._input_workers,
input_contexts,
self._container_strategy())
# TODO(priyag): Deal with OutOfRange errors once b/111349762 is fixed.
# TODO(sourabhbajaj): Remove the initial_loop_values parameter when we have
# a mechanism to infer the outputs of `fn`. Pending b/110550782.
def _experimental_run_steps_on_iterator(
self, fn, multi_worker_iterator, iterations, initial_loop_values=None):
# Wrap `fn` for repeat.
if initial_loop_values is None:
initial_loop_values = {}
initial_loop_values = nest.flatten(initial_loop_values)
ctx = input_lib.MultiStepContext()
def run_fn(inputs):
"""Single step on the TPU device."""
fn_result = fn(ctx, inputs)
flat_last_step_outputs = nest.flatten(ctx.last_step_outputs)
if flat_last_step_outputs:
with ops.control_dependencies([fn_result]):
return [array_ops.identity(f) for f in flat_last_step_outputs]
else:
return fn_result
# We capture the control_flow_context at this point, before we run `fn`
# inside a while_loop and TPU replicate context. This is useful in cases
# where we might need to exit these contexts and get back to the outer
# context to do some things, for e.g. create an op which should be
# evaluated only once at the end of the loop on the host. One such usage
# is in creating metrics' value op.
self._outer_control_flow_context = (
ops.get_default_graph()._get_control_flow_context()) # pylint: disable=protected-access
def rewrite_fn(*args):
"""The rewritten step fn running on TPU."""
del args
per_replica_inputs = multi_worker_iterator.get_next()
replicate_inputs = []
for replica_id in range(self._num_replicas_in_sync):
select_replica = lambda x: values.select_replica(replica_id, x) # pylint: disable=cell-var-from-loop
replicate_inputs.append((nest.map_structure(
select_replica, per_replica_inputs),))
replicate_outputs = tpu.replicate(
run_fn, replicate_inputs, device_assignment=self._device_assignment)
# If run_fn has tensor outputs, tpu.replicate returns a list of list. We
# will flatten it in this case. If run_fn has no tensor outputs,
# tpu.replicate returns a list of no_ops, we will keep the output as it
# is.
if isinstance(replicate_outputs[0], list):
replicate_outputs = nest.flatten(replicate_outputs)
return replicate_outputs
# TODO(sourabhbajaj): The input to while loop should be based on the
# output type of the step_fn
assert isinstance(initial_loop_values, list)
initial_loop_values = initial_loop_values * self._num_replicas_in_sync
# Put the while loop op on TPU host 0.
with ops.device(self._host_device):
if self.steps_per_run == 1:
replicate_outputs = rewrite_fn()
else:
replicate_outputs = training_loop.repeat(iterations, rewrite_fn,
initial_loop_values)
del self._outer_control_flow_context
ctx.run_op = control_flow_ops.group(replicate_outputs)
if isinstance(replicate_outputs, list):
# Filter out any ops from the outputs, typically this would be the case
# when there were no tensor outputs.
last_step_tensor_outputs = [
x for x in replicate_outputs if not isinstance(x, ops.Operation)
]
# Outputs are currently of the structure (flattened)
# [output0_device0, output1_device0, output2_device0,
# output0_device1, output1_device1, output2_device1,
# ...]
# Convert this to the following structure instead: (grouped by output)
# [[output0_device0, output0_device1],
# [output1_device0, output1_device1],
# [output2_device0, output2_device1]]
output_num = len(last_step_tensor_outputs) // self._num_replicas_in_sync
last_step_tensor_outputs = [
last_step_tensor_outputs[i::output_num] for i in range(output_num)
]
else:
# no tensors returned.
last_step_tensor_outputs = []
_set_last_step_outputs(ctx, last_step_tensor_outputs)
return ctx
def _call_for_each_replica(self, fn, args, kwargs):
# TODO(jhseu): Consider making it so call_for_each_replica implies that
# we're in a tpu.rewrite(), and update TPUMirroredVariable accordingly.
with _TPUReplicaContext(self._container_strategy()):
return fn(*args, **kwargs)
def _experimental_initialize_system(self):
"""Experimental method added to be used by Estimator.
This is a private method only to be used by Estimator. Other frameworks
should directly be calling `tf.tpu.experimental.initialize_tpu_system`
"""
tpu_strategy_util.initialize_tpu_system(self._tpu_cluster_resolver)
def _create_variable(self, next_creator, *args, **kwargs):
"""Create a TPUMirroredVariable. See `DistributionStrategy.scope`."""
colocate_with = kwargs.pop("colocate_with", None)
if colocate_with is None:
device_map = self._device_map
logical_device = 0 # TODO(josh11b): Get logical device from scope here.
elif isinstance(colocate_with, numpy_dataset.SingleDevice):
with ops.device(colocate_with.device):
return next_creator(*args, **kwargs)
else:
device_map = colocate_with.device_map
logical_device = colocate_with.logical_device
def _real_mirrored_creator(devices, *args, **kwargs): # pylint: disable=g-missing-docstring
initial_value = None
value_list = []
for i, d in enumerate(devices):
with ops.device(d):
if i == 0:
initial_value = kwargs["initial_value"]
# Note: some v1 code expects variable initializer creation to happen
# inside a init_scope.
with maybe_init_scope():
initial_value = initial_value() if callable(
initial_value) else initial_value
if i > 0:
# Give replicas meaningful distinct names:
var0name = value_list[0].name.split(":")[0]
# We append a / to variable names created on replicas with id > 0 to
# ensure that we ignore the name scope and instead use the given
# name as the absolute name of the variable.
kwargs["name"] = "%s/replica_%d/" % (var0name, i)
kwargs["initial_value"] = initial_value
with context.device_policy(context.DEVICE_PLACEMENT_SILENT):
v = next_creator(*args, **kwargs)
assert not isinstance(v, values.TPUMirroredVariable)
value_list.append(v)
return value_list
return _create_tpu_mirrored_variable(
self._container_strategy(), device_map, logical_device,
_real_mirrored_creator, *args, **kwargs)
def _reduce_to(self, reduce_op, value, destinations):
if values._enclosing_tpu_context() is not None: # pylint: disable=protected-access
if reduce_op == reduce_util.ReduceOp.MEAN:
# TODO(jhseu): Revisit once we support model-parallelism.
value *= (1. / self._num_replicas_in_sync)
elif reduce_op != reduce_util.ReduceOp.SUM:
raise NotImplementedError(
"Currently only support sum & mean in TPUStrategy.")
return tpu_ops.cross_replica_sum(value)
if not isinstance(value, values.DistributedValues):
# This function handles reducing values that are not PerReplica or
# Mirrored values. For example, the same value could be present on all
# replicas in which case `value` would be a single value or value could
# be 0.
return cross_device_ops_lib.reduce_non_distributed_value(
reduce_op, self._device_map, value, destinations)
# TODO(cjfj): Detect when it is possible to use `cross_replica_sum`.
# Always performs the reduction on the TPU host.
with ops.device(self._host_device):
output = math_ops.add_n(value.values)
if reduce_op == reduce_util.ReduceOp.MEAN:
output *= (1. / len(value.values))
devices = cross_device_ops_lib.get_devices_from(destinations)
if len(devices) == 1:
# If necessary, copy to requested destination.
dest_canonical = device_util.canonicalize(devices[0])
host_canonical = device_util.canonicalize(self._host_device)
if dest_canonical != host_canonical:
with ops.device(dest_canonical):
output = array_ops.identity(output)
else:
output = cross_device_ops_lib.simple_broadcast(output, destinations)
return output
def _update(self, var, fn, args, kwargs, group):
assert isinstance(var, values.TPUMirroredVariable) or isinstance(
var, resource_variable_ops.BaseResourceVariable)
if values._enclosing_tpu_context() is not None: # pylint: disable=protected-access
if group:
return fn(var, *args, **kwargs)
else:
return (fn(var, *args, **kwargs),)
# Otherwise, we revert to MirroredStrategy behavior and update each variable
# directly.
updates = []
for i, (d, v) in enumerate(zip(var.devices, var.values)):
name = "update_%d" % i
with ops.device(d), distribute_lib.UpdateContext(d), ops.name_scope(name):
# If args and kwargs are not mirrored, the value is returned as is.
updates.append(fn(v,
*values.select_device_mirrored(d, args),
**values.select_device_mirrored(d, kwargs)))
return values.update_regroup(self, self._device_map, updates, group)
def read_var(self, var):
assert isinstance(var, values.TPUMirroredVariable) or isinstance(
var, resource_variable_ops.BaseResourceVariable)
return var.read_value()
def _local_results(self, val):
if isinstance(val, values.DistributedValues):
# Return in a deterministic order.
return tuple(val.get(device=d) for d in sorted(val.devices))
elif isinstance(val, list):
# TODO(josh11b): We need to remove this case; per device values should
# be represented using a PerReplica wrapper instead of a list with
# one entry per device.
return tuple(val)
elif isinstance(val, values.TPUMirroredVariable):
# pylint: disable=protected-access
if values._enclosing_tpu_context() is not None:
return (val,)
return val.values
return (val,)
def value_container(self, value):
return value
def _broadcast_to(self, tensor, destinations):
del destinations
return tensor
@property
def num_hosts(self):
if self._device_assignment is None:
return self._tpu_metadata.num_hosts
return len(set([self._device_assignment.host_device(r)
for r in range(self._device_assignment.num_replicas)]))
@property
def num_replicas_per_host(self):
if self._device_assignment is None:
return self._tpu_metadata.num_of_cores_per_host
# TODO(sourabhbajaj): Remove this method we use inputs and remove infeed
# as the computation of num_replicas_per_host is not a constant
# when using device_assignment. This is a temporary workaround to support
# StatefulRNN as everything is 1 in that case.
# This method needs to take host_id as input for correct computation.
max_models_per_host = (self._tpu_metadata.num_of_cores_per_host //
self._device_assignment.num_cores_per_replica)
models_per_host = min(self._device_assignment.num_replicas,
max_models_per_host)
return models_per_host * self._device_assignment.num_cores_per_replica
@property
def _num_replicas_in_sync(self):
if self._device_assignment is None:
return self._tpu_metadata.num_cores
return (self._device_assignment.num_replicas *
self._device_assignment.num_cores_per_replica)
@property
def experimental_between_graph(self):
return False
@property
def experimental_should_init(self):
return True
@property
def should_checkpoint(self):
return True
@property
def should_save_summary(self):
return True
@property
def worker_devices(self):
return self._tpu_devices
@property
def parameter_devices(self):
return self._tpu_devices
def non_slot_devices(self, var_list):
return self._host_device
def _update_non_slot(self, colocate_with, fn, args, kwargs, group):
del colocate_with
with ops.device(self._host_device), distribute_lib.UpdateContext(
self._host_device):
result = fn(*args, **kwargs)
if group:
return result
else:
return nest.map_structure(self._local_results, result)
def _configure(self,
session_config=None,
cluster_spec=None,
task_type=None,
task_id=None):
del cluster_spec, task_type, task_id
if session_config:
session_config.CopyFrom(self._update_config_proto(session_config))
def _update_config_proto(self, config_proto):
updated_config = copy.deepcopy(config_proto)
updated_config.isolate_session_state = True
cluster_spec = self._tpu_cluster_resolver.cluster_spec()
if cluster_spec:
updated_config.cluster_def.CopyFrom(cluster_spec.as_cluster_def())
return updated_config
# TODO(priyag): Delete this once all strategies use global batch size.
@property
def _global_batch_size(self):
"""`make_dataset_iterator` and `make_numpy_iterator` use global batch size.
`make_input_fn_iterator` assumes per-replica batching.
Returns:
Boolean.
"""
return True
def tpu_run(self, fn, args, kwargs):
func = self._tpu_function_creator(fn)
return func(args, kwargs)
def _tpu_function_creator(self, fn):
if fn in self._tpu_function_cache:
return self._tpu_function_cache[fn]
strategy = self._container_strategy()
def tpu_function(args, kwargs):
"""TF Function used to replicate the user computation."""
if kwargs is None:
kwargs = {}
# Remove None at the end of args as they are not replicatable
# If there are None in the middle we can't do anything about it
# so let those cases fail.
# For example when Keras model predict is used they pass the targets as
# None. We want to handle it here so all client libraries don't have to
# do this as other strategies can handle None values better.
while args and args[-1] is None:
args = args[:-1]
# Used to re-structure flattened output tensors from `tpu.replicate()`
# into a structured format.
result = [[]]
def replicated_fn(replica_id, replica_args, replica_kwargs):
"""Wraps user function to provide replica ID and `Tensor` inputs."""
with _TPUReplicaContext(strategy, replica_id_in_sync_group=replica_id):
result[0] = fn(*replica_args, **replica_kwargs)
return result[0]
replicate_inputs = [] # By replica.
for i in range(strategy.num_replicas_in_sync):
replicate_inputs.append(
[constant_op.constant(i, dtype=dtypes.int32),
values.select_replica(i, args),
values.select_replica(i, kwargs)])
# Construct and pass `maximum_shapes` so that we could support dynamic
# shapes using dynamic padder.
if replicate_inputs:
maximum_shapes = []
flattened_list = nest.flatten(replicate_inputs[0])
for input_tensor in flattened_list:
if tensor_util.is_tensor(input_tensor):
maximum_shape = input_tensor.get_shape()
else:
maximum_shape = tensor_shape.TensorShape(np.shape(input_tensor))
maximum_shapes.append(maximum_shape)
maximum_shapes = nest.pack_sequence_as(replicate_inputs[0],
maximum_shapes)
else:
maximum_shapes = None
with strategy.scope():
replicate_outputs = tpu.replicate(
replicated_fn,
replicate_inputs,
device_assignment=self._device_assignment,
maximum_shapes=maximum_shapes)
# Remove all no ops that may have been added during 'tpu.replicate()'
if isinstance(result[0], list):
result[0] = [
output for output in result[0] if tensor_util.is_tensor(output)
]
# Workaround for `tpu.replicate` behaviour when single `Tensor` returned.
if result[0] is None:
replicate_outputs = [None] * len(replicate_outputs)
else:
replicate_outputs = [
nest.pack_sequence_as(result[0], nest.flatten(replica_output))
for replica_output in replicate_outputs
]
device_map = self._device_map # pylint: disable=protected-access
return values.regroup(device_map, replicate_outputs)
if context.executing_eagerly():
tpu_function = def_function.function(tpu_function)
self._tpu_function_cache[fn] = tpu_function
return tpu_function
class _TPUReplicaContext(distribute_lib.ReplicaContext):
"""Replication Context class for TPU Strategy."""
# TODO(sourabhbajaj): Call for each replica should be updating this.
# TODO(b/118385803): Always properly initialize replica_id.
def __init__(self, strategy, replica_id_in_sync_group=None):
if replica_id_in_sync_group is None:
replica_id_in_sync_group = constant_op.constant(0, dtypes.int32)
distribute_lib.ReplicaContext.__init__(
self, strategy, replica_id_in_sync_group=replica_id_in_sync_group)
@property
def devices(self):
distribute_lib.require_replica_context(self)
ds = self._strategy
replica_id = tensor_util.constant_value(self._replica_id_in_sync_group)
if replica_id is None: # Non-constant `Tensor` inside `tpu.replicate`.
# TODO(cjfj): Return other devices when model parallelism is supported.
return (tpu.core(0),)
else:
return (ds.extended.worker_devices[replica_id],)
def _set_last_step_outputs(ctx, last_step_tensor_outputs):
"""Sets the last step outputs on the given context."""
# Convert replicate_outputs to the original dict structure of
# last_step_outputs.
last_step_tensor_outputs_dict = nest.pack_sequence_as(
ctx.last_step_outputs, last_step_tensor_outputs)
for name, reduce_op in ctx._last_step_outputs_reduce_ops.items(): # pylint: disable=protected-access
output = last_step_tensor_outputs_dict[name]
# For outputs that have already been reduced, take the first value
# from the list as each value should be the same. Else return the full
# list of values.
# TODO(josh11b): If reduce_op is NONE, we should return a PerReplica
# value.
if reduce_op is not None:
# TODO(priyag): Should this return the element or a list with 1 element
last_step_tensor_outputs_dict[name] = output[0]
ctx._set_last_step_outputs(last_step_tensor_outputs_dict) # pylint: disable=protected-access
|
tensorflow-master
|
tensorflow/python/distribute/tpu_strategy.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for CrossDeviceOps."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import itertools
from absl.testing import parameterized
import numpy as np
from tensorflow.core.protobuf import config_pb2
from tensorflow.python.distribute import collective_all_reduce_strategy
from tensorflow.python.distribute import combinations
from tensorflow.python.distribute import cross_device_ops as cross_device_ops_lib
from tensorflow.python.distribute import cross_device_utils
from tensorflow.python.distribute import device_util
from tensorflow.python.distribute import multi_worker_test_base
from tensorflow.python.distribute import reduce_util
from tensorflow.python.distribute import values as value_lib
from tensorflow.python.eager import context
from tensorflow.python.framework import kernels
from tensorflow.python.eager import test
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import math_ops
def _get_devices(devices):
if isinstance(devices, (tuple, list)):
return tuple(device_util.resolve(d) for d in devices)
elif isinstance(devices, value_lib.DistributedValues):
return devices.devices
return (device_util.resolve(devices),)
def _make_per_replica(values, devices, regroup=False):
devices = _get_devices(devices)
assert len(values) == len(devices)
# We simulate the result of regroup called on PerReplica which strips the
# PerReplica wrapper if it has only one value.
if len(values) == 1 and regroup:
with ops.device(devices[0]):
placed_v = array_ops.identity(values[0])
return placed_v
index = []
for d, v in zip(devices, values):
with ops.device(d):
placed_v = array_ops.identity(v)
index.append(placed_v)
return value_lib.PerReplica(value_lib.ReplicaDeviceMap(devices), index)
# pylint: disable=g-doc-args,g-doc-return-or-yield
def _fake_mirrored(value, devices):
"""Create a faked Mirrored object for testing.
All components of the returned Mirrored have the same objects, which is not
true in reality.
"""
devices = _get_devices(devices)
return value_lib.Mirrored(value_lib.ReplicaDeviceMap(devices),
[value] * len(devices))
def _make_indexed_slices(values, indices, dense_shape, device):
with ops.device(device):
tensor = ops.IndexedSlices(
values=constant_op.constant(values),
indices=constant_op.constant(indices),
dense_shape=constant_op.constant(dense_shape))
return tensor
def _make_mirrored_indexed_slices(devices, values, indices, dense_shape):
values = [_make_indexed_slices(values, indices, dense_shape, d)
for d in devices]
return value_lib.Mirrored(value_lib.ReplicaDeviceMap(devices), values)
_cpu_device = "/device:CPU:0"
class CrossDeviceOpsTestBase(test.TestCase, parameterized.TestCase):
def _assert_indexed_slices_equal(self, left, right):
self.assertIsInstance(left, ops.IndexedSlices)
self.assertIsInstance(right, ops.IndexedSlices)
self.assertEqual(device_util.resolve(left.device),
device_util.resolve(right.device))
self.assertAllEqual(
self.evaluate(ops.convert_to_tensor(left)),
self.evaluate(ops.convert_to_tensor(right)))
def _assert_values_equal(self, left, right):
if isinstance(left, list):
for l, r in zip(left, right):
self._assert_values_equal(l, r)
else:
self.assertEqual(type(left), type(right))
self.assertEqual(set(left.devices), set(right.devices))
if isinstance(left.values[0], ops.IndexedSlices):
for d in left.devices:
self._assert_indexed_slices_equal(left.get(d), right.get(d))
elif context.executing_eagerly():
self.assertEqual([v.numpy() for v in left.values],
list(right.values))
else:
with self.cached_session() as sess:
self.assertEqual(
sess.run(list(left.values)), list(right.values))
def _testReductionAndBroadcast(self, cross_device_ops, devices):
if context.num_gpus() < sum(1 for d in devices if "GPU" in d.upper()):
self.skipTest("Not enough GPUs")
values = [constant_op.constant(float(d)) for d in range(len(devices))]
per_replica = _make_per_replica(values, devices)
mean = (len(devices) - 1.) / 2.
values_2 = [constant_op.constant(d + 1.0) for d in range(len(devices))]
per_replica_2 = _make_per_replica(values_2, devices)
mean_2 = mean + 1.
destination_mirrored = _fake_mirrored(1., devices)
destination_different = _fake_mirrored(1., _cpu_device)
destination_str = _cpu_device
all_destinations = [
destination_mirrored, destination_different, destination_str,
]
# test reduce()
for destinations in all_destinations:
self._assert_values_equal(
cross_device_ops.reduce(
reduce_util.ReduceOp.MEAN,
per_replica,
destinations=destinations),
_fake_mirrored(mean, destinations))
self._assert_values_equal(
cross_device_ops.reduce(
reduce_util.ReduceOp.MEAN,
per_replica_2,
destinations=destinations),
_fake_mirrored(mean_2, destinations))
self._assert_values_equal(
cross_device_ops.reduce(
reduce_util.ReduceOp.SUM, per_replica,
destinations=destinations),
_fake_mirrored(mean * len(devices), destinations))
self._assert_values_equal(
cross_device_ops.reduce(
reduce_util.ReduceOp.SUM,
per_replica_2,
destinations=destinations),
_fake_mirrored(mean_2 * len(devices), destinations))
# test batch_reduce()
for d1, d2 in itertools.product(all_destinations, all_destinations):
self._assert_values_equal(
cross_device_ops.batch_reduce(
reduce_util.ReduceOp.MEAN,
[(per_replica, d1), (per_replica_2, d2)]),
[
_fake_mirrored(mean, d1),
_fake_mirrored(mean_2, d2)
])
self._assert_values_equal(
cross_device_ops.batch_reduce(
reduce_util.ReduceOp.SUM,
[(per_replica, d1), (per_replica_2, d2)]),
[
_fake_mirrored(mean * len(devices), d1),
_fake_mirrored(mean_2 * len(devices), d2)
])
# test broadcast()
for destinations in all_destinations:
self._assert_values_equal(
cross_device_ops.broadcast(constant_op.constant(1.), destinations),
_fake_mirrored(1., destinations))
def _testIndexedSlicesAllReduce(self, devices, cross_device_ops_instance,
reduce_op, batch_reduce):
dense_shape = [5, 2]
t0 = _make_indexed_slices([[1., 2.]], [1], dense_shape, devices[0])
t1 = _make_indexed_slices([[3., 4.], [5., 6.]], [1, 3], dense_shape,
devices[1])
per_replica = value_lib.PerReplica(
value_lib.ReplicaDeviceMap(devices), (t0, t1))
if batch_reduce:
result = cross_device_ops_instance.batch_reduce(
reduce_op, [(per_replica, per_replica)])
else:
result = cross_device_ops_instance.reduce(reduce_op, per_replica,
per_replica)
total_indices_with_dups = [1, 1, 3]
total_indices_without_dups = [1, 3]
if reduce_op == reduce_util.ReduceOp.SUM:
total_values_with_dups = [[1., 2.], [3., 4.], [5., 6.]]
total_values_without_dups = [[4., 6.], [5., 6.]]
else:
assert reduce_op == reduce_util.ReduceOp.MEAN
total_values_with_dups = [[0.5, 1.], [1.5, 2.], [2.5, 3.]]
total_values_without_dups = [[2., 3.], [2.5, 3.]]
total_mirrored_with_dups = _make_mirrored_indexed_slices(
devices, total_values_with_dups, total_indices_with_dups, dense_shape)
total_mirrored_without_dups = _make_mirrored_indexed_slices(
devices, total_values_without_dups, total_indices_without_dups,
dense_shape)
# Test that the result is semantically equal to both the concatenated
# IndexedSlices, as well as when the duplicate indices are summed up.
if batch_reduce:
total_mirrored_with_dups = [total_mirrored_with_dups]
total_mirrored_without_dups = [total_mirrored_without_dups]
self._assert_values_equal(total_mirrored_with_dups, result)
self._assert_values_equal(total_mirrored_without_dups, result)
class SingleWorkerCrossDeviceOpsTest(CrossDeviceOpsTestBase):
reduction_to_one_combinations = combinations.combine(
cross_device_ops=[
combinations.NamedObject("DefaultReductionToOneDevice",
cross_device_ops_lib.ReductionToOneDevice()),
combinations.NamedObject(
"ReductionToCPUDeviceCrossDeviceOps",
cross_device_ops_lib.ReductionToOneDevice(
reduce_to_device=_cpu_device)),
combinations.NamedObject(
"AccumulateNCrossDeviceOp",
cross_device_ops_lib.ReductionToOneDevice(
accumulation_fn=math_ops.accumulate_n)),
],
devices=[
["/cpu:0"],
["/cpu:0", "/gpu:0"],
["/gpu:0", "/gpu:1"],
],
mode=["graph", "eager"])
allreduce_combinations = combinations.combine(
cross_device_ops=[
combinations.NamedObject(
"AllReduce",
cross_device_ops_lib.AllReduceCrossDeviceOps("nccl", 1, 0, 0)),
combinations.NamedObject(
"AllReduceNoGradientRepacking",
cross_device_ops_lib.AllReduceCrossDeviceOps("nccl", 0, 0, 0)),
combinations.NamedObject("NcclAllReduce",
cross_device_ops_lib.NcclAllReduce()),
combinations.NamedObject(
"HierarchicalCopy",
cross_device_ops_lib.HierarchicalCopyAllReduce(8)),
combinations.NamedObject(
"HierarchicalCopyAggregateSmallTensors",
cross_device_ops_lib.AllReduceCrossDeviceOps(
"hierarchical_copy", 0, 100, 10))
],
devices=[
["/gpu:0", "/gpu:1"],
],
mode=["graph", "eager"])
@combinations.generate(reduction_to_one_combinations + allreduce_combinations)
def testReductionAndBroadcast(self, cross_device_ops, devices):
self._testReductionAndBroadcast(cross_device_ops, devices)
def testChooseAlgorithm(self):
# Not use nccl if there is any cpu device.
self.assertIsInstance(
cross_device_ops_lib.choose_the_best(["/cpu:0"]),
cross_device_ops_lib.ReductionToOneDevice)
# Not use nccl if requested device is not visible to TensorFlow.
self.assertIsInstance(
cross_device_ops_lib.choose_the_best(["/gpu:100"]),
cross_device_ops_lib.ReductionToOneDevice)
if context.num_gpus() < 1:
return
devices = ["/gpu:0"]
def mock_get_registered_kernels_for_op(op):
if op == "NcclAllReduce":
return [object]
else:
return []
# Use nccl if nccl kernel is found.
with test.mock.patch.object(kernels, "get_registered_kernels_for_op",
mock_get_registered_kernels_for_op):
self.assertIsInstance(
cross_device_ops_lib.choose_the_best(devices),
cross_device_ops_lib.NcclAllReduce)
# Not use nccl if nccl kernel is not found.
with test.mock.patch.object(kernels,
"get_registered_kernels_for_op", lambda _: []):
self.assertIsInstance(
cross_device_ops_lib.choose_the_best(devices),
cross_device_ops_lib.ReductionToOneDevice)
@combinations.generate(combinations.combine(
mode=["graph", "eager"],
required_gpus=1))
def testSimpleReduceWithIndexedSlices(self):
devices = ["/cpu:0", "/gpu:0"]
t0 = _make_indexed_slices([[1., 2.]], [1], [5, 2], devices[0])
t1 = _make_indexed_slices([[3., 4.], [5., 6.]], [1, 3], [5, 2], devices[1])
per_replica = value_lib.PerReplica(
value_lib.ReplicaDeviceMap(devices), (t0, t1))
result = cross_device_ops_lib._simple_reduce(
per_replica, devices[0], math_ops.add_n, reduce_util.ReduceOp.SUM)
# Test that the result is semantically equal to both the concatenated
# IndexedSlices with and without duplicate indices.
total_with_dups = _make_indexed_slices(
[[1., 2.], [3., 4.], [5., 6.]], [1, 1, 3], [5, 2], devices[0])
total_without_dups = _make_indexed_slices(
[[4., 6.], [5., 6.]], [1, 3], [5, 2], devices[0])
self._assert_indexed_slices_equal(total_with_dups, result)
self._assert_indexed_slices_equal(total_without_dups, result)
@combinations.generate(
combinations.combine(
cross_device_ops_instance=[
combinations.NamedObject(
"ReductionToOneDevice",
cross_device_ops_lib.ReductionToOneDevice()),
combinations.NamedObject(
"AllReduceCrossDeviceOps",
cross_device_ops_lib.AllReduceCrossDeviceOps())
],
reduce_op=[reduce_util.ReduceOp.SUM, reduce_util.ReduceOp.MEAN],
batch_reduce=[True, False],
mode=["graph", "eager"],
required_gpus=1))
def testIndexedSlicesAllReduce(self, cross_device_ops_instance, reduce_op,
batch_reduce):
devices = ["/cpu:0", "/gpu:0"]
self._testIndexedSlicesAllReduce(devices, cross_device_ops_instance,
reduce_op, batch_reduce)
class MultiWorkerCrossDeviceOpsTest(multi_worker_test_base.MultiWorkerTestBase,
CrossDeviceOpsTestBase):
worker_devices = [
"/job:worker/replica:0/task:0", "/job:worker/replica:0/task:1"
]
multi_worker_allreduce_combinations = combinations.combine(
cross_device_ops=[
combinations.NamedObject(
"MultiWorkerAllReduce",
cross_device_ops_lib.MultiWorkerAllReduce(worker_devices, 2,
("pscpu/pscpu", 2, -1),
0, 0, 0)),
combinations.NamedObject(
"MultiWorkerAllReducePack",
cross_device_ops_lib.MultiWorkerAllReduce(worker_devices, 2,
("pscpu/pscpu", 2, -1),
1, 0, 0)),
combinations.NamedObject(
"MultiWorkerAllReduceAggregation",
cross_device_ops_lib.MultiWorkerAllReduce(worker_devices, 2,
("pscpu/pscpu", 2, -1),
0, 100, 10)),
combinations.NamedObject(
"MultiWorkerAllReduceMultipleSpecs",
cross_device_ops_lib.MultiWorkerAllReduce(
worker_devices, 2, [("pscpu/pscpu", 2, 100),
("xring", 2, -1)], 0, 0, 0)),
],
devices=[
[
"/job:worker/replica:0/task:0/device:CPU:0",
"/job:worker/replica:0/task:1/device:CPU:0"
],
[
"/job:worker/replica:0/task:0/device:GPU:0",
"/job:worker/replica:0/task:1/device:GPU:0"
],
[
"/job:worker/replica:0/task:0/device:GPU:0",
"/job:worker/replica:0/task:0/device:GPU:1",
"/job:worker/replica:0/task:1/device:GPU:0",
"/job:worker/replica:0/task:1/device:GPU:1"
],
],
mode=["graph"])
@combinations.generate(multi_worker_allreduce_combinations)
def testReductionAndBroadcast(self, cross_device_ops, devices):
self._testReductionAndBroadcast(cross_device_ops, devices)
NUM_WORKERS = 3
class CollectiveAllReduceTest(multi_worker_test_base.MultiWorkerTestBase,
parameterized.TestCase):
collective_key_base = 100000
@classmethod
def setUpClass(cls):
"""Create a local cluster with 3 workers."""
cls._cluster_spec = multi_worker_test_base.create_in_process_cluster(
num_workers=NUM_WORKERS, num_ps=0)
def setUp(self):
super(CollectiveAllReduceTest, self).setUp()
# Reusing keys is not supported well. So we have to give a different
# collective key base for different tests.
CollectiveAllReduceTest.collective_key_base += 100000
def _get_test_objects(self,
task_type,
task_id,
num_gpus=0,
use_strategy_object=False,
local_mode=False):
collective_keys = cross_device_utils.CollectiveKeys(
group_key_start=10 + CollectiveAllReduceTest.collective_key_base,
op_instance_key_start=100 + CollectiveAllReduceTest.collective_key_base,
variable_instance_key_start=10000 +
CollectiveAllReduceTest.collective_key_base)
if local_mode:
if num_gpus:
devices = ["/device:GPU:%d" % i for i in range(num_gpus)]
else:
devices = ["/device:CPU:0"]
if use_strategy_object:
strategy = collective_all_reduce_strategy.CollectiveAllReduceStrategy()
strategy.extended._collective_keys = collective_keys
strategy.extended._cross_device_ops._collective_keys = collective_keys
return strategy, devices, ""
else:
collective_all_reduce_ops = cross_device_ops_lib.CollectiveAllReduce(
1, num_gpus, collective_keys=collective_keys)
return collective_all_reduce_ops, devices, ""
else:
if num_gpus:
devices = [
"/job:%s/task:%d/replica:0/device:GPU:%d" % (task_type, task_id, i)
for i in range(num_gpus)
]
else:
devices = [
"/job:%s/task:%d/replica:0/device:CPU:0" % (task_type, task_id)
]
if use_strategy_object:
strategy = collective_all_reduce_strategy.CollectiveAllReduceStrategy()
strategy.configure(
cluster_spec=self._cluster_spec,
task_type=task_type,
task_id=task_id)
strategy.extended._collective_keys = collective_keys
strategy.extended._cross_device_ops._collective_keys = collective_keys
return (strategy, devices,
"grpc://" + self._cluster_spec[task_type][task_id])
else:
collective_all_reduce_ops = cross_device_ops_lib.CollectiveAllReduce(
NUM_WORKERS, num_gpus, collective_keys=collective_keys)
return (collective_all_reduce_ops, devices,
"grpc://" + self._cluster_spec[task_type][task_id])
def _assert_values_equal(self, left, right, sess):
if isinstance(left, list):
for l, r in zip(left, right):
self._assert_values_equal(l, r, sess)
else:
self.assertEqual(type(left), type(right))
self.assertEqual(set(left.devices), set(right.devices))
run_options = config_pb2.RunOptions()
run_options.experimental.collective_graph_key = 6
left_values = np.array(
sess.run(list(left.values), options=run_options)).flatten()
right_values = np.array(list(right.values)).flatten()
self.assertEqual(len(left_values), len(right_values))
for l, r in zip(left_values, right_values):
self.assertEqual(l, r)
def _test_reduction(self,
task_type,
task_id,
num_gpus,
use_strategy_object=False,
local_mode=False):
collective_all_reduce, devices, master_target = self._get_test_objects(
task_type,
task_id,
num_gpus,
use_strategy_object=use_strategy_object,
local_mode=local_mode)
if local_mode:
num_workers = 1
worker_device = None
else:
num_workers = len(self._cluster_spec.get("chief", [])) + len(
self._cluster_spec.get("worker", []))
worker_device = "/job:%s/task:%d" % (task_type, task_id)
def _reduce(test_object, reduce_op, per_replica, destinations):
if use_strategy_object:
with test_object.scope():
# Mimic the behavior that distribution strategy usually strips the
# wrapper if there is only one value.
if len(per_replica.values) == 1:
per_replica = per_replica.values[0]
return test_object.extended.reduce_to(reduce_op, per_replica,
destinations)
else:
return test_object.reduce(reduce_op, per_replica, destinations)
def _batch_reduce(test_object, reduce_op, value_destination_pairs):
if use_strategy_object:
with test_object.scope():
return test_object.extended.batch_reduce_to(reduce_op,
value_destination_pairs)
else:
return test_object.batch_reduce(reduce_op, value_destination_pairs)
with ops.Graph().as_default(), \
ops.device(worker_device), \
self.cached_session(target=master_target) as sess:
# Collective ops doesn't support scalar tensors, so we have to construct
# 1-d tensors.
values = [constant_op.constant([float(d)]) for d in range(len(devices))]
per_replica = _make_per_replica(values, devices)
mean = np.array([(len(devices) - 1.) / 2.])
values_2 = [constant_op.constant([d + 1.0]) for d in range(len(devices))]
per_replica_2 = _make_per_replica(values_2, devices)
mean_2 = np.array([mean[0] + 1.])
destination_mirrored = _fake_mirrored(1., devices)
destination_different = _fake_mirrored(1., _cpu_device)
destination_str = _cpu_device
all_destinations = [
destination_different, destination_mirrored, destination_str
]
# test reduce()
for destinations in all_destinations:
self._assert_values_equal(
_reduce(
collective_all_reduce,
reduce_util.ReduceOp.MEAN,
per_replica,
destinations=destinations), _fake_mirrored(mean, destinations),
sess)
self._assert_values_equal(
_reduce(
collective_all_reduce,
reduce_util.ReduceOp.MEAN,
per_replica_2,
destinations=destinations), _fake_mirrored(
mean_2, destinations), sess)
self._assert_values_equal(
_reduce(
collective_all_reduce,
reduce_util.ReduceOp.SUM,
per_replica,
destinations=destinations),
_fake_mirrored(mean * len(devices) * num_workers, destinations),
sess)
self._assert_values_equal(
_reduce(
collective_all_reduce,
reduce_util.ReduceOp.SUM,
per_replica_2,
destinations=destinations),
_fake_mirrored(mean_2 * len(devices) * num_workers, destinations),
sess)
# test batch_reduce()
for d1, d2 in itertools.product(all_destinations, all_destinations):
self._assert_values_equal(
_batch_reduce(collective_all_reduce, reduce_util.ReduceOp.MEAN,
[(per_replica, d1), (per_replica_2, d2)]),
[_fake_mirrored(mean, d1),
_fake_mirrored(mean_2, d2)], sess)
self._assert_values_equal(
_batch_reduce(collective_all_reduce, reduce_util.ReduceOp.SUM,
[(per_replica, d1), (per_replica_2, d2)]),
[
_fake_mirrored(mean * len(devices) * num_workers, d1),
_fake_mirrored(mean_2 * len(devices) * num_workers, d2)
], sess)
return True
def _get_indexed_slices(self, devices, start_i, as_per_replica=True):
dense_shape = [10, 2]
values = ([[1., 2.]], [[3., 4.]], [[2., 1.]], [[0., 0.]], [[3., 1.]],
[[2., 1.]])
indices = ([1], [2], [3], [4], [5], [6])
indexed_slices = []
for i, d in enumerate(devices):
idx = i + start_i
indexed_slices.append(
_make_indexed_slices(values[idx], indices[idx], dense_shape, d))
if as_per_replica:
per_replica = value_lib.PerReplica(
value_lib.ReplicaDeviceMap(devices), indexed_slices)
return per_replica
else:
return indexed_slices
def _test_reduce_indexed_slices(self,
task_type,
task_id,
num_gpus,
batch_reduce,
local_mode=False):
collective_all_reduce, devices, master_target = self._get_test_objects(
task_type, task_id, num_gpus, local_mode=local_mode)
if local_mode:
num_workers = 1
worker_device = None
else:
num_workers = len(self._cluster_spec.get("chief", [])) + len(
self._cluster_spec.get("worker", []))
worker_device = "/job:%s/task:%d" % (task_type, task_id)
with ops.Graph().as_default(), \
ops.device(worker_device), \
self.cached_session(target=master_target) as sess:
per_replica = self._get_indexed_slices(devices,
(task_id or 0) * max(num_gpus, 1))
if batch_reduce:
result = collective_all_reduce.batch_reduce(
reduce_util.ReduceOp.SUM, [(per_replica, per_replica)])[0]
else:
result = collective_all_reduce.reduce(reduce_util.ReduceOp.SUM,
per_replica, per_replica)
self.assertIsInstance(result, value_lib.Mirrored)
run_options = config_pb2.RunOptions()
run_options.experimental.collective_graph_key = 7
result = sess.run([ops.convert_to_tensor(v) for v in result.values],
options=run_options)[0]
# Reduce the same indexed slices on CPU locally as our expected results.
devices_cpu = [(worker_device or "") + "/device:CPU:0"] * (
max(num_gpus, 1) * num_workers)
per_replica_on_cpu = self._get_indexed_slices(
devices_cpu, 0, as_per_replica=False)
expected_result = cross_device_utils.aggregate_tensors_or_indexed_slices(
per_replica_on_cpu)
expected_result = sess.run(ops.convert_to_tensor(expected_result))
self.assertAllEqual(expected_result, result)
return True
@combinations.generate(
combinations.combine(
mode=["graph"],
num_gpus=[0, 1, 2],
required_gpus=1,
use_strategy_object=[True, False]))
def testReductionDistributed(self, num_gpus, use_strategy_object):
if context.num_gpus() < num_gpus:
return
self._run_between_graph_clients(
self._test_reduction,
self._cluster_spec,
num_gpus,
use_strategy_object=use_strategy_object)
@combinations.generate(
combinations.combine(
mode=["graph"],
num_gpus=[0, 1, 2],
required_gpus=1,
batch_reduce=[True]))
def testReduceIndexedSlicesDistributed(self, num_gpus, batch_reduce):
if context.num_gpus() < num_gpus:
return
self._run_between_graph_clients(self._test_reduce_indexed_slices,
self._cluster_spec, num_gpus, batch_reduce)
# Collective ops doesn't support strategy with one device.
@combinations.generate(
combinations.combine(
mode=["graph"],
num_gpus=[2],
required_gpus=2,
use_strategy_object=[True, False]))
def testReductionLocal(self, num_gpus, use_strategy_object):
if context.num_gpus() < num_gpus:
return
self._test_reduction(
None,
None,
num_gpus,
use_strategy_object=use_strategy_object,
local_mode=True)
@combinations.generate(
combinations.combine(
mode=["graph"],
num_gpus=[2],
required_gpus=2,
batch_reduce=[True, False]))
def testReduceIndexedSlicesLocal(self, num_gpus, batch_reduce):
self._test_reduce_indexed_slices(
None, None, num_gpus, batch_reduce, local_mode=True)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/cross_device_ops_test.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Multi-GPU tests for MirroredStrategy."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import json
import sys
from absl.testing import parameterized
import numpy as np
from tensorflow.core.protobuf import config_pb2
from tensorflow.python.data.ops import dataset_ops
from tensorflow.python.distribute import combinations
from tensorflow.python.distribute import cross_device_ops as cross_device_ops_lib
from tensorflow.python.distribute import device_util
from tensorflow.python.distribute import distribution_strategy_context as ds_context
from tensorflow.python.distribute import mirrored_strategy
from tensorflow.python.distribute import multi_worker_test_base
from tensorflow.python.distribute import reduce_util
from tensorflow.python.distribute import strategy_combinations
from tensorflow.python.distribute import strategy_test_lib
from tensorflow.python.distribute import values
from tensorflow.python.eager import backprop
from tensorflow.python.eager import context
from tensorflow.python.eager import def_function
from tensorflow.python.eager import function
from tensorflow.python.eager import test
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import func_graph
from tensorflow.python.framework import ops
from tensorflow.python.framework import tensor_shape
from tensorflow.python.framework import tensor_util
from tensorflow.python.keras.engine import training as keras_training
from tensorflow.python.keras.layers import core as keras_core
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import variable_scope
from tensorflow.python.ops import variables
from tensorflow.python.training import gradient_descent
from tensorflow.python.training import optimizer as optimizer_lib
from tensorflow.python.training import server_lib
GPU_TEST = "test_gpu" in sys.argv[0]
@combinations.generate(
combinations.combine(
distribution=[
strategy_combinations.mirrored_strategy_with_gpu_and_cpu,
strategy_combinations.mirrored_strategy_with_two_gpus,
],
mode=["graph", "eager"]))
class MirroredTwoDeviceDistributionTest(
strategy_test_lib.DistributionTestBase,
strategy_test_lib.TwoDeviceDistributionTestBase,
parameterized.TestCase):
def testMinimizeLoss(self, distribution):
if context.executing_eagerly():
self._test_minimize_loss_eager(distribution)
else:
self._test_minimize_loss_graph(distribution)
def testReplicaId(self, distribution):
self._test_replica_id(distribution)
def testNumReplicasInSync(self, distribution):
self.assertEqual(2, distribution.num_replicas_in_sync)
def testCallAndMergeExceptions(self, distribution):
self._test_call_and_merge_exceptions(distribution)
def testRunRegroupError(self, distribution):
def run_fn():
replica_id = int(self.evaluate(_replica_id()))
# Generates a list with different lengths on different devices.
# Will fail in _regroup() (if more than one device).
return list(range(replica_id))
with distribution.scope(), self.assertRaises(AssertionError):
distribution.extended.call_for_each_replica(run_fn)
def testReduceToCpu(self, distribution):
with distribution.scope():
result = distribution.extended.call_for_each_replica(_replica_id)
reduced = distribution.reduce(reduce_util.ReduceOp.SUM, result, axis=None)
expected = sum(range(distribution.num_replicas_in_sync))
self.assertEqual(expected, self.evaluate(reduced))
def reduce_axis_helper(self, distribution, replica_squared_fn):
with distribution.scope():
num_replicas = distribution.num_replicas_in_sync
result = distribution.extended.call_for_each_replica(replica_squared_fn)
# sum
reduced = distribution.reduce(reduce_util.ReduceOp.SUM, result, axis=0)
expected = sum(x * (x + 1) for x in range(num_replicas))
self.assertNear(expected, self.evaluate(reduced), 0.00001)
# mean
reduced = distribution.reduce(reduce_util.ReduceOp.MEAN, result, axis=0)
expected /= sum(x + 1 for x in range(num_replicas))
self.assertNear(expected, self.evaluate(reduced), 0.00001)
def testReduceAxisToCpu(self, distribution):
for dtype in (dtypes.float32, dtypes.int32):
def replica_squared_fn(dtype=dtype):
# Lists with different lengths on different replicas.
replica_id = _replica_id_as_int()
return math_ops.cast([replica_id] * (replica_id + 1), dtype)
self.reduce_axis_helper(distribution, replica_squared_fn)
def set_v2_tensorshape(self, v2):
if v2:
tensor_shape.enable_v2_tensorshape()
else:
tensor_shape.disable_v2_tensorshape()
def testReduceAxisToCpuUnknownShape(self, distribution):
original_v2 = tensor_shape._TENSORSHAPE_V2_OVERRIDE # pylint: disable=protected-access
try:
for v2 in (False, True):
self.set_v2_tensorshape(v2)
for dtype in (dtypes.float32, dtypes.int32):
for shape in ((None,), None): # Test both unknown size and rank.
def replica_squared_fn(dtype=dtype, shape=shape):
# Lists with different lengths on different replicas.
replica_id = _replica_id_as_int()
tensor = math_ops.cast([replica_id] * (replica_id + 1), dtype)
# Erase shape information
return array_ops.placeholder_with_default(tensor, shape=shape)
self.reduce_axis_helper(distribution, replica_squared_fn)
finally:
self.set_v2_tensorshape(original_v2)
def testReplicateDataset(self, distribution):
dataset_fn = lambda: dataset_ops.Dataset.range(10)
expected_values = [[i, i+1] for i in range(0, 10, 2)]
input_fn = self._input_fn_to_test_input_context(
dataset_fn,
expected_num_replicas_in_sync=2,
expected_num_input_pipelines=1,
expected_input_pipeline_id=0)
self._test_input_fn_iterable(distribution, input_fn, expected_values)
def testMakeInputFnIteratorWithDataset(self, distribution):
dataset_fn = lambda: dataset_ops.Dataset.range(10)
expected_values = [[i, i+1] for i in range(0, 10, 2)]
input_fn = self._input_fn_to_test_input_context(
dataset_fn,
expected_num_replicas_in_sync=2,
expected_num_input_pipelines=1,
expected_input_pipeline_id=0)
iterator = distribution.make_input_fn_iterator(input_fn)
self._test_input_fn_iterator(iterator, distribution.extended.worker_devices,
expected_values)
def testMakeInputFnIteratorWithCallable(self, distribution):
def fn():
dataset = dataset_ops.Dataset.range(2).interleave(
(lambda _: dataset_ops.Dataset.range(10)), cycle_length=2)
it = dataset.make_one_shot_iterator()
return it.get_next
expected_values = [[i, i] for i in range(0, 10)]
input_fn = self._input_fn_to_test_input_context(
fn,
expected_num_replicas_in_sync=2,
expected_num_input_pipelines=1,
expected_input_pipeline_id=0)
iterator = distribution.make_input_fn_iterator(input_fn)
self._test_input_fn_iterator(iterator, distribution.extended.worker_devices,
expected_values, test_reinitialize=False,
ignore_order=True)
def testNumpyDataset(self, distribution):
self._test_numpy_dataset(distribution)
def testGlobalStepUpdate(self, distribution):
self._test_global_step_update(distribution)
def testRun(self, distribution):
self._test_run(distribution)
def testAllReduceSum(self, distribution):
self._test_all_reduce_sum(distribution)
def testAllReduceSumGradients(self, distribution):
self._test_all_reduce_sum_gradients(distribution)
def testAllReduceSumGradientTape(self, distribution):
self._test_all_reduce_sum_gradient_tape(distribution)
def testAllReduceMean(self, distribution):
self._test_all_reduce_mean(distribution)
def testAllReduceMeanGradients(self, distribution):
self._test_all_reduce_mean_gradients(distribution)
def testAllReduceMeanGradientTape(self, distribution):
self._test_all_reduce_mean_gradient_tape(distribution)
def testSummaryForReplicaZeroOnly(self, distribution):
self._test_summary_for_replica_zero_only(distribution)
def testTrainableVariables(self, distribution):
self._test_trainable_variable(distribution)
def one_device_combinations():
return combinations.combine(
distribution=[
strategy_combinations.mirrored_strategy_with_one_cpu,
strategy_combinations.mirrored_strategy_with_one_gpu,
],
mode=["graph", "eager"])
@combinations.generate(one_device_combinations())
class MirroredOneDeviceDistributionTest(
strategy_test_lib.DistributionTestBase,
strategy_test_lib.OneDeviceDistributionTestBase,
parameterized.TestCase):
def testMinimizeLoss(self, distribution):
if context.executing_eagerly():
self._test_minimize_loss_eager(distribution)
else:
self._test_minimize_loss_graph(distribution)
def testReplicaId(self, distribution):
self._test_replica_id(distribution)
def testCallAndMergeExceptions(self, distribution):
self._test_call_and_merge_exceptions(distribution)
def testRun(self, distribution):
self._test_run(distribution)
def testAllReduceSum(self, distribution):
self._test_all_reduce_sum(distribution)
def testAllReduceSumGradients(self, distribution):
self._test_all_reduce_sum_gradients(distribution)
def testAllReduceSumGradientTape(self, distribution):
self._test_all_reduce_sum_gradient_tape(distribution)
def testAllReduceMean(self, distribution):
self._test_all_reduce_mean(distribution)
def testAllReduceMeanGradients(self, distribution):
self._test_all_reduce_mean_gradients(distribution)
def testAllReduceMeanGradientTape(self, distribution):
self._test_all_reduce_mean_gradient_tape(distribution)
class MirroredStrategyVariableCreatorStackTest(
test.TestCase, parameterized.TestCase):
@combinations.generate(
combinations.combine(
distribution=[
strategy_combinations.mirrored_strategy_with_gpu_and_cpu,
],
mode=["graph"]))
def testCreatorStacksAreThreadLocal(self, distribution):
def model_fn():
replica_id_str = str(self.evaluate(_replica_id()))
def thread_creator_fn(next_creator, *args, **kwargs):
return next_creator(*args, **kwargs) + ":thread_" + replica_id_str
with variable_scope.variable_creator_scope(thread_creator_fn):
# Create a variable in this scope.
v = variable_scope.variable(1.0)
# This will pause the current thread, and execute the other thread.
ds_context.get_replica_context().merge_call(lambda _: _)
return v
def main_thread_creator(next_creator, *args, **kwargs):
# We are not using the underlying next_creator for test purposes.
del next_creator, args, kwargs
return "main_thread"
with context.graph_mode(), \
distribution.scope(), \
variable_scope.variable_creator_scope(main_thread_creator):
result = distribution.extended.call_for_each_replica(model_fn)
result = distribution.experimental_local_results(result)
expected = ("main_thread:thread_0", "main_thread:thread_1")
self.assertEqual(expected, result)
@combinations.generate(
combinations.combine(
distribution=[
strategy_combinations.mirrored_strategy_with_gpu_and_cpu,
],
mode=["graph", "eager"]))
class MirroredStrategyCallForEachReplicaTest(test.TestCase):
def testExecutingEagerlyOutsideFunction(self, distribution):
"""Verify we preserve the value of executing_eagerly_outside_functions()."""
def model_fn():
return ops.executing_eagerly_outside_functions()
originally = ops.executing_eagerly_outside_functions()
with distribution.scope():
in_scope = ops.executing_eagerly_outside_functions()
in_model_fn = distribution.extended.call_for_each_replica(model_fn)
unwrapped = distribution.experimental_local_results(in_model_fn)
self.assertEqual(in_scope, unwrapped[0])
self.assertEqual(in_scope, originally)
# Verify this all again, but this time in a FuncGraph.
with func_graph.FuncGraph("fg").as_default(), distribution.scope():
in_scope = ops.executing_eagerly_outside_functions()
in_model_fn = distribution.extended.call_for_each_replica(model_fn)
unwrapped = distribution.experimental_local_results(in_model_fn)
self.assertEqual(in_scope, unwrapped[0])
self.assertEqual(in_scope, originally)
def testFunctionInCallForEachReplica(self, distribution):
traces = []
@def_function.function
def model_fn():
traces.append(1)
return ds_context.get_replica_context().replica_id_in_sync_group
with distribution.scope():
result = distribution.extended.call_for_each_replica(model_fn)
self.assertEqual((0, 1), self.evaluate(result.values))
self.assertLen(traces, distribution.num_replicas_in_sync)
def testFunctionInCallForEachReplicaInsideAnotherFunction(self, distribution):
traces = []
@def_function.function
def model_fn():
traces.append(1)
return ds_context.get_replica_context().replica_id_in_sync_group
@def_function.function
def step():
return distribution.extended.call_for_each_replica(model_fn)
with distribution.scope():
result = step()
self.assertEqual((0, 1), self.evaluate(result.values))
self.assertLen(traces, distribution.num_replicas_in_sync)
def testFunctionInCallForEachReplicaWithMergeCall(self, distribution):
def merge_fn(_):
pass
@def_function.function
def model_fn():
ds_context.get_replica_context().merge_call(merge_fn)
return 0.
with distribution.scope():
with self.assertRaisesRegexp(
RuntimeError, "`merge_call` called while defining a new graph."):
distribution.extended.call_for_each_replica(model_fn)
@combinations.generate(
combinations.combine(
distribution=[
strategy_combinations.mirrored_strategy_with_gpu_and_cpu,
],
mode=["graph"]))
class MirroredStrategyNameScopeTest(test.TestCase):
# NOTE(priyag): Names and name scopes are ignored in eager, hence we are not
# testing this in eager mode.
def testNameScope(self, distribution):
def model_fn():
with ops.name_scope("foo"):
a = constant_op.constant(1.0, name="a")
ds_context.get_replica_context().merge_call(lambda _: _)
b = constant_op.constant(1.0, name="b")
return a, b
with context.graph_mode(), distribution.scope():
with ops.name_scope("main"):
result = distribution.extended.call_for_each_replica(model_fn)
self.assertEqual(2, len(result))
for v, name in zip(result, ["a", "b"]):
self.assertIsInstance(v, values.DistributedValues)
v0, v1 = distribution.experimental_local_results(v)
self.assertEqual("main/foo/" + name + ":0", v0.name)
self.assertEqual("main/replica_1/foo/" + name + ":0", v1.name)
def testWithDefaultName(self, distribution):
def model_fn():
with ops.name_scope(None, "foo"):
a = constant_op.constant(1.0, name="a")
ds_context.get_replica_context().merge_call(lambda _: _)
b = constant_op.constant(2.0, name="b")
return a, b
with context.graph_mode(), distribution.scope():
result = distribution.extended.call_for_each_replica(model_fn)
self.assertEqual(2, len(result))
for v, name in zip(result, ["a", "b"]):
self.assertIsInstance(v, values.DistributedValues)
v0, v1 = distribution.experimental_local_results(v)
self.assertEqual("foo/" + name + ":0", v0.name)
self.assertEqual("replica_1/foo/" + name + ":0", v1.name)
# variable_scope.variable() respects name scopes when creating
# variables. On the other hand variable_scope.get_variable() ignores name
# scopes but respects variable scope when creating variables. We test both
# methods of creating variables to make sure that we have the same
# variable names in both cases.
def testNameScopeWithVariable(self, distribution):
def in_cross_replica(_):
c = variable_scope.variable(1.0, name="c")
return c
def model_fn():
b = variable_scope.variable(1.0, name="b")
with ops.name_scope("foo"):
c = ds_context.get_replica_context().merge_call(in_cross_replica)
return b, c
with context.graph_mode(), distribution.scope():
with ops.name_scope("main"):
a = variable_scope.variable(1.0, name="a")
result = distribution.extended.call_for_each_replica(model_fn)
result_b = result[0]
result_c = result[1]
self.assertIsInstance(result_b, values.DistributedValues)
self.assertIsInstance(result_c, values.DistributedValues)
a0, a1 = distribution.experimental_local_results(a)
b0, b1 = distribution.experimental_local_results(result_b)
c0, c1 = distribution.experimental_local_results(result_c)
self.assertEqual("main/a:0", a0.name)
self.assertEqual("main/a/replica_1:0", a1.name)
self.assertEqual("main/b:0", b0.name)
self.assertEqual("main/b/replica_1:0", b1.name)
self.assertEqual("main/foo/c:0", c0.name)
self.assertEqual("main/foo/c/replica_1:0", c1.name)
def testNameScopeWithGetVariable(self, distribution):
def in_cross_replica(_):
c = variable_scope.get_variable("c", [1])
return c
def model_fn():
b = variable_scope.get_variable("b", [1])
with ops.name_scope("foo"):
c = ds_context.get_replica_context().merge_call(in_cross_replica)
return b, c
with context.graph_mode(), distribution.scope():
with ops.name_scope("main"):
a = variable_scope.get_variable("a", [1])
result = distribution.extended.call_for_each_replica(model_fn)
result_b = result[0]
result_c = result[1]
self.assertIsInstance(result_b, values.DistributedValues)
self.assertIsInstance(result_c, values.DistributedValues)
a0, a1 = distribution.experimental_local_results(a)
b0, b1 = distribution.experimental_local_results(result_b)
c0, c1 = distribution.experimental_local_results(result_c)
self.assertEqual("a:0", a0.name)
self.assertEqual("a/replica_1:0", a1.name)
self.assertEqual("b:0", b0.name)
self.assertEqual("b/replica_1:0", b1.name)
self.assertEqual("c:0", c0.name)
self.assertEqual("c/replica_1:0", c1.name)
def testVariableScopeWithGetVariable(self, distribution):
def in_cross_replica(_):
c = variable_scope.get_variable("c", [1])
return c
def model_fn():
b = variable_scope.get_variable("b", [1])
with variable_scope.variable_scope("foo"):
c = ds_context.get_replica_context().merge_call(in_cross_replica)
return b, c
with context.graph_mode(), distribution.scope():
with variable_scope.variable_scope("main"):
a = variable_scope.get_variable("a", [1])
result = distribution.extended.call_for_each_replica(model_fn)
result_b = result[0]
result_c = result[1]
self.assertIsInstance(result_b, values.DistributedValues)
self.assertIsInstance(result_c, values.DistributedValues)
a0, a1 = distribution.experimental_local_results(a)
b0, b1 = distribution.experimental_local_results(result_b)
c0, c1 = distribution.experimental_local_results(result_c)
self.assertEqual("main/a:0", a0.name)
self.assertEqual("main/a/replica_1:0", a1.name)
self.assertEqual("main/b:0", b0.name)
self.assertEqual("main/b/replica_1:0", b1.name)
self.assertEqual("main/foo/c:0", c0.name)
self.assertEqual("main/foo/c/replica_1:0", c1.name)
@combinations.generate(
combinations.combine(
distribution=[
combinations.NamedDistribution(
"Mirrored3Devices",
# pylint: disable=g-long-lambda
lambda: mirrored_strategy.MirroredStrategy(
["/device:GPU:0", "/device:GPU:1", "/device:CPU:0"]),
required_gpus=2)
],
mode=["graph", "eager"]))
class MirroredThreeDeviceDistributionTest(
strategy_test_lib.DistributionTestBase,
parameterized.TestCase):
def testThreeDevices(self, distribution):
def model_fn():
v = variable_scope.variable(1.0, name="foo")
ds_context.get_replica_context().merge_call(lambda _: _)
return v
with distribution.scope():
result = distribution.extended.call_for_each_replica(model_fn)
self.assertIsInstance(result, values.MirroredVariable)
self.assertEqual("foo:0", result.name)
@combinations.generate(
combinations.combine(
distribution=[
strategy_combinations.mirrored_strategy_with_gpu_and_cpu,
],
mode=["graph", "eager"]))
class MirroredVariableUpdateTest(test.TestCase):
# The following tests check assign, assign_add and assign_sub on Mirrored
# variables in replica and cross replica context.
def testAssignMirroredVarReplicaContextWithoutAggregationType(self,
distribution):
# Test that we always have an aggregation type set on the mirrored variable
# if we assign to it in replica mode.
def var_fn():
v = variable_scope.variable(1.0, name="foo")
return v
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.evaluate(variables.global_variables_initializer())
def model_fn():
return mirrored_var.assign(5.0)
with self.assertRaisesRegexp(
ValueError, "You must specify an aggregation method to update a "
"MirroredVariable in Replica Context. You can do so by"):
self.evaluate(distribution.experimental_local_results(
distribution.extended.call_for_each_replica(model_fn)))
def testAssignMirroredVarReplicaContextWithSum(self, distribution):
# Test that we don't reduce a non-per-replica value with the "sum"
# aggregation type.
def var_fn():
v = variable_scope.variable(
1.0, name="foo", aggregation=variable_scope.VariableAggregation.SUM)
return v
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.evaluate(variables.global_variables_initializer())
def model_fn():
return mirrored_var.assign(5.0)
with self.assertRaisesRegexp(
ValueError, "A non-DistributedValues value 5.0 cannot be reduced "
"with the given reduce op ReduceOp.SUM."):
self.evaluate(distribution.experimental_local_results(
distribution.extended.call_for_each_replica(model_fn)))
def testAssignMirroredVarCrossDeviceContext(self, distribution):
def var_fn():
return variable_scope.variable(1.0, name="foo")
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.evaluate(variables.global_variables_initializer())
self.assertEqual(1.0, self.evaluate(mirrored_var))
mirrored_var_result = self.evaluate(mirrored_var.assign(6.0))
self.assertEqual(6.0, mirrored_var_result)
def testAssignMirroredVarReplicaContext(self, distribution):
def var_fn():
return variable_scope.variable(
1.0, name="foo", aggregation=variable_scope.VariableAggregation.MEAN)
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.evaluate(variables.global_variables_initializer())
self.assertEqual(1.0, self.evaluate(mirrored_var))
def model_fn():
value = math_ops.cast(
ds_context.get_replica_context().replica_id_in_sync_group,
mirrored_var.dtype)
return mirrored_var.assign(value)
self.evaluate(distribution.experimental_local_results(
distribution.extended.call_for_each_replica(model_fn)))
self.assertEqual(0.5, self.evaluate(mirrored_var))
def testAssignMirroredVarReplicaContextWithSingleValue(self, distribution):
def var_fn():
return variable_scope.variable(
1.0, name="foo", aggregation=variable_scope.VariableAggregation.MEAN)
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.evaluate(variables.global_variables_initializer())
self.assertEqual(1.0, self.evaluate(mirrored_var))
def model_fn():
return mirrored_var.assign(5.0)
self.evaluate(distribution.experimental_local_results(
distribution.extended.call_for_each_replica(model_fn)))
self.assertEqual(5.0, self.evaluate(mirrored_var))
def testAssignAddMirroredVarCrossDeviceContext(self, distribution):
def var_fn():
return variable_scope.variable(1.0, name="foo")
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.evaluate(variables.global_variables_initializer())
self.assertEqual(1.0, self.evaluate(mirrored_var))
# read_value == True
mirrored_var_result = self.evaluate(
mirrored_var.assign_add(6.0, read_value=True))
self.assertEqual(7.0, mirrored_var_result)
self.assertEqual(7.0, self.evaluate(mirrored_var.get("/device:CPU:0")))
self.assertEqual(7.0, self.evaluate(mirrored_var.get("/device:GPU:0")))
# read_value == False
self.evaluate(mirrored_var.assign_add(2.0, read_value=False))
self.assertEqual(9.0, self.evaluate(mirrored_var.get("/device:CPU:0")))
self.assertEqual(9.0, self.evaluate(mirrored_var.get("/device:GPU:0")))
def testAssignAddMirroredVarReplicaContext(self, distribution):
def var_fn():
return variable_scope.variable(
1.0, name="foo", aggregation=variable_scope.VariableAggregation.MEAN)
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.evaluate(variables.global_variables_initializer())
self.assertEqual(1.0, self.evaluate(mirrored_var))
def model_fn():
value = math_ops.cast(
ds_context.get_replica_context().replica_id_in_sync_group,
mirrored_var.dtype)
return mirrored_var.assign_add(value)
self.evaluate(distribution.experimental_local_results(
distribution.extended.call_for_each_replica(model_fn)))
self.assertEqual(1.5, self.evaluate(mirrored_var))
def testAssignAddMirroredVarReplicaContextWithSingleValue(self, distribution):
def var_fn():
return variable_scope.variable(
1.0, name="foo", aggregation=variable_scope.VariableAggregation.MEAN)
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.evaluate(variables.global_variables_initializer())
self.assertEqual(1.0, self.evaluate(mirrored_var))
def model_fn():
return mirrored_var.assign_add(5.0)
self.evaluate(distribution.experimental_local_results(
distribution.extended.call_for_each_replica(model_fn)))
self.assertEqual(6.0, self.evaluate(mirrored_var))
def testAssignSubMirroredVarCrossDeviceContext(self, distribution):
def var_fn():
return variable_scope.variable(5.0, name="foo")
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.evaluate(variables.global_variables_initializer())
self.assertEqual(5.0, self.evaluate(mirrored_var))
mirrored_var_result = self.evaluate(mirrored_var.assign_sub(2.0))
self.assertEqual(3.0, mirrored_var_result)
self.assertEqual(3.0, self.evaluate(mirrored_var.get("/device:GPU:0")))
self.assertEqual(3.0, self.evaluate(mirrored_var.get("/device:CPU:0")))
def testAssignSubMirroredVarReplicaContext(self, distribution):
def var_fn():
return variable_scope.variable(
5.0, name="foo", aggregation=variable_scope.VariableAggregation.MEAN)
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.evaluate(variables.global_variables_initializer())
self.assertEqual(5.0, self.evaluate(mirrored_var))
def model_fn():
value = math_ops.cast(
ds_context.get_replica_context().replica_id_in_sync_group,
mirrored_var.dtype)
return mirrored_var.assign_sub(value)
self.evaluate(distribution.experimental_local_results(
distribution.extended.call_for_each_replica(model_fn)))
self.assertEqual(4.5, self.evaluate(mirrored_var))
def testAssignSubMirroredVarReplicaContextWithSingleValue(self, distribution):
def var_fn():
return variable_scope.variable(
5.0, name="foo", aggregation=variable_scope.VariableAggregation.MEAN)
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.evaluate(variables.global_variables_initializer())
self.assertEqual(5.0, self.evaluate(mirrored_var))
def model_fn():
return mirrored_var.assign_sub(1.0)
self.evaluate(distribution.experimental_local_results(
distribution.extended.call_for_each_replica(model_fn)))
self.assertEqual(4.0, self.evaluate(mirrored_var))
@combinations.generate(
combinations.combine(
distribution=[
strategy_combinations.mirrored_strategy_with_gpu_and_cpu,
],
mode=["graph", "eager"]))
class MirroredAndSyncOnReadVariableInitializerTest(test.TestCase):
def testAssignMirroredVarInitializer(self, distribution):
# This test is not eager compatible since in eager variables are initialized
# upon construction instead of once the initialization op is run.
with context.graph_mode():
def var_fn():
v = variable_scope.variable(1.0, name="foo")
return v
with distribution.scope():
mirrored_var = distribution.extended.call_for_each_replica(var_fn)
self.assertIsInstance(mirrored_var, values.MirroredVariable)
self.assertFalse(self.evaluate(mirrored_var.is_initialized()))
self.evaluate(mirrored_var.initializer)
self.assertTrue(self.evaluate(mirrored_var.is_initialized()))
def testAssignReplicaLocalVarInitializer(self, distribution):
# This test is not eager compatible since in eager variables are initialized
# upon construction instead of once the initialization op is run.
with context.graph_mode():
def model_fn():
v_sum = variable_scope.variable(
1.0,
synchronization=variable_scope.VariableSynchronization.ON_READ,
aggregation=variable_scope.VariableAggregation.SUM)
self.assertIsInstance(v_sum, values.SyncOnReadVariable)
return v_sum
with distribution.scope():
sync_on_read_var = distribution.extended.call_for_each_replica(
model_fn)
self.assertIsInstance(sync_on_read_var, values.SyncOnReadVariable)
self.assertFalse(self.evaluate(sync_on_read_var.is_initialized()))
self.evaluate(sync_on_read_var.initializer)
self.assertTrue(self.evaluate(sync_on_read_var.is_initialized()))
@combinations.generate(
combinations.combine(
distribution=[
strategy_combinations.mirrored_strategy_with_gpu_and_cpu,
],
mode=["graph", "eager"]))
class SyncOnReadVariableAssignTest(test.TestCase):
def testAssignReplicaLocalVarSumAggregation(self, distribution):
def model_fn():
v_sum = variable_scope.variable(
1.0,
synchronization=variable_scope.VariableSynchronization.ON_READ,
aggregation=variable_scope.VariableAggregation.SUM)
return v_sum
with distribution.scope():
sync_on_read_var = distribution.extended.call_for_each_replica(model_fn)
self.assertIsInstance(sync_on_read_var, values.SyncOnReadVariable)
self.evaluate(variables.global_variables_initializer())
# Each replica has a value of 1.0 assigned to it in replica context.
# When we read the value using `read_var` we should see the SUM of each of
# values on each of the replicas.
self.assertEqual(2.0, self.evaluate(
distribution.extended.read_var(sync_on_read_var)))
# Assigning 6.0 in cross replica context will assign a value of
# 6.0/num_replicas to each replica.
tlv_ops = sync_on_read_var.assign(6.0)
self.evaluate(tlv_ops)
# On reading the sync on read var we should get the assigned value back.
# The value on all the replicas are added before being returned by
# `read_var`.
self.assertEqual(6.0, self.evaluate(
distribution.extended.read_var(sync_on_read_var)))
def testAssignReplicaLocalVarMeanAggregation(self, distribution):
def model_fn():
v_sum = variable_scope.variable(
1.0,
synchronization=variable_scope.VariableSynchronization.ON_READ,
aggregation=variable_scope.VariableAggregation.MEAN)
return v_sum
with distribution.scope():
sync_on_read_var = distribution.extended.call_for_each_replica(model_fn)
self.assertIsInstance(sync_on_read_var, values.SyncOnReadVariable)
self.evaluate(variables.global_variables_initializer())
# Each replica has a value of 1.0 assigned to it in replica context.
# When we read the value using `read_var` we should see the MEAN of values
# on all replicas which is the value assigned in replica context.
self.assertEqual(1.0, self.evaluate(
distribution.extended.read_var(sync_on_read_var)))
tlv_ops = sync_on_read_var.assign(6.0)
self.evaluate(tlv_ops)
# On reading the sync on read var we should get the MEAN of all values
# which is equal to the value assigned.
self.assertEqual(6.0, self.evaluate(
distribution.extended.read_var(sync_on_read_var)))
class MockModel(object):
def __init__(self, two_variables=False):
self.variables = []
self.variables.append(variable_scope.variable(1.25, name="dummy_var1"))
if two_variables:
self.variables.append(variable_scope.variable(2.0, name="dummy_var2"))
def __call__(self, factor=2):
x = factor * self.variables[0]
if len(self.variables) > 1:
x += self.variables[1]
return x
class MiniModel(keras_training.Model):
"""Minimal model for mnist.
Useful for testing and debugging on slow TPU simulators.
"""
def __init__(self):
super(MiniModel, self).__init__(name="")
self.fc = keras_core.Dense(1, name="fc", kernel_initializer="ones",
bias_initializer="ones")
def call(self, inputs, training=True):
inputs = array_ops.ones([1, 10])
return self.fc(inputs)
@combinations.generate(
combinations.combine(
distribution=[
strategy_combinations.mirrored_strategy_with_gpu_and_cpu,
],
mode=["graph", "eager"]))
class MirroredStrategyDefunTest(test.TestCase):
def _call_and_check(self, distribution, model_fn, inputs, expected_result,
defuns, two_variables=False):
cpu_dev = device_util.canonicalize("CPU:0")
gpu_dev = device_util.canonicalize("GPU:0")
devices = [cpu_dev, gpu_dev]
with distribution.scope():
mock_model = MockModel(two_variables)
self.evaluate(variables.global_variables_initializer())
result = distribution.extended.call_for_each_replica(
model_fn, args=[mock_model] + inputs)
for r in range(len(devices)):
device_result = values.select_replica(r, result)
device_expected_result = values.select_replica(r, expected_result)
self.assertAllClose(device_expected_result,
self.evaluate(device_result))
for defun in defuns:
# `Function`s are specialized to the current device stack, so
# call_for_each has one trace per device. To check that the expected set
# of variables was accessed on each trace, we first retrieve each
# device-specific graph function.
per_replica_graph_functions = (
distribution.extended.call_for_each_replica(
defun.get_concrete_function, args=[mock_model] + inputs))
for device in devices:
graph_function = per_replica_graph_functions.get(device=device)
# TODO(b/129555712): re-enable an assertion here that the two sets of
# variables are the same.
# self.assertEqual(set(graph_function.graph.variables),
# set(mock_model.variables))
del graph_function
def testVariableInDefun(self, distribution):
@function.defun
def times_two(mock_model):
return mock_model()
def model_fn(mock_model):
return times_two(mock_model)
self._call_and_check(distribution, model_fn, [], 2.5, [times_two])
def testVariableInNestedDefun(self, distribution):
@function.defun
def times_two(mock_model):
return mock_model()
@function.defun
def two_x_plus_one(mock_model):
return times_two(mock_model) + 1
def model_fn(mock_model):
return two_x_plus_one(mock_model)
self._call_and_check(distribution, model_fn, [], 3.5,
[times_two, two_x_plus_one])
def testTwoVariablesInNestedDefun(self, distribution):
@function.defun
def fn1(mock_model):
return mock_model()
@function.defun
def fn2(mock_model):
return fn1(mock_model) + 1
def model_fn(mock_model):
return fn2(mock_model)
self._call_and_check(distribution, model_fn, [], 5.5, [fn1, fn2],
two_variables=True)
def testGradientTapeOverNestedDefuns(self, distribution):
@function.defun
def fn1(mock_model):
return mock_model()
@function.defun
def fn2(mock_model):
return fn1(mock_model) + 1
def model_fn(mock_model):
with backprop.GradientTape(persistent=True) as gtape:
result = fn2(mock_model)
grads = gtape.gradient(result,
[v.get() for v in mock_model.variables])
return grads
self._call_and_check(distribution, model_fn, [], [2.0, 1.0], [fn1, fn2],
two_variables=True)
def testPassPerReplica(self, distribution):
@function.defun
def fn1(mock_model, factor):
return mock_model(factor)
device_map = values.ReplicaDeviceMap(("/device:CPU:0", "/device:GPU:0"))
factors = values.PerReplica(device_map, (5.0, 3.0))
expected_result = values.PerReplica(device_map, (5.0 * 1.25, 3.0 * 1.25))
self._call_and_check(distribution, fn1, [factors], expected_result, [fn1])
def testTrain(self, distribution):
with distribution.scope():
mock_model = MiniModel()
mock_model.call = function.defun(mock_model.call)
def loss_fn(ctx):
del ctx
return mock_model(array_ops.ones([1, 10]))
gradients_fn = backprop.implicit_grad(loss_fn)
gradients_fn = optimizer_lib.get_filtered_grad_fn(gradients_fn)
grads_and_vars = distribution.extended.call_for_each_replica(
gradients_fn, args=(None,))
optimizer = gradient_descent.GradientDescentOptimizer(0.25)
update_ops = optimizer._distributed_apply(distribution, grads_and_vars) # pylint: disable=protected-access
if not context.executing_eagerly():
self.evaluate(variables.global_variables_initializer())
self.evaluate(update_ops)
updated_var_values = self.evaluate(mock_model.variables)
# All variables start at 1.0 and get two updates of 0.25.
self.assertAllEqual(0.5 * np.ones([10, 1]), updated_var_values[0])
self.assertAllEqual([0.5], updated_var_values[1])
@combinations.generate(
combinations.combine(
distribution=[
combinations.NamedDistribution(
"Mirrored",
# pylint: disable=g-long-lambda
lambda: mirrored_strategy.MirroredStrategy(
devices=mirrored_strategy.all_local_devices(),
cross_device_ops=cross_device_ops_lib.MultiWorkerAllReduce([
"/job:worker/task:0", "/job:worker/task:1"
], context.num_gpus())),
required_gpus=1)
],
mode=["graph"]))
class MultiWorkerMirroredStrategyTest(
multi_worker_test_base.MultiWorkerTestBase,
strategy_test_lib.DistributionTestBase):
def _configure_distribution_strategy(self, distribution):
cluster_spec = server_lib.ClusterSpec({
"worker": ["/job:worker/task:0", "/job:worker/task:1"]
})
distribution.configure(cluster_spec=cluster_spec)
def test_num_replicas_in_sync(self, distribution):
self._configure_distribution_strategy(distribution)
# We calculate the total number of gpus across the workers(2) specified in
# the cluster spec.
self.assertEqual(context.num_gpus() * 2, distribution.num_replicas_in_sync)
def testMinimizeLossGraph(self, distribution):
self._configure_distribution_strategy(distribution)
self._test_minimize_loss_graph(distribution, learning_rate=0.05)
def testDeviceScope(self, distribution):
"""Test the device scope of multi-worker MirroredStrategy."""
self._configure_distribution_strategy(distribution)
with distribution.scope():
a = constant_op.constant(1.)
with ops.device("/cpu:0"):
b = constant_op.constant(1.)
self.assertEqual(a.device, "/job:worker/task:0")
self.assertEqual(b.device, "/job:worker/task:0/device:CPU:0")
def testMakeInputFnIteratorWithDataset(self, distribution):
self._configure_distribution_strategy(distribution)
dataset_fn = lambda: dataset_ops.Dataset.range(100)
num_gpus = context.num_gpus()
num_workers = 2
expected_values = [[i+j for j in range(num_gpus)] * num_workers
for i in range(0, 100, num_gpus)]
with context.graph_mode(), self.cached_session() as sess:
# `expected_input_pipeline_id` is None because the input_fn will be called
# multiple times, each with a different input_pipeline_id.
input_fn = self._input_fn_to_test_input_context(
dataset_fn,
expected_num_replicas_in_sync=num_workers*num_gpus,
expected_num_input_pipelines=num_workers,
expected_input_pipeline_id=None)
iterator = distribution.make_input_fn_iterator(input_fn)
self._test_input_fn_iterator(
iterator, distribution.extended.worker_devices, expected_values, sess)
def testMakeInputFnIteratorWithCallable(self, distribution):
self._configure_distribution_strategy(distribution)
def fn():
dataset = dataset_ops.Dataset.range(100)
it = dataset.make_one_shot_iterator()
return it.get_next
num_gpus = context.num_gpus()
num_workers = 2
expected_values = []
for i in range(0, 100, num_gpus):
expected_values.append([i+j for j in range(num_gpus)] * num_workers)
with context.graph_mode(), self.cached_session() as sess:
# `expected_input_pipeline_id` is None because the input_fn will be called
# multiple times, each with a different input_pipeline_id.
input_fn = self._input_fn_to_test_input_context(
fn,
expected_num_replicas_in_sync=num_workers*num_gpus,
expected_num_input_pipelines=num_workers,
expected_input_pipeline_id=None)
iterator = distribution.make_input_fn_iterator(input_fn)
self._test_input_fn_iterator(
iterator, distribution.extended.worker_devices, expected_values, sess,
test_reinitialize=False, ignore_order=True)
def testUpdateConfigProto(self, distribution):
distribution.configure(cluster_spec={"worker": ["fake1", "fake2"]})
config_proto = config_pb2.ConfigProto()
new_config = distribution.update_config_proto(config_proto)
# Verify isolate_session_state
self.assertTrue(new_config.isolate_session_state)
class MultiWorkerMirroredStrategyTestWithChief(
multi_worker_test_base.MultiWorkerTestBase,
strategy_test_lib.DistributionTestBase):
@classmethod
def setUpClass(cls):
"""Create a local cluster with 2 workers and 1 chief."""
cls._cluster_spec = multi_worker_test_base.create_in_process_cluster(
num_workers=2, num_ps=0, has_chief=True)
cls._default_target = "grpc://" + cls._cluster_spec["chief"][0]
def _make_cross_device_ops(self):
return cross_device_ops_lib.MultiWorkerAllReduce(
["/job:chief/task:0", "/job:worker/task:0", "/job:worker/task:1"],
context.num_gpus())
def testMinimizeLossGraph(self):
strategy = mirrored_strategy.MirroredStrategy(
cross_device_ops=self._make_cross_device_ops())
strategy.configure(cluster_spec=self._cluster_spec)
self._test_minimize_loss_graph(strategy, learning_rate=0.05)
def testMinimizeLossGraphMirroredStrategy(self):
strategy = mirrored_strategy.MirroredStrategy(
mirrored_strategy.all_local_devices(),
cross_device_ops=self._make_cross_device_ops())
strategy.configure(cluster_spec=self._cluster_spec)
self._test_minimize_loss_graph(strategy, learning_rate=0.05)
def testMinimizeLossGraphMirroredStrategyWithOneNode(self):
cluster_spec = {}
cluster_spec["chief"] = self._cluster_spec["chief"]
tf_config = {"cluster": cluster_spec}
with test.mock.patch.dict("os.environ",
{"TF_CONFIG": json.dumps(tf_config)}):
strategy = mirrored_strategy.MirroredStrategy()
self.assertIsInstance(strategy.extended._inferred_cross_device_ops,
cross_device_ops_lib.NcclAllReduce)
self.skipTest('b/130551176, run the following once fixed.')
self._test_minimize_loss_graph(strategy, learning_rate=0.05)
def testInitializeFromTFConfig(self):
tf_config = {"cluster": self._cluster_spec}
with test.mock.patch.dict("os.environ",
{"TF_CONFIG": json.dumps(tf_config)}):
strategy = mirrored_strategy.MirroredStrategy(
cross_device_ops=self._make_cross_device_ops())
self.assertEqual(
max(context.num_gpus(), 1) * 3, strategy.num_replicas_in_sync)
def testSummaryForReplicaZeroOnly(self):
strategy = mirrored_strategy.MirroredStrategy(
mirrored_strategy.all_local_devices(),
cross_device_ops=self._make_cross_device_ops())
strategy.configure(cluster_spec=self._cluster_spec)
self._test_summary_for_replica_zero_only(strategy)
def _replica_id():
replica_id = ds_context.get_replica_context().replica_id_in_sync_group
if not isinstance(replica_id, ops.Tensor):
replica_id = constant_op.constant(replica_id)
return replica_id
def _replica_id_as_int():
replica_id = ds_context.get_replica_context().replica_id_in_sync_group
if isinstance(replica_id, ops.Tensor):
replica_id = tensor_util.constant_value(replica_id)
return replica_id
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/mirrored_strategy_test.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Class OneDeviceStrategy implementing DistributionStrategy."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.distribute import device_util
from tensorflow.python.distribute import distribute_lib
from tensorflow.python.distribute import input_lib
from tensorflow.python.distribute import numpy_dataset
from tensorflow.python.distribute import values
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.util import nest
from tensorflow.python.util.tf_export import tf_export
# TODO(josh11b): Replace asserts in this file with if ...: raise ...
# TODO(josh11b): Do we wrap values in types to generate errors if you are
# doing something that won't work with other DistributionStrategy
# implementations?
@tf_export("distribute.OneDeviceStrategy", v1=[])
class OneDeviceStrategy(distribute_lib.Strategy):
"""A distribution strategy for running on a single device."""
def __init__(self, device):
super(OneDeviceStrategy, self).__init__(OneDeviceExtended(self, device))
@tf_export(v1=["distribute.OneDeviceStrategy"])
class OneDeviceStrategyV1(distribute_lib.StrategyV1):
"""A distribution strategy for running on a single device."""
def __init__(self, device):
super(OneDeviceStrategyV1, self).__init__(OneDeviceExtended(self, device))
# TODO(josh11b): Switch to V2 after callers have been updated to only V2 APIs.
class OneDeviceExtended(distribute_lib.StrategyExtendedV1):
"""Implementation of OneDeviceStrategy."""
def __init__(self, container_strategy, device):
super(OneDeviceExtended, self).__init__(container_strategy)
self._device = device_util.canonicalize(device)
suffix_loc = self._device.rfind("/")
self._input_device = self._device[:suffix_loc] + "/device:CPU:0"
worker_device_pairs = [(self._input_device, [self._device])]
device_map = values.SingleDeviceMap(device)
self._input_workers = input_lib.InputWorkers(
device_map, worker_device_pairs)
def _create_variable(self, next_creator, *args, **kwargs):
colocate_with = kwargs.pop("colocate_with", None)
if colocate_with is None:
with ops.device(self._device):
return next_creator(*args, **kwargs)
elif isinstance(colocate_with, numpy_dataset.SingleDevice):
with ops.device(colocate_with.device):
return next_creator(*args, **kwargs)
else:
with ops.colocate_with(colocate_with):
return next_creator(*args, **kwargs)
def _validate_colocate_with_variable(self, colocate_with_variable):
values.validate_colocate(colocate_with_variable, self)
def _make_dataset_iterator(self, dataset):
"""Make iterator from dataset without splitting the batch."""
# Note that split_batch_by argument is not passed because it is always 1 in
# this strategy, and adding it adds unnecessary overhead to the dataset.
return input_lib.DatasetIterator(dataset, self._input_workers,
self._container_strategy())
def _make_input_fn_iterator(
self,
input_fn,
replication_mode=distribute_lib.InputReplicationMode.PER_WORKER):
return input_lib.InputFunctionIterator(input_fn, self._input_workers,
[distribute_lib.InputContext()],
self._container_strategy())
def _experimental_make_numpy_dataset(self, numpy_input, session):
return numpy_dataset.one_host_numpy_dataset(
numpy_input, numpy_dataset.SingleDevice(self._input_device), session)
def _broadcast_to(self, tensor, destinations):
del destinations
return tensor
def _experimental_distribute_dataset(self, dataset):
# Note that split_batch_by argument is not passed because it is always 1 in
# this strategy, and adding it adds unnecessary overhead to the dataset.
return input_lib.get_distributed_dataset(dataset, self._input_workers,
self._container_strategy())
def _experimental_distribute_datasets_from_function(self, dataset_fn):
return input_lib.DistributedDatasetsFromFunction(
dataset_fn,
self._input_workers,
[distribute_lib.InputContext()],
self._container_strategy())
# TODO(priyag): Deal with OutOfRange errors once b/111349762 is fixed.
def _experimental_run_steps_on_iterator(self, fn, iterator, iterations,
initial_loop_values=None):
if initial_loop_values is None:
initial_loop_values = {}
initial_loop_values = nest.flatten(initial_loop_values)
ctx = input_lib.MultiStepContext()
def body(i, *args):
"""A wrapper around `fn` to create the while loop body."""
del args
fn_result = fn(ctx, iterator.get_next())
flat_last_step_outputs = nest.flatten(ctx.last_step_outputs)
with ops.control_dependencies([fn_result]):
return [i + 1] + flat_last_step_outputs
# We capture the control_flow_context at this point, before we run `fn`
# inside a while_loop. This is useful in cases where we might need to exit
# these contexts and get back to the outer context to do some things, for
# e.g. create an op which should be evaluated only once at the end of the
# loop on the host. One such usage is in creating metrics' value op.
self._outer_control_flow_context = (
ops.get_default_graph()._get_control_flow_context()) # pylint: disable=protected-access
# TODO(priyag): Use max_iterations instead of an explicit counter.
cond = lambda i, *args: i < iterations
i = constant_op.constant(0)
loop_result = control_flow_ops.while_loop(
cond, body, [i] + initial_loop_values, name="",
parallel_iterations=1, back_prop=False, swap_memory=False,
return_same_structure=True)
del self._outer_control_flow_context
ctx.run_op = control_flow_ops.group(loop_result)
# Convert the last_step_outputs from a list to the original dict structure
# of last_step_outputs.
last_step_tensor_outputs = loop_result[1:]
last_step_tensor_outputs_dict = nest.pack_sequence_as(
ctx.last_step_outputs, last_step_tensor_outputs)
ctx._set_last_step_outputs(last_step_tensor_outputs_dict) # pylint: disable=protected-access
return ctx
def _call_for_each_replica(self, fn, args, kwargs):
strategy = self._container_strategy()
with ops.device(self._device), _OneDeviceReplicaContext(strategy):
return fn(*args, **kwargs)
def _reduce_to(self, reduce_op, value, destinations):
del reduce_op, destinations
return value
def _update(self, var, fn, args, kwargs, group):
# The implementations of _update() and _update_non_slot() are identical
# except _update() passes `var` as the first argument to `fn()`.
return self._update_non_slot(var, fn, (var,) + tuple(args), kwargs, group)
def _update_non_slot(self, colocate_with, fn, args, kwargs, group):
del colocate_with
with ops.device(self._device), distribute_lib.UpdateContext(self._device):
result = fn(*args, **kwargs)
if group:
return result
else:
return nest.map_structure(self._local_results, result)
def read_var(self, replica_local_var):
"""Read the aggregate value of a replica-local variable."""
return array_ops.identity(replica_local_var)
def _local_results(self, value):
return (value,)
def value_container(self, value):
return value
@property
def _num_replicas_in_sync(self):
return 1
@property
def worker_devices(self):
return (self._device,)
@property
def parameter_devices(self):
return (self._device,)
def non_slot_devices(self, var_list):
del var_list
return (self._device,)
@property
def experimental_should_init(self):
return True
@property
def should_checkpoint(self):
return True
@property
def should_save_summary(self):
return True
# TODO(priyag): Delete this once all strategies use global batch size.
@property
def _global_batch_size(self):
"""Global and per-replica batching are equivalent for OneDeviceStrategy."""
return True
@property
def _support_per_replica_values(self):
return False
class _OneDeviceReplicaContext(distribute_lib.ReplicaContext):
"""ReplicaContext for OneDeviceStrategy."""
def __init__(self, strategy):
zero = constant_op.constant(0, dtypes.int32)
distribute_lib.ReplicaContext.__init__(
self, strategy, replica_id_in_sync_group=zero)
@property
def devices(self):
return self._strategy.extended.worker_devices
|
tensorflow-master
|
tensorflow/python/distribute/one_device_strategy.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for numpy_dataset."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.distribute import numpy_dataset
from tensorflow.python.eager import test
from tensorflow.python.framework import test_util
from tensorflow.python.ops import variable_scope
class InitVarFromNumpyTest(test.TestCase):
@test_util.run_in_graph_and_eager_modes
def test_creating_var_with_numpy_arrays(self):
with self.cached_session() as session:
x = np.asarray(np.random.random((64, 3)), dtype=np.float32)
initial = np.zeros_like(x)
var_x = variable_scope.variable(initial)
numpy_dataset.init_var_from_numpy(var_x, x, session)
val = self.evaluate(var_x.value())
# Verify that the numpy value is copied to the variable.
self.assertAllEqual(x, val)
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/numpy_dataset_test.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for saving and loading using tf's saved_model APIs with DS."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.distribute import combinations
from tensorflow.python.distribute import saved_model_test_base as test_base
from tensorflow.python.eager import test
from tensorflow.python.saved_model import saved_model
class SavedModelSaveAndLoadTest(test_base.TestSavedModelBase):
def setUp(self):
self._root_dir = 'saved_model_save_load'
super(SavedModelSaveAndLoadTest, self).setUp()
def _save_model(self, model, saved_dir):
saved_model.save(model, saved_dir)
def _load_and_run_model(self, distribution, saved_dir, predict_dataset,
output_name):
return test_base.load_and_run_with_saved_model_api(distribution, saved_dir,
predict_dataset,
output_name)
@combinations.generate(test_base.simple_models_with_strategies())
def test_save_no_strategy_restore_strategy(self, model_and_input,
distribution):
self.run_test_save_no_strategy_restore_strategy(model_and_input,
distribution)
@combinations.generate(
combinations.times(test_base.simple_models_with_strategies(),
combinations.combine(save_in_scope=[True, False])))
def test_save_strategy_restore_no_strategy(self, model_and_input,
distribution, save_in_scope):
if save_in_scope:
self.skipTest(('Saving model within tf.distribute.Strategy scope is not ',
'supported.'))
self.run_test_save_strategy_restore_no_strategy(model_and_input,
distribution, save_in_scope)
@combinations.generate(
combinations.times(test_base.simple_models_with_strategy_pairs(),
combinations.combine(save_in_scope=[True, False])))
def test_save_strategy_restore_strategy(self, model_and_input,
distribution_for_saving,
distribution_for_restoring,
save_in_scope):
if save_in_scope:
self.skipTest(('Saving model within tf.distribute.Strategy scope is not ',
'supported.'))
self.run_test_save_strategy_restore_strategy(model_and_input,
distribution_for_saving,
distribution_for_restoring,
save_in_scope)
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/saved_model_save_load_test.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Experimental Distribution Strategy library."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
# pylint: disable=unused-import
from tensorflow.python.distribute import central_storage_strategy
from tensorflow.python.distribute import collective_all_reduce_strategy
from tensorflow.python.distribute import parameter_server_strategy
from tensorflow.python.distribute import tpu_strategy
# pylint: enable=unused-import
|
tensorflow-master
|
tensorflow/python/distribute/experimental/__init__.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""A simple functional keras model with one layer."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python import keras
from tensorflow.python.distribute.model_collection import model_collection_base
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.keras.optimizer_v2 import gradient_descent
_BATCH_SIZE = 10
def _get_data_for_simple_models():
x_train = constant_op.constant(np.random.rand(1000, 3), dtype=dtypes.float32)
y_train = constant_op.constant(np.random.rand(1000, 5), dtype=dtypes.float32)
x_predict = constant_op.constant(
np.random.rand(1000, 3), dtype=dtypes.float32)
return x_train, y_train, x_predict
class SimpleFunctionalModel(model_collection_base.ModelAndInput):
"""A simple functinal model and its inputs."""
def get_model(self, **kwargs):
output_name = 'output_layer'
x = keras.layers.Input(shape=(3,), dtype=dtypes.float32)
y = keras.layers.Dense(5, dtype=dtypes.float32, name=output_name)(x)
model = keras.Model(inputs=x, outputs=y)
optimizer = gradient_descent.SGD(learning_rate=0.001)
model.compile(loss='mse', metrics=['mae'], optimizer=optimizer)
return model, output_name
def get_data(self):
return _get_data_for_simple_models()
def get_batch_size(self):
return _BATCH_SIZE
class SimpleSequentialModel(model_collection_base.ModelAndInput):
"""A simple sequential model and its inputs."""
def get_model(self, **kwargs):
output_name = 'output_layer'
model = keras.Sequential()
y = keras.layers.Dense(
5, dtype=dtypes.float32, name=output_name, input_dim=3)
model.add(y)
optimizer = gradient_descent.SGD(learning_rate=0.001)
model.compile(loss='mse', metrics=['mae'], optimizer=optimizer)
return model, output_name
def get_data(self):
return _get_data_for_simple_models()
def get_batch_size(self):
return _BATCH_SIZE
class _SimpleModel(keras.Model):
output_name = 'output_layer'
def __init__(self):
self._dense_layer = keras.layers.Dense(
5, dtype=dtypes.float32, name=self.output_name)
def call(self, inputs):
return self._dense_layer(inputs)
class SimpleSubclassModel(model_collection_base.ModelAndInput):
"""A simple subclass model and its data."""
def get_model(self, **kwargs):
model = _SimpleModel()
optimizer = gradient_descent.SGD(learning_rate=0.001)
model.compile(
loss='mse', metrics=['mae'], cloning=False, optimizer=optimizer)
return model, model.output_name
def get_data(self):
return _get_data_for_simple_models()
def get_batch_size(self):
return _BATCH_SIZE
|
tensorflow-master
|
tensorflow/python/distribute/model_collection/simple_models.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""A base class to provid a model and corresponding input data for testing."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
class ModelAndInput(object):
"""Base class to provide model and its corresponding inputs."""
def get_model(self):
"""Returns a compiled keras model object, together with output name.
Returns:
model: a keras model object
output_name: a string for the name of the output layer
"""
raise NotImplementedError("must be implemented in descendants")
def get_data(self):
"""Returns data for training and predicting.
Returns:
x_train: data used for training
y_train: label used for training
x_predict: data used for predicting
"""
raise NotImplementedError("must be implemented in descendants")
def get_batch_size(self):
"""Returns the batch_size used by the model."""
raise NotImplementedError("must be implemented in descendants")
|
tensorflow-master
|
tensorflow/python/distribute/model_collection/model_collection_base.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Implementation of Cluster Resolvers for Kubernetes."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import ClusterResolver
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import format_master_url
from tensorflow.python.training import server_lib
from tensorflow.python.util.tf_export import tf_export
_KUBERNETES_API_CLIENT_INSTALLED = True
try:
from kubernetes import client as k8sclient # pylint: disable=g-import-not-at-top
from kubernetes import config as k8sconfig # pylint: disable=g-import-not-at-top
except ImportError:
_KUBERNETES_API_CLIENT_INSTALLED = False
@tf_export('distribute.cluster_resolver.KubernetesClusterResolver')
class KubernetesClusterResolver(ClusterResolver):
"""Cluster Resolver for Kubernetes.
This is an implementation of cluster resolvers for Kubernetes. When given the
the Kubernetes namespace and label selector for pods, we will retrieve the
pod IP addresses of all running pods matching the selector, and return a
ClusterSpec based on that information.
"""
def __init__(self,
job_to_label_mapping=None,
tf_server_port=8470,
rpc_layer='grpc',
override_client=None):
"""Initializes a new KubernetesClusterResolver.
This initializes a new Kubernetes Cluster Resolver. The Cluster Resolver
will attempt to talk to the Kubernetes master to retrieve all the instances
of pods matching a label selector.
Args:
job_to_label_mapping: A mapping of TensorFlow jobs to label selectors.
This allows users to specify many TensorFlow jobs in one Cluster
Resolver, and each job can have pods belong with different label
selectors. For example, a sample mapping might be
```
{'worker': ['job-name=worker-cluster-a', 'job-name=worker-cluster-b'],
'ps': ['job-name=ps-1', 'job-name=ps-2']}
```
tf_server_port: The port the TensorFlow server is listening on.
rpc_layer: (Optional) The RPC layer TensorFlow should use to communicate
between tasks in Kubernetes. Defaults to 'grpc'.
override_client: The Kubernetes client (usually automatically retrieved
using `from kubernetes import client as k8sclient`). If you pass this
in, you are responsible for setting Kubernetes credentials manually.
Raises:
ImportError: If the Kubernetes Python client is not installed and no
`override_client` is passed in.
RuntimeError: If autoresolve_task is not a boolean or a callable.
"""
if _KUBERNETES_API_CLIENT_INSTALLED:
k8sconfig.load_kube_config()
if not job_to_label_mapping:
job_to_label_mapping = {'worker': ['job-name=tensorflow']}
if not override_client and not _KUBERNETES_API_CLIENT_INSTALLED:
raise ImportError('The Kubernetes Python client must be installed before'
'using the Kubernetes Cluster Resolver. To install the'
'Kubernetes Python client, run `pip install '
'kubernetes` on your command line.')
self._job_to_label_mapping = job_to_label_mapping
self._tf_server_port = tf_server_port
self._override_client = override_client
self.task_type = None
self.task_id = None
self.rpc_layer = rpc_layer
def master(self, task_type=None, task_id=None, rpc_layer=None):
"""Returns the master address to use when creating a session.
You must have set the task_type and task_id object properties before
calling this function, or pass in the `task_type` and `task_id`
parameters when using this function. If you do both, the function parameters
will override the object properties.
Args:
task_type: (Optional) The type of the TensorFlow task of the master.
task_id: (Optional) The index of the TensorFlow task of the master.
rpc_layer: (Optional) The RPC protocol for the given cluster.
Returns:
The name or URL of the session master.
"""
task_type = task_type if task_type is not None else self.task_type
task_id = task_id if task_id is not None else self.task_id
if task_type is not None and task_id is not None:
return format_master_url(
self.cluster_spec().task_address(task_type, task_id),
rpc_layer or self.rpc_layer)
return ''
def cluster_spec(self):
"""Returns a ClusterSpec object based on the latest info from Kubernetes.
We retrieve the information from the Kubernetes master every time this
method is called.
Returns:
A ClusterSpec containing host information returned from Kubernetes.
Raises:
RuntimeError: If any of the pods returned by the master is not in the
`Running` phase.
"""
if not self._override_client:
k8sconfig.load_kube_config()
client = self._override_client or k8sclient.CoreV1Api()
cluster_map = {}
for tf_job in self._job_to_label_mapping:
all_pods = []
for selector in self._job_to_label_mapping[tf_job]:
ret = client.list_pod_for_all_namespaces(label_selector=selector)
selected_pods = []
# Sort the list by the name to make sure it doesn't change call to call.
for pod in sorted(ret.items, key=lambda x: x.metadata.name):
if pod.status.phase == 'Running':
selected_pods.append(
'%s:%s' % (pod.status.host_ip, self._tf_server_port))
else:
raise RuntimeError('Pod "%s" is not running; phase: "%s"' %
(pod.metadata.name, pod.status.phase))
all_pods.extend(selected_pods)
cluster_map[tf_job] = all_pods
return server_lib.ClusterSpec(cluster_map)
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/kubernetes_cluster_resolver.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the 'License');
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an 'AS IS' BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for TFCONFIGClusterResolver."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
from tensorflow.python import eager
from tensorflow.python.client import session
from tensorflow.python.distribute.cluster_resolver import TFConfigClusterResolver
from tensorflow.python.framework import test_util
from tensorflow.python.platform import test
from tensorflow.python.training import server_lib
mock = test.mock
@test_util.run_all_in_graph_and_eager_modes
class TFConfigClusterResolverTest(test.TestCase):
def _verifyClusterSpecEquality(self, cluster_spec, expected_proto):
self.assertProtoEquals(expected_proto, cluster_spec.as_cluster_def())
self.assertProtoEquals(
expected_proto, server_lib.ClusterSpec(cluster_spec).as_cluster_def())
self.assertProtoEquals(
expected_proto,
server_lib.ClusterSpec(cluster_spec.as_cluster_def()).as_cluster_def())
self.assertProtoEquals(
expected_proto,
server_lib.ClusterSpec(cluster_spec.as_dict()).as_cluster_def())
def testNormalClusterSpecRead(self):
os.environ['TF_CONFIG'] = """
{
"cluster": {
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
},
"task": {
"type": "ps",
"index": 0
}
}
"""
cluster_resolver = TFConfigClusterResolver()
expected_proto = """
job { name: 'ps' tasks { key: 0 value: 'ps0:2222' }
tasks { key: 1 value: 'ps1:2222' } }
job { name: 'worker' tasks { key: 0 value: 'worker0:2222' }
tasks { key: 1 value: 'worker1:2222' }
tasks { key: 2 value: 'worker2:2222' } }
"""
actual_cluster_spec = cluster_resolver.cluster_spec()
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
def testAutomaticMasterRead(self):
os.environ['TF_CONFIG'] = """
{
"cluster": {
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
},
"task": {
"type": "ps",
"index": 0
}
}
"""
cluster_resolver = TFConfigClusterResolver()
self.assertEqual('ps0:2222', cluster_resolver.master())
def testSpecifiedTaskTypeAndIndexMasterRead(self):
os.environ['TF_CONFIG'] = """
{
"cluster": {
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
},
"task": {
"type": "ps",
"index": 0
}
}
"""
cluster_resolver = TFConfigClusterResolver()
self.assertEqual('worker1:2222', cluster_resolver.master('worker', 1))
def testSessionMasterRead(self):
os.environ['TF_CONFIG'] = """
{
"cluster": {
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
},
"session_master": "sessionmaster:2222",
"task": {
"type": "ps",
"index": 0
}
}
"""
cluster_resolver = TFConfigClusterResolver()
self.assertEqual('sessionmaster:2222', cluster_resolver.master())
def testRpcLayerRead(self):
os.environ['TF_CONFIG'] = """
{
"cluster": {
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
},
"rpc_layer": "grpc",
"task": {
"type": "ps",
"index": 0
}
}
"""
cluster_resolver = TFConfigClusterResolver()
self.assertEqual('grpc://ps0:2222', cluster_resolver.master())
def testTaskTypeIndexRpcRead(self):
os.environ['TF_CONFIG'] = """
{
"cluster": {
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
},
"rpc_layer": "grpc",
"task": {
"type": "ps",
"index": 0
}
}
"""
cluster_resolver = TFConfigClusterResolver()
self.assertEqual('ps', cluster_resolver.task_type)
self.assertEqual(0, cluster_resolver.task_id)
self.assertEqual('grpc', cluster_resolver.rpc_layer)
def testParameterOverrides(self):
os.environ['TF_CONFIG'] = """
{
"cluster": {
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
},
"rpc_layer": "grpc",
"task": {
"type": "ps",
"index": 1
}
}
"""
cluster_resolver = TFConfigClusterResolver(task_type='ps', task_id=0)
self.assertEqual('grpc://ps0:2222', cluster_resolver.master())
self.assertEqual('ps', cluster_resolver.task_type)
self.assertEqual(0, cluster_resolver.task_id)
cluster_resolver.task_type = 'worker'
cluster_resolver.task_id = 1
cluster_resolver.rpc_layer = 'test'
self.assertEqual('test://worker1:2222', cluster_resolver.master())
self.assertEqual('worker', cluster_resolver.task_type)
self.assertEqual(1, cluster_resolver.task_id)
self.assertEqual('test', cluster_resolver.rpc_layer)
def testTaskTypeCastToString(self):
os.environ['TF_CONFIG'] = """
{
"cluster": {
"123456": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
},
"rpc_layer": "grpc",
"task": {
"type": 123456,
"index": 0
}
}
"""
cluster_resolver = TFConfigClusterResolver()
self.assertEqual('123456', cluster_resolver.task_type)
def testTaskIndexCastToInteger(self):
os.environ['TF_CONFIG'] = """
{
"cluster": {
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
},
"rpc_layer": "grpc",
"task": {
"type": "ps",
"index": "1"
}
}
"""
cluster_resolver = TFConfigClusterResolver()
self.assertEqual(1, cluster_resolver.task_id)
def testZeroItemsInClusterSpecMasterRead(self):
os.environ['TF_CONFIG'] = """
{}
"""
cluster_resolver = TFConfigClusterResolver()
self.assertEqual('', cluster_resolver.master())
def testOneItemInClusterSpecMasterRead(self):
os.environ['TF_CONFIG'] = """
{
"cluster": {
"worker": ["worker0:2222"]
}
}
"""
cluster_resolver = TFConfigClusterResolver()
self.assertEqual('', cluster_resolver.master())
@mock.patch.object(eager.context, 'list_devices')
@mock.patch.object(session.BaseSession, 'list_devices')
def testNumAcceleratorsFilterTasksByEnvVar(self, mock_list_devices,
mock_eager_list_devices):
os.environ['TF_CONFIG'] = """
{
"cluster": {
"worker1": ["w10:2222"],
"worker2": ["w21:2222", "w22:2222", "w23:2222", "w24:2222"]
},
"rpc_layer": "grpc",
"task": {
"type": "worker1",
"index": "0"
}
}
"""
device_names = [
'/job:worker1/task:0/device:TPU:0',
'/job:worker1/task:0/device:TPU:1',
'/job:worker1/task:0/device:GPU:0',
'/job:worker1/task:0/device:GPU:1',
'/job:worker2/task:1/device:TPU:2',
'/job:worker2/task:2/device:TPU:3',
'/job:worker2/task:3/device:GPU:2',
'/job:worker2/task:4/device:GPU:3',
]
device_list = [
session._DeviceAttributes(name, name[27:30], 1024, 0)
for name in device_names
]
mock_eager_list_devices.return_value = device_names
mock_list_devices.return_value = device_list
resolver = TFConfigClusterResolver()
# By default we read from TF_CONFIG
self.assertEqual(resolver.num_accelerators(), {'TPU': 2, 'GPU': 2})
# Override still works when we want it to
self.assertEqual(resolver.num_accelerators(task_type='worker2', task_id=3),
{'GPU': 1})
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/tfconfig_cluster_resolver_test.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for GCEClusterResolver."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.distribute.cluster_resolver import GCEClusterResolver
from tensorflow.python.distribute.cluster_resolver import UnionClusterResolver
from tensorflow.python.platform import test
from tensorflow.python.training import server_lib
mock = test.mock
class GCEClusterResolverTest(test.TestCase):
def _verifyClusterSpecEquality(self, cluster_spec, expected_proto):
self.assertProtoEquals(expected_proto, cluster_spec.as_cluster_def())
self.assertProtoEquals(
expected_proto, server_lib.ClusterSpec(cluster_spec).as_cluster_def())
self.assertProtoEquals(
expected_proto,
server_lib.ClusterSpec(cluster_spec.as_cluster_def()).as_cluster_def())
self.assertProtoEquals(
expected_proto,
server_lib.ClusterSpec(cluster_spec.as_dict()).as_cluster_def())
def standard_mock_instance_groups(self, instance_map=None):
if instance_map is None:
instance_map = [
{'instance': 'https://gce.example.com/res/gce-instance-1'}
]
mock_instance_group_request = mock.MagicMock()
mock_instance_group_request.execute.return_value = {
'items': instance_map
}
service_attrs = {
'listInstances.return_value': mock_instance_group_request,
'listInstances_next.return_value': None,
}
mock_instance_groups = mock.Mock(**service_attrs)
return mock_instance_groups
def standard_mock_instances(self, instance_to_ip_map=None):
if instance_to_ip_map is None:
instance_to_ip_map = {
'gce-instance-1': '10.123.45.67'
}
mock_get_request = mock.MagicMock()
mock_get_request.execute.return_value = {
'networkInterfaces': [
{'networkIP': '10.123.45.67'}
]
}
def get_side_effect(project, zone, instance):
del project, zone # Unused
if instance in instance_to_ip_map:
mock_get_request = mock.MagicMock()
mock_get_request.execute.return_value = {
'networkInterfaces': [
{'networkIP': instance_to_ip_map[instance]}
]
}
return mock_get_request
else:
raise RuntimeError('Instance %s not found!' % instance)
service_attrs = {
'get.side_effect': get_side_effect,
}
mock_instances = mock.MagicMock(**service_attrs)
return mock_instances
def standard_mock_service_client(
self,
mock_instance_groups=None,
mock_instances=None):
if mock_instance_groups is None:
mock_instance_groups = self.standard_mock_instance_groups()
if mock_instances is None:
mock_instances = self.standard_mock_instances()
mock_client = mock.MagicMock()
mock_client.instanceGroups.return_value = mock_instance_groups
mock_client.instances.return_value = mock_instances
return mock_client
def gen_standard_mock_service_client(self, instances=None):
name_to_ip = {}
instance_list = []
for instance in instances:
name_to_ip[instance['name']] = instance['ip']
instance_list.append({
'instance': 'https://gce.example.com/gce/res/' + instance['name']
})
mock_instance = self.standard_mock_instances(name_to_ip)
mock_instance_group = self.standard_mock_instance_groups(instance_list)
return self.standard_mock_service_client(mock_instance_group, mock_instance)
def testSimpleSuccessfulRetrieval(self):
gce_cluster_resolver = GCEClusterResolver(
project='test-project',
zone='us-east1-d',
instance_group='test-instance-group',
port=8470,
credentials=None,
service=self.standard_mock_service_client())
actual_cluster_spec = gce_cluster_resolver.cluster_spec()
expected_proto = """
job { name: 'worker' tasks { key: 0 value: '10.123.45.67:8470' } }
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
def testMasterRetrieval(self):
gce_cluster_resolver = GCEClusterResolver(
project='test-project',
zone='us-east1-d',
instance_group='test-instance-group',
task_id=0,
port=8470,
credentials=None,
service=self.standard_mock_service_client())
self.assertEqual(gce_cluster_resolver.master(), 'grpc://10.123.45.67:8470')
def testMasterRetrievalWithCustomTasks(self):
name_to_ip = [
{'name': 'instance1', 'ip': '10.1.2.3'},
{'name': 'instance2', 'ip': '10.2.3.4'},
{'name': 'instance3', 'ip': '10.3.4.5'},
]
gce_cluster_resolver = GCEClusterResolver(
project='test-project',
zone='us-east1-d',
instance_group='test-instance-group',
port=8470,
credentials=None,
service=self.gen_standard_mock_service_client(name_to_ip))
self.assertEqual(
gce_cluster_resolver.master('worker', 2, 'test'),
'test://10.3.4.5:8470')
def testOverrideParameters(self):
name_to_ip = [
{'name': 'instance1', 'ip': '10.1.2.3'},
{'name': 'instance2', 'ip': '10.2.3.4'},
{'name': 'instance3', 'ip': '10.3.4.5'},
]
gce_cluster_resolver = GCEClusterResolver(
project='test-project',
zone='us-east1-d',
instance_group='test-instance-group',
task_type='testworker',
port=8470,
credentials=None,
service=self.gen_standard_mock_service_client(name_to_ip))
gce_cluster_resolver.task_id = 1
gce_cluster_resolver.rpc_layer = 'test'
self.assertEqual(gce_cluster_resolver.task_type, 'testworker')
self.assertEqual(gce_cluster_resolver.task_id, 1)
self.assertEqual(gce_cluster_resolver.rpc_layer, 'test')
self.assertEqual(gce_cluster_resolver.master(), 'test://10.2.3.4:8470')
def testOverrideParametersWithZeroOrEmpty(self):
name_to_ip = [
{'name': 'instance1', 'ip': '10.1.2.3'},
{'name': 'instance2', 'ip': '10.2.3.4'},
{'name': 'instance3', 'ip': '10.3.4.5'},
]
gce_cluster_resolver = GCEClusterResolver(
project='test-project',
zone='us-east1-d',
instance_group='test-instance-group',
task_type='',
task_id=1,
port=8470,
credentials=None,
service=self.gen_standard_mock_service_client(name_to_ip))
self.assertEqual(gce_cluster_resolver.master(
task_type='', task_id=0), 'grpc://10.1.2.3:8470')
def testCustomJobNameAndPortRetrieval(self):
gce_cluster_resolver = GCEClusterResolver(
project='test-project',
zone='us-east1-d',
instance_group='test-instance-group',
task_type='custom',
port=2222,
credentials=None,
service=self.standard_mock_service_client())
actual_cluster_spec = gce_cluster_resolver.cluster_spec()
expected_proto = """
job { name: 'custom' tasks { key: 0 value: '10.123.45.67:2222' } }
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
def testMultipleInstancesRetrieval(self):
name_to_ip = [
{'name': 'instance1', 'ip': '10.1.2.3'},
{'name': 'instance2', 'ip': '10.2.3.4'},
{'name': 'instance3', 'ip': '10.3.4.5'},
]
gce_cluster_resolver = GCEClusterResolver(
project='test-project',
zone='us-east1-d',
instance_group='test-instance-group',
port=8470,
credentials=None,
service=self.gen_standard_mock_service_client(name_to_ip))
actual_cluster_spec = gce_cluster_resolver.cluster_spec()
expected_proto = """
job { name: 'worker' tasks { key: 0 value: '10.1.2.3:8470' }
tasks { key: 1 value: '10.2.3.4:8470' }
tasks { key: 2 value: '10.3.4.5:8470' } }
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
def testUnionMultipleInstanceRetrieval(self):
worker1_name_to_ip = [
{'name': 'instance1', 'ip': '10.1.2.3'},
{'name': 'instance2', 'ip': '10.2.3.4'},
{'name': 'instance3', 'ip': '10.3.4.5'},
]
worker2_name_to_ip = [
{'name': 'instance4', 'ip': '10.4.5.6'},
{'name': 'instance5', 'ip': '10.5.6.7'},
{'name': 'instance6', 'ip': '10.6.7.8'},
]
ps_name_to_ip = [
{'name': 'ps1', 'ip': '10.100.1.2'},
{'name': 'ps2', 'ip': '10.100.2.3'},
]
worker1_gce_cluster_resolver = GCEClusterResolver(
project='test-project',
zone='us-east1-d',
instance_group='test-instance-group',
task_type='worker',
port=8470,
credentials=None,
service=self.gen_standard_mock_service_client(worker1_name_to_ip))
worker2_gce_cluster_resolver = GCEClusterResolver(
project='test-project',
zone='us-east1-d',
instance_group='test-instance-group',
task_type='worker',
port=8470,
credentials=None,
service=self.gen_standard_mock_service_client(worker2_name_to_ip))
ps_gce_cluster_resolver = GCEClusterResolver(
project='test-project',
zone='us-east1-d',
instance_group='test-instance-group',
task_type='ps',
port=2222,
credentials=None,
service=self.gen_standard_mock_service_client(ps_name_to_ip))
union_cluster_resolver = UnionClusterResolver(worker1_gce_cluster_resolver,
worker2_gce_cluster_resolver,
ps_gce_cluster_resolver)
actual_cluster_spec = union_cluster_resolver.cluster_spec()
expected_proto = """
job { name: 'ps' tasks { key: 0 value: '10.100.1.2:2222' }
tasks { key: 1 value: '10.100.2.3:2222' } }
job { name: 'worker' tasks { key: 0 value: '10.1.2.3:8470' }
tasks { key: 1 value: '10.2.3.4:8470' }
tasks { key: 2 value: '10.3.4.5:8470' }
tasks { key: 3 value: '10.4.5.6:8470' }
tasks { key: 4 value: '10.5.6.7:8470' }
tasks { key: 5 value: '10.6.7.8:8470' } }
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/gce_cluster_resolver_test.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Implementation of Cluster Resolvers for Slurm workload manager."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import os
import subprocess
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import ClusterResolver
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import format_master_url
from tensorflow.python.training.server_lib import ClusterSpec
from tensorflow.python.util.tf_export import tf_export
@tf_export('distribute.cluster_resolver.SlurmClusterResolver')
class SlurmClusterResolver(ClusterResolver):
"""Cluster Resolver for system with Slurm workload manager.
This is an implementation of cluster resolvers for Slurm clusters. This allows
the specification of jobs and task counts, number of tasks per node, number of
GPUs on each node and number of GPUs for each task, It retrieves system
attributes by Slurm environment variables, resolves allocated computing node
names, construct a cluster and return a Cluster Resolver object which an be
use for distributed TensorFlow.
"""
def _resolve_hostnames(self):
"""Resolve host names of nodes allocated in current jobs.
Returns:
A list of node names as strings.
"""
hostlist = (subprocess.check_output(['scontrol', 'show', 'hostname']).
decode('utf-8').strip().split('\n'))
return hostlist
def __init__(self,
jobs,
port_base=8888,
gpus_per_node=1,
gpus_per_task=1,
tasks_per_node=None,
auto_set_gpu=True,
rpc_layer='grpc'):
"""Creates a new SlurmClusterResolver object.
This takes in parameters and creates a SlurmClusterResolver object. It uses
those parameters to check which nodes will processes reside and resolves
their hostnames. With the number of the GPUs on each node and number of GPUs
for each task it offsets the port number for each processes and allocate
GPUs to tasks by setting environment variables. The resolver currently
supports homogeneous tasks and default Slurm process allocation.
Args:
jobs: Dictionary with job names as key and number of tasks in the job as
value
port_base: The first port number to start with for processes on a node.
gpus_per_node: Number of GPUs available on each node.
gpus_per_task: Number of GPUs to be used for each task.
tasks_per_node: Number of tasks to run on each node, if not set defaults
to Slurm's output environment variable SLURM_NTASKS_PER_NODE.
auto_set_gpu: Set the visible CUDA devices automatically while resolving
the cluster by setting CUDA_VISIBLE_DEVICES environment variable.
Defaults to True.
rpc_layer: (Optional) The protocol TensorFlow uses to communicate between
nodes. Defaults to 'grpc'.
Returns:
A ClusterResolver object which can be used with distributed TensorFlow.
Raises:
RuntimeError: If requested more GPUs per node then available or requested
more tasks then assigned tasks.
"""
# check if launched by mpirun
if 'OMPI_COMM_WORLD_RANK' in os.environ:
self._rank = int(os.environ['OMPI_COMM_WORLD_RANK'])
num_tasks = int(os.environ['OMPI_COMM_WORLD_SIZE'])
else:
self._rank = int(os.environ['SLURM_PROCID'])
num_tasks = int(os.environ['SLURM_NTASKS'])
self._jobs = collections.OrderedDict(sorted(jobs.items()))
self._port_base = port_base
# user specification overrides SLURM specification
if tasks_per_node is not None:
self._tasks_per_node = tasks_per_node
elif tasks_per_node is None and 'SLURM_NTASKS_PER_NODE' in os.environ:
self._tasks_per_node = int(os.environ['SLURM_NTASKS_PER_NODE'])
else:
raise RuntimeError('Neither `tasks_per_node` or '
'SLURM_NTASKS_PER_NODE is set.')
self._gpus_per_node = gpus_per_node
self._gpus_per_task = gpus_per_task
self._auto_set_gpu = auto_set_gpu
self.task_type = None
self.task_id = None
self.rpc_layer = rpc_layer
self._gpu_allocation = []
self._cluster_allocation = {}
if self._tasks_per_node * self._gpus_per_task > self._gpus_per_node:
raise RuntimeError('Requested more GPUs per node then available.')
if sum(self._jobs.values()) != num_tasks:
raise RuntimeError('Requested more tasks then assigned tasks.')
def cluster_spec(self):
"""Returns a ClusterSpec object based on the latest instance group info.
This returns a ClusterSpec object for use based on information from the
specified initialization parameters and Slurm environment variables. The
cluster specification is resolved each time this function is called. The
resolver extract hostnames of nodes by scontrol and pack tasks in that
order until a node a has number of tasks that is equal to specification.
GPUs on nodes are allocated to tasks by specification through setting
CUDA_VISIBLE_DEVICES environment variable.
Returns:
A ClusterSpec containing host information retrieved from Slurm's
environment variables.
"""
hostlist = self._resolve_hostnames()
task_list = []
self._gpu_allocation = []
self._cluster_allocation = {}
for host in hostlist:
for port_offset, gpu_offset in zip(
range(self._tasks_per_node),
range(0, self._gpus_per_node, self._gpus_per_task)):
host_addr = '%s:%d' % (host, self._port_base + port_offset)
task_list.append(host_addr)
gpu_id_list = []
for gpu_id in range(gpu_offset, gpu_offset + self._gpus_per_task):
gpu_id_list.append(str(gpu_id))
self._gpu_allocation.append(','.join(gpu_id_list))
cluster_rank_offset_start = 0
cluster_rank_offset_end = 0
for task_type, num_tasks in self._jobs.items():
cluster_rank_offset_end = cluster_rank_offset_start + num_tasks
self._cluster_allocation[task_type] = (
task_list[cluster_rank_offset_start:cluster_rank_offset_end])
if cluster_rank_offset_start <= self._rank < cluster_rank_offset_end:
self.task_type = task_type
self.task_id = self._rank - cluster_rank_offset_start
cluster_rank_offset_start = cluster_rank_offset_end
if self._auto_set_gpu is True:
os.environ['CUDA_VISIBLE_DEVICES'] = self._gpu_allocation[self._rank]
return ClusterSpec(self._cluster_allocation)
def get_task_info(self):
"""Returns job name and task_id for the process which calls this.
This returns the job name and task index for the process which calls this
function according to its rank and cluster specification. The job name and
task index are set after a cluster is constructed by cluster_spec otherwise
defaults to None.
Returns:
A string specifying job name the process belongs to and an integner
specifying the task index the process belongs to in that job.
"""
return self.task_type, self.task_id
def master(self, task_type=None, task_id=None, rpc_layer=None):
"""Returns the master string for connecting to a TensorFlow master.
Args:
task_type: (Optional) Overrides the default auto-selected task type.
task_id: (Optional) Overrides the default auto-slected task index.
rpc_layer: (Optional) Overrides the default RPC protocol TensorFlow uses
to communicate across nodes.
Returns:
A connection string for connecting to a TensorFlow master.
"""
task_type = task_type if task_type is not None else self.task_type
task_id = task_id if task_id is not None else self.task_id
if task_type is not None and task_id is not None:
return format_master_url(
self.cluster_spec().task_address(task_type, task_id),
rpc_layer or self.rpc_layer)
return ''
def num_accelerators(self,
task_type=None,
task_id=None,
config_proto=None):
# Unused, since this is set in __init__ manually.
del task_type, task_id, config_proto
return {'GPU': self._gpus_per_node}
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/slurm_cluster_resolver.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Implementation of Cluster Resolvers for GCE Instance Groups."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import ClusterResolver
from tensorflow.python.training.server_lib import ClusterSpec
from tensorflow.python.util.tf_export import tf_export
_GOOGLE_API_CLIENT_INSTALLED = True
try:
from googleapiclient import discovery # pylint: disable=g-import-not-at-top
from oauth2client.client import GoogleCredentials # pylint: disable=g-import-not-at-top
except ImportError:
_GOOGLE_API_CLIENT_INSTALLED = False
@tf_export('distribute.cluster_resolver.GCEClusterResolver')
class GCEClusterResolver(ClusterResolver):
"""Cluster Resolver for Google Compute Engine.
This is an implementation of cluster resolvers for the Google Compute Engine
instance group platform. By specifying a project, zone, and instance group,
this will retrieve the IP address of all the instances within the instance
group and return a Cluster Resolver object suitable for use for distributed
TensorFlow.
"""
def __init__(self,
project,
zone,
instance_group,
port,
task_type='worker',
task_id=0,
rpc_layer='grpc',
credentials='default',
service=None):
"""Creates a new GCEClusterResolver object.
This takes in a few parameters and creates a GCEClusterResolver project. It
will then use these parameters to query the GCE API for the IP addresses of
each instance in the instance group.
Args:
project: Name of the GCE project.
zone: Zone of the GCE instance group.
instance_group: Name of the GCE instance group.
port: Port of the listening TensorFlow server (default: 8470)
task_type: Name of the TensorFlow job this GCE instance group of VM
instances belong to.
task_id: The task index for this particular VM, within the GCE
instance group. In particular, every single instance should be assigned
a unique ordinal index within an instance group manually so that they
can be distinguished from each other.
rpc_layer: The RPC layer TensorFlow should use to communicate across
instances.
credentials: GCE Credentials. If nothing is specified, this defaults to
GoogleCredentials.get_application_default().
service: The GCE API object returned by the googleapiclient.discovery
function. (Default: discovery.build('compute', 'v1')). If you specify a
custom service object, then the credentials parameter will be ignored.
Raises:
ImportError: If the googleapiclient is not installed.
"""
self._project = project
self._zone = zone
self._instance_group = instance_group
self._task_type = task_type
self._task_id = task_id
self._rpc_layer = rpc_layer
self._port = port
self._credentials = credentials
if credentials == 'default':
if _GOOGLE_API_CLIENT_INSTALLED:
self._credentials = GoogleCredentials.get_application_default()
if service is None:
if not _GOOGLE_API_CLIENT_INSTALLED:
raise ImportError('googleapiclient must be installed before using the '
'GCE cluster resolver')
self._service = discovery.build(
'compute', 'v1',
credentials=self._credentials)
else:
self._service = service
def cluster_spec(self):
"""Returns a ClusterSpec object based on the latest instance group info.
This returns a ClusterSpec object for use based on information from the
specified instance group. We will retrieve the information from the GCE APIs
every time this method is called.
Returns:
A ClusterSpec containing host information retrieved from GCE.
"""
request_body = {'instanceState': 'RUNNING'}
request = self._service.instanceGroups().listInstances(
project=self._project,
zone=self._zone,
instanceGroups=self._instance_group,
body=request_body,
orderBy='name')
worker_list = []
while request is not None:
response = request.execute()
items = response['items']
for instance in items:
instance_name = instance['instance'].split('/')[-1]
instance_request = self._service.instances().get(
project=self._project,
zone=self._zone,
instance=instance_name)
if instance_request is not None:
instance_details = instance_request.execute()
ip_address = instance_details['networkInterfaces'][0]['networkIP']
instance_url = '%s:%s' % (ip_address, self._port)
worker_list.append(instance_url)
request = self._service.instanceGroups().listInstances_next(
previous_request=request,
previous_response=response)
worker_list.sort()
return ClusterSpec({self._task_type: worker_list})
def master(self, task_type=None, task_id=None, rpc_layer=None):
task_type = task_type if task_type is not None else self._task_type
task_id = task_id if task_id is not None else self._task_id
if task_type is not None and task_id is not None:
master = self.cluster_spec().task_address(task_type, task_id)
if rpc_layer or self._rpc_layer:
return '%s://%s' % (rpc_layer or self._rpc_layer, master)
else:
return master
return ''
@property
def task_type(self):
return self._task_type
@property
def task_id(self):
return self._task_id
@task_type.setter
def task_type(self, task_type):
raise RuntimeError(
'You cannot reset the task_type of the GCEClusterResolver after it has '
'been created.')
@task_id.setter
def task_id(self, task_id):
self._task_id = task_id
@property
def rpc_layer(self):
return self._rpc_layer
@rpc_layer.setter
def rpc_layer(self, rpc_layer):
self._rpc_layer = rpc_layer
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/gce_cluster_resolver.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Library Imports for Cluster Resolvers."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import ClusterResolver
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import SimpleClusterResolver
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import UnionClusterResolver
from tensorflow.python.distribute.cluster_resolver.gce_cluster_resolver import GCEClusterResolver
from tensorflow.python.distribute.cluster_resolver.kubernetes_cluster_resolver import KubernetesClusterResolver
from tensorflow.python.distribute.cluster_resolver.slurm_cluster_resolver import SlurmClusterResolver
from tensorflow.python.distribute.cluster_resolver.tfconfig_cluster_resolver import TFConfigClusterResolver
from tensorflow.python.distribute.cluster_resolver.tpu_cluster_resolver import TPUClusterResolver
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/__init__.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for K8sClusterResolver."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.distribute.cluster_resolver import KubernetesClusterResolver
from tensorflow.python.platform import test
from tensorflow.python.training import server_lib
mock = test.mock
def _mock_kubernetes_client(ret):
mock_client = mock.MagicMock()
mock_client.list_pod_for_all_namespaces.side_effect = (
lambda *args, **kwargs: ret[kwargs['label_selector']])
return mock_client
def _get_mock_pod_item(name, phase, host_ip):
mock_status = mock.Mock()
mock_status.configure_mock(phase=phase, host_ip=host_ip)
mock_metadata = mock.Mock()
mock_metadata.configure_mock(name=name)
mock_item = mock.Mock()
mock_item.configure_mock(status=mock_status, metadata=mock_metadata)
return mock_item
def _create_pod_list(*args):
return mock.MagicMock(items=[_get_mock_pod_item(*x) for x in args])
class KubernetesClusterResolverTest(test.TestCase):
def _verifyClusterSpecEquality(self, cluster_spec, expected_proto):
"""Verifies that the ClusterSpec generates the correct proto.
We are testing this four different ways to ensure that the ClusterSpec
returned by the TPUClusterResolver behaves identically to a normal
ClusterSpec when passed into the generic ClusterSpec libraries.
Args:
cluster_spec: ClusterSpec returned by the TPUClusterResolver
expected_proto: Expected protobuf
"""
self.assertProtoEquals(expected_proto, cluster_spec.as_cluster_def())
self.assertProtoEquals(
expected_proto,
server_lib.ClusterSpec(cluster_spec).as_cluster_def())
self.assertProtoEquals(expected_proto,
server_lib.ClusterSpec(
cluster_spec.as_cluster_def()).as_cluster_def())
self.assertProtoEquals(expected_proto,
server_lib.ClusterSpec(
cluster_spec.as_dict()).as_cluster_def())
def testSingleItemSuccessfulRetrieval(self):
ret = _create_pod_list(('tensorflow-abc123', 'Running', '10.1.2.3'),)
cluster_resolver = KubernetesClusterResolver(
override_client=_mock_kubernetes_client(
{'job-name=tensorflow': ret}))
actual_cluster_spec = cluster_resolver.cluster_spec()
expected_proto = """
job {
name: 'worker'
tasks { key: 0 value: '10.1.2.3:8470' }
}
"""
self._verifyClusterSpecEquality(actual_cluster_spec, str(expected_proto))
def testSuccessfulRetrievalWithSort(self):
ret = _create_pod_list(
('tensorflow-abc123', 'Running', '10.1.2.3'),
('tensorflow-def456', 'Running', '10.1.2.4'),
('tensorflow-999999', 'Running', '10.1.2.5'))
cluster_resolver = KubernetesClusterResolver(
override_client=_mock_kubernetes_client(
{'job-name=tensorflow': ret}))
actual_cluster_spec = cluster_resolver.cluster_spec()
expected_proto = """
job {
name: 'worker'
tasks { key: 0 value: '10.1.2.5:8470' }
tasks { key: 1 value: '10.1.2.3:8470' }
tasks { key: 2 value: '10.1.2.4:8470' }
}
"""
self._verifyClusterSpecEquality(actual_cluster_spec, str(expected_proto))
def testGetMasterWithOverrideParameters(self):
ret = _create_pod_list(
('worker-0', 'Running', '10.1.2.3'),
('worker-1', 'Running', '10.1.2.4'),
('worker-2', 'Running', '10.1.2.5'))
cluster_resolver = KubernetesClusterResolver(
override_client=_mock_kubernetes_client(
{'job-name=tensorflow': ret}))
cluster_resolver.task_type = 'worker'
cluster_resolver.task_id = 0
self.assertEqual(cluster_resolver.task_type, 'worker')
self.assertEqual(cluster_resolver.task_id, 0)
self.assertEqual(cluster_resolver.master(), 'grpc://10.1.2.3:8470')
self.assertEqual(cluster_resolver.master('worker', 2),
'grpc://10.1.2.5:8470')
def testNonRunningPod(self):
ret = _create_pod_list(('tensorflow-abc123', 'Failed', '10.1.2.3'),)
cluster_resolver = KubernetesClusterResolver(
override_client=_mock_kubernetes_client(
{'job-name=tensorflow': ret}))
error_msg = 'Pod "tensorflow-abc123" is not running; phase: "Failed"'
with self.assertRaisesRegexp(RuntimeError, error_msg):
cluster_resolver.cluster_spec()
def testMultiplePodSelectorsAndWorkers(self):
worker1 = _create_pod_list(
('tensorflow-abc123', 'Running', '10.1.2.3'),
('tensorflow-def456', 'Running', '10.1.2.4'),
('tensorflow-999999', 'Running', '10.1.2.5'))
worker2 = _create_pod_list(
('tensorflow-abc124', 'Running', '10.1.2.6'),
('tensorflow-def457', 'Running', '10.1.2.7'),
('tensorflow-999990', 'Running', '10.1.2.8'))
ps = _create_pod_list(
('tensorflow-ps-1', 'Running', '10.1.2.1'),
('tensorflow-ps-2', 'Running', '10.1.2.2'))
cluster_resolver = KubernetesClusterResolver(
job_to_label_mapping={
'worker': ['job-name=worker1', 'job-name=worker2'],
'ps': ['job-name=ps']
},
override_client=_mock_kubernetes_client({
'job-name=worker1': worker1,
'job-name=worker2': worker2,
'job-name=ps': ps
}))
actual_cluster_spec = cluster_resolver.cluster_spec()
expected_proto = """
job {
name: 'ps'
tasks { key: 0 value: '10.1.2.1:8470' }
tasks { key: 1 value: '10.1.2.2:8470' }
}
job {
name: 'worker'
tasks { key: 0 value: '10.1.2.5:8470' }
tasks { key: 1 value: '10.1.2.3:8470' }
tasks { key: 2 value: '10.1.2.4:8470' }
tasks { key: 3 value: '10.1.2.8:8470' }
tasks { key: 4 value: '10.1.2.6:8470' }
tasks { key: 5 value: '10.1.2.7:8470' }
}
"""
self._verifyClusterSpecEquality(actual_cluster_spec, str(expected_proto))
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/kubernetes_cluster_resolver_test.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Implementation of Cluster Resolvers for TF_CONFIG Environment Variables."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import json
import os
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import ClusterResolver
from tensorflow.python.training.server_lib import ClusterSpec
from tensorflow.python.util.tf_export import tf_export
_TF_CONFIG_ENV = 'TF_CONFIG'
_SESSION_MASTER_KEY = 'session_master'
_RPC_LAYER_KEY = 'rpc_layer'
_TASK_KEY = 'task'
def format_master_url(master, rpc_layer=None):
if rpc_layer:
return '%s://%s' % (rpc_layer, master)
else:
return master
def _load_tf_config():
return json.loads(os.environ.get(_TF_CONFIG_ENV, '{}'))
def _get_value_in_tfconfig(key, default=None):
tf_config = _load_tf_config()
return tf_config[key] if key in tf_config else default
@tf_export('distribute.cluster_resolver.TFConfigClusterResolver')
class TFConfigClusterResolver(ClusterResolver):
"""Implementation of a ClusterResolver which reads the TF_CONFIG EnvVar."""
def __init__(self,
task_type=None,
task_id=None,
rpc_layer=None,
environment=None):
"""Creates a new TFConfigClusterResolver.
Args:
task_type: (String, optional) Overrides the task type specified in the
TF_CONFIG environment variable.
task_id: (Integer, optional) Overrides the task index specified in the
TF_CONFIG environment variable.
rpc_layer: (String, optional) Overrides the rpc layer TensorFlow uses.
environment: (String, optional) Overrides the environment TensorFlow
operates in.
"""
self._task_type = task_type
self._task_id = task_id
self._rpc_layer = rpc_layer
self._environment = environment
@property
def task_type(self):
if self._task_type is None:
task_info = _get_value_in_tfconfig(_TASK_KEY, {})
return str(task_info['type']) if 'type' in task_info else None
else:
return str(self._task_type)
@property
def task_id(self):
if self._task_type is None:
task_info = _get_value_in_tfconfig(_TASK_KEY, {})
return int(task_info['index']) if 'index' in task_info else None
else:
return int(self._task_id)
@task_type.setter
def task_type(self, task_type):
self._task_type = task_type
@task_id.setter
def task_id(self, task_id):
self._task_id = task_id
@property
def environment(self):
return self._environment
@property
def rpc_layer(self):
if self._rpc_layer is None:
return _get_value_in_tfconfig(_RPC_LAYER_KEY)
else:
return self._rpc_layer
@rpc_layer.setter
def rpc_layer(self, rpc_layer):
self._rpc_layer = rpc_layer
def num_accelerators(self,
task_type=None,
task_id=None,
config_proto=None):
task_type = self.task_type if task_type is None else task_type
task_id = self.task_id if task_id is None else task_id
return super(TFConfigClusterResolver, self).num_accelerators(
task_type, task_id, config_proto)
def cluster_spec(self):
"""Returns a ClusterSpec based on the TF_CONFIG environment variable.
Returns:
A ClusterSpec with information from the TF_CONFIG environment variable.
"""
tf_config = _load_tf_config()
if 'cluster' not in tf_config:
return ClusterSpec({})
return ClusterSpec(tf_config['cluster'])
def master(self, task_type=None, task_id=None, rpc_layer=None):
"""Returns the master address to use when creating a TensorFlow session.
Args:
task_type: (String, optional) Overrides and sets the task_type of the
master.
task_id: (Integer, optional) Overrides and sets the task id of the
master.
rpc_layer: (String, optional) Overrides and sets the protocol over which
TensorFlow nodes communicate with each other.
Returns:
The address of the master.
Raises:
RuntimeError: If the task_type or task_id is not specified and the
`TF_CONFIG` environment variable does not contain a task section.
"""
# If `session_master` is set, just use that.
session_master = _get_value_in_tfconfig(_SESSION_MASTER_KEY)
if session_master is not None:
return session_master
# Return an empty string if we are the only job in the ClusterSpec.
cluster_spec = self.cluster_spec()
if (not cluster_spec.jobs or
(len(cluster_spec.jobs) == 1 and
len(cluster_spec.job_tasks(cluster_spec.jobs[0])) == 1)):
return ''
# We try to auto-detect the task type and id, but uses the user-supplied one
# where available
task_type = task_type if task_type is not None else self.task_type
task_id = task_id if task_id is not None else self.task_id
return format_master_url(cluster_spec.task_address(task_type, task_id),
self.rpc_layer)
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/tfconfig_cluster_resolver.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for SlurmClusterResolver."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
from tensorflow.python.distribute.cluster_resolver import SlurmClusterResolver
from tensorflow.python.platform import test
from tensorflow.python.training import server_lib
mock = test.mock
class SlurmClusterResolverTest(test.TestCase):
def mock_resolve_hostnames_output(self):
return ['t02n13', 't02n41', 't02n43', 't02n44']
def _verifyClusterSpecEquality(self, cluster_spec, expected_proto):
self.assertProtoEquals(expected_proto, cluster_spec.as_cluster_def())
self.assertProtoEquals(
expected_proto,
server_lib.ClusterSpec(cluster_spec).as_cluster_def())
self.assertProtoEquals(
expected_proto,
server_lib.ClusterSpec(cluster_spec.as_cluster_def()).as_cluster_def())
self.assertProtoEquals(
expected_proto,
server_lib.ClusterSpec(cluster_spec.as_dict()).as_cluster_def())
@mock.patch.dict(os.environ, {'SLURM_PROCID': '0', 'SLURM_NTASKS': '3'})
@mock.patch.object(SlurmClusterResolver, '_resolve_hostnames',
mock_resolve_hostnames_output)
def testSimpleSuccessfulRetrieval(self):
slurm_cluster_resolver = SlurmClusterResolver(
jobs={
'ps': 1,
'worker': 2
},
port_base=8888,
tasks_per_node=1,
gpus_per_node=1,
gpus_per_task=1,
auto_set_gpu=False)
actual_cluster_spec = slurm_cluster_resolver.cluster_spec()
expected_proto = """
job { name: 'ps' tasks { value: 't02n13:8888' } }
job { name: 'worker' tasks { key: 0 value: 't02n41:8888' }
tasks { key: 1 value: 't02n43:8888' } }
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
@mock.patch.dict(os.environ, {'SLURM_PROCID': '0', 'SLURM_NTASKS': '3'})
@mock.patch.object(SlurmClusterResolver, '_resolve_hostnames',
mock_resolve_hostnames_output)
def testSimpleMasterRetrieval(self):
slurm_cluster_resolver = SlurmClusterResolver(
jobs={
'ps': 1,
'worker': 2
},
port_base=8888,
tasks_per_node=1,
gpus_per_node=1,
gpus_per_task=1,
auto_set_gpu=False)
slurm_cluster_resolver.task_type = 'worker'
slurm_cluster_resolver.task_id = 1
self.assertEqual(slurm_cluster_resolver.master(), 'grpc://t02n43:8888')
slurm_cluster_resolver.rpc_layer = 'ab'
self.assertEqual(slurm_cluster_resolver.master('ps', 0), 'ab://t02n13:8888')
self.assertEqual(
slurm_cluster_resolver.master('ps', 0, rpc_layer='test'),
'test://t02n13:8888')
@mock.patch.dict(os.environ, {
'SLURM_PROCID': '0',
'SLURM_NTASKS': '3',
'SLURM_NTASKS_PER_NODE': '1'
})
@mock.patch.object(SlurmClusterResolver, '_resolve_hostnames',
mock_resolve_hostnames_output)
def testTaskPerNodeNotSetRetrieval(self):
slurm_cluster_resolver = SlurmClusterResolver(
jobs={
'ps': 1,
'worker': 2
},
port_base=8888,
gpus_per_node=1,
gpus_per_task=1,
auto_set_gpu=False)
actual_cluster_spec = slurm_cluster_resolver.cluster_spec()
expected_proto = """
job { name: 'ps' tasks { value: 't02n13:8888' } }
job { name: 'worker' tasks { key: 0 value: 't02n41:8888' }
tasks { key: 1 value: 't02n43:8888' } }
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
@mock.patch.dict(
os.environ, {
'SLURM_PROCID': '1',
'SLURM_NTASKS': '5',
'SLURM_NTASKS_PER_NODE': '2',
'CUDA_VISIBLE_DEVICES': ''
})
@mock.patch.object(SlurmClusterResolver, '_resolve_hostnames',
mock_resolve_hostnames_output)
def testMultiTaskPerNodeRetrieval(self):
slurm_cluster_resolver = SlurmClusterResolver(
jobs={
'ps': 1,
'worker': 4
},
port_base=8888,
gpus_per_node=2,
gpus_per_task=1,
auto_set_gpu=True)
actual_cluster_spec = slurm_cluster_resolver.cluster_spec()
expected_proto = """
job { name: 'ps' tasks { value: 't02n13:8888' } }
job { name: 'worker' tasks { key: 0 value: 't02n13:8889' }
tasks { key: 1 value: 't02n41:8888' }
tasks { key: 2 value: 't02n41:8889' }
tasks { key: 3 value: 't02n43:8888' } }
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
assert os.environ['CUDA_VISIBLE_DEVICES'] == '1'
@mock.patch.dict(
os.environ, {
'SLURM_PROCID': '1',
'SLURM_NTASKS': '5',
'SLURM_NTASKS_PER_NODE': '2',
'CUDA_VISIBLE_DEVICES': ''
})
@mock.patch.object(SlurmClusterResolver, '_resolve_hostnames',
mock_resolve_hostnames_output)
def testMultipleGpusPerTaskRetrieval(self):
slurm_cluster_resolver = SlurmClusterResolver(
jobs={
'ps': 1,
'worker': 4
},
port_base=8888,
gpus_per_node=4,
gpus_per_task=2,
auto_set_gpu=True)
actual_cluster_spec = slurm_cluster_resolver.cluster_spec()
expected_proto = """
job { name: 'ps' tasks { value: 't02n13:8888' } }
job { name: 'worker' tasks { key: 0 value: 't02n13:8889' }
tasks { key: 1 value: 't02n41:8888' }
tasks { key: 2 value: 't02n41:8889' }
tasks { key: 3 value: 't02n43:8888' } }
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
assert os.environ['CUDA_VISIBLE_DEVICES'] == '2,3'
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/slurm_cluster_resolver_test.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Implementation of Cluster Resolvers for Cloud TPUs."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import os
import re
from six.moves import urllib
from six.moves.urllib.error import URLError
from six.moves.urllib.request import Request
from six.moves.urllib.request import urlopen
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import ClusterResolver
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import format_master_url
from tensorflow.python.distribute.cluster_resolver.cluster_resolver import get_accelerator_devices
from tensorflow.python.framework import errors
from tensorflow.python.framework import ops
from tensorflow.python.platform import tf_logging as logging
from tensorflow.python.training import server_lib
from tensorflow.python.util import compat
from tensorflow.python.util.tf_export import tf_export
_GOOGLE_API_CLIENT_INSTALLED = True
try:
from googleapiclient import discovery # pylint: disable=g-import-not-at-top
from oauth2client.client import GoogleCredentials # pylint: disable=g-import-not-at-top
except ImportError:
_GOOGLE_API_CLIENT_INSTALLED = False
_GKE_ENV_VARIABLE = 'KUBE_GOOGLE_CLOUD_TPU_ENDPOINTS'
_ENDPOINTS_SEPARATOR = ','
_DEFAULT_ENV_VARIABLE = 'TPU_NAME'
_DISCOVERY_SERVICE_URL_ENV_VARIABLE = 'TPU_API_DISCOVERY_URL'
_TPU_DEVICE_REGEX = re.compile(
r'.*task:(?P<host_id>\d+)/.*device:TPU:(?P<core_id>\d+)$')
_TPU_CONN_RETRIES = 120
_GCE_METADATA_ENDPOINT = 'http://metadata.google.internal'
DeviceDetails = collections.namedtuple(
'DeviceDetails', ['device_map', 'total_cores'])
def is_running_in_gce():
"""Checks for GCE presence by attempting to query the metadata service."""
try:
req = Request(
'%s/computeMetadata/v1' % _GCE_METADATA_ENDPOINT,
headers={'Metadata-Flavor': 'Google'})
resp = urllib.request.urlopen(req, timeout=1)
info = resp.info()
if 'Metadata-Flavor' in info and info['Metadata-Flavor'] == 'Google':
return True
except URLError:
pass
return False
@tf_export('distribute.cluster_resolver.TPUClusterResolver')
class TPUClusterResolver(ClusterResolver):
"""Cluster Resolver for Google Cloud TPUs.
This is an implementation of cluster resolvers for the Google Cloud TPU
service. As Cloud TPUs are in alpha, you will need to specify a API definition
file for this to consume, in addition to a list of Cloud TPUs in your Google
Cloud Platform project.
TPUClusterResolver supports the following distinct environments:
Google Compute Engine
Google Kubernetes Engine
Google internal
"""
def _tpu_service(self):
"""Creates a new Cloud TPU API object.
This works around an issue where the underlying HTTP connection sometimes
times out when the script has been running for too long. Other methods in
this object calls this method to get a new API object whenever they need
to communicate with the Cloud API.
Returns:
A Google Cloud TPU API object.
"""
if self._service:
return self._service
credentials = self._credentials
if credentials is None or credentials == 'default':
credentials = GoogleCredentials.get_application_default()
if self._discovery_url:
return discovery.build(
'tpu', 'v1alpha1', credentials=credentials,
discoveryServiceUrl=self._discovery_url, cache_discovery=False)
else:
return discovery.build(
'tpu', 'v1alpha1', credentials=credentials, cache_discovery=False)
def _request_compute_metadata(self, path):
req = Request('%s/computeMetadata/v1/%s' % (_GCE_METADATA_ENDPOINT, path),
headers={'Metadata-Flavor': 'Google'})
resp = urlopen(req)
return compat.as_bytes(resp.read())
def _is_google_environment(self):
return (
self._tpu == compat.as_bytes('') or
self._tpu == compat.as_bytes('local') or
self._tpu.startswith(compat.as_bytes('localhost:')) or
self._tpu.startswith(compat.as_bytes('/bns')) or
self._tpu.startswith(compat.as_bytes('uptc://')))
def _should_resolve(self):
if isinstance(self._should_resolve_override, bool):
return self._should_resolve_override
else:
return not (self._tpu.startswith(compat.as_bytes('grpc://')) or
self._is_google_environment())
@staticmethod
def _get_device_dict_and_cores(devices):
"""Returns a dict of hosts to cores and total cores given devices names.
Returns a namedtuple with two attributes:
device_map: A map of host_ids to a list of core_ids.
total_cores: The total number of cores within the TPU system.
Args:
devices: A list of devices returned by session.list_devices()
"""
device_map = collections.defaultdict(list)
num_cores = 0
for device in devices:
match = _TPU_DEVICE_REGEX.match(device.name)
if match:
host_id = match.group('host_id')
core_id = match.group('core_id')
device_map[host_id].append(core_id)
num_cores += 1
return DeviceDetails(device_map, num_cores)
@staticmethod
def _verify_and_return_same_core_count(device_dict):
"""Verifies that every device in device_dict has the same # of cores."""
num_cores_per_host_set = (
{len(core_ids) for core_ids in device_dict.values()})
if len(num_cores_per_host_set) != 1:
raise RuntimeError('TPU cores on each device is not the same. This '
'should never happen. Devices: {}'.format(device_dict))
return num_cores_per_host_set.pop()
@staticmethod
def _in_gke():
"""When running in GKE, the environment variable will be set."""
return _GKE_ENV_VARIABLE in os.environ
@staticmethod
def _gke_endpoints():
return os.environ[_GKE_ENV_VARIABLE]
@staticmethod
def _env_var_fallback():
if _DEFAULT_ENV_VARIABLE in os.environ:
return os.environ[_DEFAULT_ENV_VARIABLE]
return None
@staticmethod
def _environment_discovery_url():
return os.environ.get(_DISCOVERY_SERVICE_URL_ENV_VARIABLE)
def __init__(self,
tpu=None,
zone=None,
project=None,
job_name='worker',
coordinator_name=None,
coordinator_address=None,
credentials='default',
service=None,
discovery_url=None):
"""Creates a new TPUClusterResolver object.
The ClusterResolver will then use the parameters to query the Cloud TPU APIs
for the IP addresses and ports of each Cloud TPU listed.
Args:
tpu: A string corresponding to the TPU to use. If the string is the empty
string, the string 'local', or a string that begins with 'grpc://' or
'/bns', then it is assumed to not correspond with a Cloud TPU and will
instead be passed as the session master and no ClusterSpec propagation
will be done. In the future, this may also support a list of strings
when multiple Cloud TPUs are used.
zone: Zone where the TPUs are located. If omitted or empty, we will assume
that the zone of the TPU is the same as the zone of the GCE VM, which we
will try to discover from the GCE metadata service.
project: Name of the GCP project containing Cloud TPUs. If omitted or
empty, we will try to discover the project name of the GCE VM from the
GCE metadata service.
job_name: Name of the TensorFlow job the TPUs belong to.
coordinator_name: The name to use for the coordinator. Set to None if the
coordinator should not be included in the computed ClusterSpec.
coordinator_address: The address of the coordinator (typically an ip:port
pair). If set to None, a TF server will be started. If coordinator_name
is None, a TF server will not be started even if coordinator_address is
None.
credentials: GCE Credentials. If None, then we use default credentials
from the oauth2client
service: The GCE API object returned by the googleapiclient.discovery
function. If you specify a custom service object, then the credentials
parameter will be ignored.
discovery_url: A URL template that points to the location of the discovery
service. It should have two parameters {api} and {apiVersion} that when
filled in produce an absolute URL to the discovery document for that
service. The environment variable 'TPU_API_DISCOVERY_URL' will override
this.
Raises:
ImportError: If the googleapiclient is not installed.
ValueError: If no TPUs are specified.
RuntimeError: If an empty TPU name is specified and this is running in a
Google Cloud environment.
"""
if isinstance(tpu, list):
if not tpu:
raise ValueError('At least one TPU must be specified.')
if len(tpu) != 1:
raise NotImplementedError(
'Using multiple TPUs in a single session is not yet implemented')
tpu = tpu[0]
in_gke = self._in_gke()
# When using GKE with Cloud TPUs, the env variable will be set.
if tpu is None:
if in_gke:
tpu = self._gke_endpoints()
else:
tpu = self._env_var_fallback()
if tpu is None:
raise ValueError('Please provide a TPU Name to connect to.')
self._tpu = compat.as_bytes(tpu) # self._tpu is always bytes
# If we are running in Cloud and don't specify a TPU name
if is_running_in_gce() and not self._tpu:
raise RuntimeError('You need to specify a TPU Name if you are running in '
'the Google Cloud environment.')
# By default the task_type is 'worker` and the task_id is 0 (which is the
# first worker in the task).
self.task_type = job_name
self.task_id = 0
if self._is_google_environment():
self._environment = 'google'
self.rpc_layer = None
else:
self._environment = ''
self.rpc_layer = 'grpc'
# Setting this overrides the return value of self._should_resolve()
self._should_resolve_override = None
# We strip out the protocol if it is included, and override the
# shouldResolve function to never resolve. We are adding the protocol back
# in later in self.master().
if self.rpc_layer is not None and tpu.startswith(self.rpc_layer + '://'):
tpu = tpu[len(self.rpc_layer + '://'):]
self._tpu = compat.as_bytes(tpu) # self._tpu is always bytes
self._should_resolve_override = False
# Whether we should actually attempt to contact Cloud APIs
should_resolve = self._should_resolve()
# We error out if we are in a non-Cloud environment which cannot talk to the
# Cloud APIs using the standard class and a special object is not passed in.
self._service = service
if (self._service is None and should_resolve and
not _GOOGLE_API_CLIENT_INSTALLED):
raise ImportError('googleapiclient and oauth2client must be installed '
'before using the TPU cluster resolver. Execute: '
'`pip install --upgrade google-api-python-client` '
'and `pip install --upgrade oauth2client` to '
'install with pip.')
# We save user-passed credentials, unless the user didn't pass in anything.
self._credentials = credentials
if (credentials == 'default' and should_resolve and
_GOOGLE_API_CLIENT_INSTALLED):
self._credentials = None
# Automatically detect project and zone if unspecified.
if not project and should_resolve:
project = compat.as_str(
self._request_compute_metadata('project/project-id'))
if not zone and should_resolve:
zone_path = compat.as_str(self._request_compute_metadata('instance/zone'))
zone = zone_path.split('/')[-1]
self._project = project
self._zone = zone
self._discovery_url = self._environment_discovery_url() or discovery_url
self._coordinator_name = coordinator_name
if (coordinator_name and not coordinator_address and
(should_resolve or in_gke)):
self._start_local_server()
else:
self._coordinator_address = coordinator_address
def master(self, task_type=None, task_id=None, rpc_layer=None):
"""Get the Master string to be used for the session.
In the normal case, this returns the grpc path (grpc://1.2.3.4:8470) of
first instance in the ClusterSpec returned by the cluster_spec function.
If a non-TPU name is used when constructing a TPUClusterResolver, that will
be returned instead (e.g. If the tpus argument's value when constructing
this TPUClusterResolver was 'grpc://10.240.1.2:8470',
'grpc://10.240.1.2:8470' will be returned).
Args:
task_type: (Optional, string) The type of the TensorFlow task of the
master.
task_id: (Optional, integer) The index of the TensorFlow task of the
master.
rpc_layer: (Optional, string) The RPC protocol TensorFlow should use to
communicate with TPUs.
Returns:
string, the connection string to use when creating a session.
Raises:
ValueError: If none of the TPUs specified exists.
"""
if self._should_resolve():
# We are going to communicate with the Cloud TPU APIs to get a Cluster.
cluster_spec = self.cluster_spec()
if task_type is not None and task_id is not None:
# task_type and task_id is from the function parameter
master = cluster_spec.task_address(task_type, task_id)
elif self.task_type is not None and self.task_id is not None:
# task_type and task_id is from the object
master = cluster_spec.task_address(self.task_type, self.task_id)
else:
# by default we take the first item in the cluster with the right name
job_tasks = cluster_spec.job_tasks(self.task_type)
if not job_tasks:
raise ValueError('No TPUs with the specified names exist.')
master = job_tasks[0]
else:
if isinstance(self._tpu, (bytes, bytearray)):
master = compat.as_text(self._tpu).split(_ENDPOINTS_SEPARATOR)[0]
else:
master = self._tpu.split(_ENDPOINTS_SEPARATOR)[0]
return format_master_url(master, rpc_layer or self.rpc_layer)
def get_master(self):
return self.master()
def get_job_name(self):
if ops.executing_eagerly_outside_functions() or self._should_resolve(
) or is_running_in_gce():
return self.task_type
def cluster_spec(self):
"""Returns a ClusterSpec object based on the latest TPU information.
We retrieve the information from the GCE APIs every time this method is
called.
Returns:
A ClusterSpec containing host information returned from Cloud TPUs.
Raises:
RuntimeError: If the provided TPU is not healthy.
"""
############################################################################
# There are 5 potential cases this code must handle:
# 1. [Normal case.] We should resolve the TPU name to a set of tasks, and
# a. Create a ClusterSpec that includes the coordinator job
# b. Create a ClusterSpec without the coordinator job.
# 2. [GKE / No API Access.] We should not resolve the TPU name to a set of
# tasks and
# a. Create a ClusterSpec with the coordinator
# b. Create a ClusterSpec without the coordinator
# 3. [Other (legacy non-gRPC).] We should return an empty ClusterSpec.
############################################################################
if self._should_resolve():
# Case 1.
response = self._fetch_cloud_tpu_metadata() # pylint: disable=protected-access
if 'state' in response and response['state'] != 'READY':
raise RuntimeError('TPU "%s" is not yet ready; state: "%s"' %
(compat.as_text(self._tpu), response['state']))
if 'networkEndpoints' in response:
worker_list = [
'%s:%s' % (endpoint['ipAddress'], endpoint['port'])
for endpoint in response['networkEndpoints']
]
else:
# Fall back to the deprecated response format
instance_url = '%s:%s' % (response['ipAddress'], response['port'])
worker_list = [instance_url]
cluster_spec = {self.task_type: worker_list}
else:
if self.rpc_layer is None:
# Case 3.
return None
# Case 2.
tpus = []
for tpu in compat.as_text(self._tpu).split(_ENDPOINTS_SEPARATOR):
# We are working around the fact that GKE environment variable that is
# supplied to us has the protocol string embedded in it, but we want
# to strip it out for the ClusterSpec.
if (self.rpc_layer is not None and
tpu.startswith(self.rpc_layer + '://')):
tpus.append(tpu[len(self.rpc_layer + '://'):])
else:
tpus.append(tpu)
cluster_spec = {self.task_type: tpus}
if self._coordinator_address:
# {1, 2}.a
cluster_spec[self._coordinator_name] = [self._coordinator_address]
return server_lib.ClusterSpec(cluster_spec)
def _fetch_cloud_tpu_metadata(self):
"""Returns the TPU metadata object from the TPU Get API call."""
res = []
try:
full_name = 'projects/%s/locations/%s/nodes/%s' % (
self._project, self._zone, compat.as_text(self._tpu))
service = self._tpu_service()
request = service.projects().locations().nodes().get(name=full_name)
res = request.execute()
except: # pylint: disable=bare-except
pass
finally:
return res # pylint: disable=lost-exception
def num_accelerators(self,
task_type=None,
task_id=None,
config_proto=None):
"""Returns the number of TPU cores per worker.
Connects to the master and list all the devices present in the master,
and counts them up. Also verifies that the device counts per host in the
cluster is the same before returning the number of TPU cores per host.
Args:
task_type: Unused.
task_id: Unused.
config_proto: Used to create a connection to a TPU master in order to
retrieve the system metadata.
Raises:
RuntimeError: If we cannot talk to a TPU worker after retrying or if the
number of TPU devices per host is different.
"""
retry_count = 1
# TODO(b/120564445): Replace with standard library for retries.
while True:
try:
device_details = TPUClusterResolver._get_device_dict_and_cores(
get_accelerator_devices(self.master(), config_proto=config_proto))
break
except errors.DeadlineExceededError:
error_message = ('Failed to connect to master. The TPU might not be '
'ready (e.g. still scheduling) or the master '
'address is incorrect: got (%s)' % self.master())
if retry_count <= _TPU_CONN_RETRIES:
logging.warning(error_message)
logging.warning('Retrying (%d/%d)...', retry_count, _TPU_CONN_RETRIES)
retry_count += 1
else:
raise RuntimeError(error_message)
if device_details.total_cores:
return {'TPU': TPUClusterResolver._verify_and_return_same_core_count(
device_details.device_map)}
return {'TPU': 0}
@property
def environment(self):
"""Returns the current environment which TensorFlow is running in."""
return self._environment
def _start_local_server(self):
address = compat.as_text(
self._request_compute_metadata('instance/network-interfaces/0/ip'))
self._server = server_lib.Server(
{
'local': ['0.0.0.0:0']
}, protocol='grpc', config=None, start=True)
# self._server.target is of the form: grpc://ipaddress:port
target = compat.as_bytes(self._server.target)
splits = target.split(compat.as_bytes(':'))
assert len(splits) == 3, self._server.target
assert splits[0] == compat.as_bytes('grpc'), self._server.target
self._coordinator_port = compat.as_text(splits[2])
self._coordinator_address = '%s:%s' % (
address, compat.as_text(self._coordinator_port))
def __deepcopy__(self, memo):
# TODO(b/73668574): Remove this once RunConfig avoids performing deepcopy.
return self
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/tpu_cluster_resolver.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Cluster Resolvers are used for dynamic cluster IP/hostname resolution."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import abc
import collections
import re
import six
from tensorflow.python.client import session
from tensorflow.python.eager import context
from tensorflow.python.framework import ops
from tensorflow.python.training.server_lib import ClusterSpec
from tensorflow.python.util.tf_export import tf_export
DEVICE_TYPE_REGEX = re.compile('.*device:([^:]+).*')
def format_master_url(master, rpc_layer=None):
if rpc_layer:
return '%s://%s' % (rpc_layer, master)
else:
return master
def get_accelerator_devices(master, config_proto):
"""Returns accelerator devices given a master and a configuration."""
if context.executing_eagerly():
device_names = context.list_devices() # list_devices returns list(string)
devices = []
for name in device_names:
device_type = 'GPU' # default device type is GPU
device_match = DEVICE_TYPE_REGEX.match(name)
if device_match:
device_type = device_match.group(1)
if device_type == 'CPU' or device_type == 'XLA_CPU': # Filter CPUs
continue
devices.append(session._DeviceAttributes(name, device_type, 0, 0)) # pylint: disable=protected-access
return devices
else:
with ops.Graph().as_default():
with session.Session(master, config=config_proto) as s:
devices = s.list_devices()
return devices
@tf_export('distribute.cluster_resolver.ClusterResolver')
@six.add_metaclass(abc.ABCMeta)
class ClusterResolver(object):
"""Abstract class for all implementations of ClusterResolvers.
This defines the skeleton for all implementations of ClusterResolvers.
ClusterResolvers are a way for TensorFlow to communicate with various cluster
management systems (e.g. GCE, AWS, etc...).
By letting TensorFlow communicate with these systems, we will be able to
automatically discover and resolve IP addresses for various TensorFlow
workers. This will eventually allow us to automatically recover from
underlying machine failures and scale TensorFlow worker clusters up and down.
Note to Implementors: In addition to these abstract methods, you must also
implement the task_type, task_id, and rpc_layer attributes. You may choose
to implement them either as properties with getters or setters or directly
set the attributes.
- task_type is the name of the server's current named job (e.g. 'worker',
'ps' in a distributed parameterized training job).
- task_id is the ordinal index of the server within the task type.
- rpc_layer is the protocol used by TensorFlow to communicate with other
TensorFlow servers in a distributed environment.
"""
@abc.abstractmethod
def cluster_spec(self):
"""Retrieve the current state of the cluster and returns a ClusterSpec.
Returns:
A ClusterSpec representing the state of the cluster at the moment this
function is called.
Implementors of this function must take care in ensuring that the
ClusterSpec returned is up-to-date at the time of calling this function.
This usually means retrieving the information from the underlying cluster
management system every time this function is invoked and reconstructing
a cluster_spec, rather than attempting to cache anything.
"""
raise NotImplementedError()
@abc.abstractmethod
def master(self, task_type=None, task_id=None, rpc_layer=None):
"""Retrieves the name or URL of the session master.
Args:
task_type: (Optional) The type of the TensorFlow task of the master.
task_id: (Optional) The index of the TensorFlow task of the master.
rpc_layer: (Optional) The RPC protocol for the given cluster.
Returns:
The name or URL of the session master.
Implementors of this function must take care in ensuring that the master
returned is up-to-date at the time to calling this function. This usually
means retrieving the master every time this function is invoked.
"""
raise NotImplementedError()
def num_accelerators(self,
task_type=None,
task_id=None,
config_proto=None):
"""Returns the number of accelerator cores per worker.
This returns the number of accelerator cores (such as GPUs and TPUs)
available per worker.
Optionally, we allow callers to specify the task_type, and task_id, for
if they want to target a specific TensorFlow process to query
the number of accelerators. This is to support heterogenous environments,
where the number of accelerators cores per host is different.
Args:
task_type: (Optional) The type of the TensorFlow task of the machine we
want to query.
task_id: (Optional) The index of the TensorFlow task of the machine we
want to query.
config_proto: (Optional) Configuration for starting a new session to
query how many accelerator cores it has.
Returns:
A map of accelerator types to number of cores.
"""
master = self.master(task_type, task_id)
devices = get_accelerator_devices(master, config_proto)
mapping = collections.defaultdict(int)
for device in devices:
if task_type is not None and task_id is not None:
job_path = '/job:%s' % task_type
task_path = '/task:%s' % task_id
if job_path not in device.name or task_path not in device.name:
continue
mapping[device.device_type] += 1
return mapping
@property
def environment(self):
"""Returns the current environment which TensorFlow is running in.
There are two possible return values, "google" (when TensorFlow is running
in a Google-internal environment) or an empty string (when TensorFlow is
running elsewhere).
If you are implementing a ClusterResolver that works in both the Google
environment and the open-source world (for instance, a TPU ClusterResolver
or similar), you will have to return the appropriate string depending on the
environment, which you will have to detect.
Otherwise, if you are implementing a ClusterResolver that will only work
in open-source TensorFlow, you do not need to implement this property.
"""
return ''
@tf_export('distribute.cluster_resolver.SimpleClusterResolver')
class SimpleClusterResolver(ClusterResolver):
"""Simple implementation of ClusterResolver that accepts a ClusterSpec."""
def __init__(self, cluster_spec, master='', task_type=None, task_id=None,
environment='', num_accelerators=None,
rpc_layer=None):
"""Creates a SimpleClusterResolver from a ClusterSpec."""
super(SimpleClusterResolver, self).__init__()
self._task_type = task_type
self._task_id = task_id
self._environment = environment
self._num_accelerators = num_accelerators
self._rpc_layer = rpc_layer
if not isinstance(cluster_spec, ClusterSpec):
raise TypeError('cluster_spec must be a ClusterSpec.')
self._cluster_spec = cluster_spec
if not isinstance(master, str):
raise TypeError('master must be a string.')
self._master = master
def cluster_spec(self):
"""Returns the ClusterSpec passed into the constructor."""
return self._cluster_spec
def master(self, task_type=None, task_id=None, rpc_layer=None):
"""Returns the master address to use when creating a session.
Args:
task_type: (Optional) The type of the TensorFlow task of the master.
task_id: (Optional) The index of the TensorFlow task of the master.
rpc_layer: (Optional) The RPC used by distributed TensorFlow.
Returns:
The name or URL of the session master.
If a task_type and task_id is given, this will override the `master`
string passed into the initialization function.
"""
if task_type is not None and task_id is not None:
master = self.cluster_spec().task_address(task_type, task_id)
else:
master = self._master
return format_master_url(master, rpc_layer=rpc_layer or self._rpc_layer)
@property
def task_type(self):
return self._task_type
@property
def task_id(self):
return self._task_id
@task_type.setter
def task_type(self, task_type):
self._task_type = task_type
@task_id.setter
def task_id(self, task_id):
self._task_id = task_id
@property
def environment(self):
return self._environment
def num_accelerators(self,
task_type=None,
task_id=None,
config_proto=None):
"""Returns the number of accelerator cores per worker.
The SimpleClusterResolver does not do automatic detection of accelerators,
so a TensorFlow session will never be created, and thus all arguments are
unused and we simply assume that the type of accelerator is a GPU and return
the value in provided to us in the constructor.
Args:
task_type: Unused.
task_id: Unused.
config_proto: Unused.
"""
# Unused
del task_type, task_id, config_proto
if self._num_accelerators is None:
return {}
return self._num_accelerators
@property
def rpc_layer(self):
return self._rpc_layer
@rpc_layer.setter
def rpc_layer(self, rpc_layer):
self._rpc_layer = rpc_layer
@tf_export('distribute.cluster_resolver.UnionResolver')
class UnionClusterResolver(ClusterResolver):
"""Performs a union on underlying ClusterResolvers.
This class performs a union given two or more existing ClusterResolvers. It
merges the underlying ClusterResolvers, and returns one unified ClusterSpec
when cluster_spec is called. The details of the merge function is
documented in the cluster_spec function.
For additional Cluster Resolver properties such as task type, task index,
rpc layer, environment, etc..., we will return the value from the first
ClusterResolver in the union.
"""
def __init__(self, *args, **kwargs):
"""Initializes a UnionClusterResolver with other ClusterResolvers.
Args:
*args: `ClusterResolver` objects to be unionized.
**kwargs:
rpc_layer - (Optional) Override value for the RPC layer used by
TensorFlow.
task_type - (Optional) Override value for the current task type.
task_id - (Optional) Override value for the current task index.
Raises:
TypeError: If any argument is not a subclass of `ClusterResolvers`.
ValueError: If there are no arguments passed.
"""
super(UnionClusterResolver, self).__init__()
self._rpc_layer = kwargs.pop('rpc_layer', None)
self._task_type = kwargs.pop('task_type', None)
self._task_id = kwargs.pop('task_id', None)
if kwargs:
raise ValueError('Unexpected kwargs provided {!r}'.format(kwargs))
if not args:
raise ValueError('At least one ClusterResolver is required.')
for cluster_resolver in args:
if not isinstance(cluster_resolver, ClusterResolver):
raise TypeError('All arguments must be a sub-class of '
'`ClusterResolver.`')
self._cluster_resolvers = args
def cluster_spec(self):
"""Returns a union of all the ClusterSpecs from the ClusterResolvers.
Returns:
A ClusterSpec containing host information merged from all the underlying
ClusterResolvers.
Raises:
KeyError: If there are conflicting keys detected when merging two or
more dictionaries, this exception is raised.
Note: If there are multiple ClusterResolvers exposing ClusterSpecs with the
same job name, we will merge the list/dict of workers.
If *all* underlying ClusterSpecs expose the set of workers as lists, we will
concatenate the lists of workers, starting with the list of workers from
the first ClusterResolver passed into the constructor.
If *any* of the ClusterSpecs expose the set of workers as a dict, we will
treat all the sets of workers as dicts (even if they are returned as lists)
and will only merge them into a dict if there is no conflicting keys. If
there is a conflicting key, we will raise a `KeyError`.
"""
merged_cluster = {}
# We figure out whether it is all lists for a particular job, or whether
# there are dicts inside.
for cluster_resolver in self._cluster_resolvers:
cluster_spec = cluster_resolver.cluster_spec()
cluster_dict = cluster_spec.as_dict()
for job_name, tasks in cluster_dict.items():
if job_name in merged_cluster:
# If we see a dict, then we write a dict out regardless.
if isinstance(tasks, dict):
merged_cluster[job_name] = {}
else:
# We take whichever type is present.
if isinstance(tasks, list):
merged_cluster[job_name] = []
else:
merged_cluster[job_name] = {}
# We then do the merge as appropriate in merged_cluster[job].
for cluster_resolver in self._cluster_resolvers:
cluster_spec = cluster_resolver.cluster_spec()
cluster_dict = cluster_spec.as_dict()
for job_name, tasks in cluster_dict.items():
if isinstance(merged_cluster[job_name], list):
# We all have lists, we can just concatenate and be done.
merged_cluster[job_name].extend(tasks)
else:
if isinstance(tasks, list):
# We convert to a dictionary if the type is a list.
task_dict = dict(zip(range(0, len(tasks)), tasks))
else:
# We can simply make a copy (for update) and be done.
task_dict = tasks.copy()
# We detect if there are duplicates, and raise an error if so.
task_keys = set(task_dict)
merged_keys = set(merged_cluster[job_name].keys())
intersected_keys = task_keys.intersection(merged_keys)
if intersected_keys:
raise KeyError('Duplicate keys detected when merging two '
'ClusterSpecs: %s' % repr(intersected_keys))
# We do the merge after all the processing.
merged_cluster[job_name].update(task_dict)
return ClusterSpec(merged_cluster)
def master(self, task_type=None, task_id=None, rpc_layer=None):
"""Returns the master address to use when creating a session.
This usually returns the master from the first ClusterResolver passed in,
but you can override this by specifying the task_type and task_id.
Args:
task_type: (Optional) The type of the TensorFlow task of the master.
task_id: (Optional) The index of the TensorFlow task of the master.
rpc_layer: (Optional) The RPC protocol for the given cluster.
Returns:
The name or URL of the session master.
"""
if task_type is not None and task_id is not None:
master = self.cluster_spec().task_address(task_type, task_id)
return format_master_url(master, rpc_layer or self._rpc_layer)
return self._cluster_resolvers[0].master(rpc_layer=rpc_layer)
@property
def task_type(self):
return self._task_type or self._cluster_resolvers[0].task_type
@property
def task_id(self):
return self._task_id or self._cluster_resolvers[0].task_id
@task_type.setter
def task_type(self, task_type):
self._task_type = task_type
@task_id.setter
def task_id(self, task_id):
self._task_id = task_id
@property
def environment(self):
return self._cluster_resolvers[0].environment
def num_accelerators(self,
task_type=None,
task_id=None,
config_proto=None):
return self._cluster_resolvers[0].num_accelerators(
task_type, task_id, config_proto)
@property
def rpc_layer(self):
return self._rpc_layer or self._cluster_resolvers[0].rpc_layer
@rpc_layer.setter
def rpc_layer(self, rpc_layer):
self._rpc_layer = rpc_layer
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/cluster_resolver.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for TPUClusterResolver."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import six
from six.moves.urllib.error import URLError
from tensorflow.python import eager
from tensorflow.python.client import session
from tensorflow.python.distribute.cluster_resolver import tpu_cluster_resolver as resolver
from tensorflow.python.framework import errors
from tensorflow.python.framework import test_util
from tensorflow.python.platform import test
from tensorflow.python.training import server_lib
from tensorflow.python.util import compat
mock = test.mock
class MockRequestClass(object):
def __init__(self, name, tpu_map):
self._name = name
self._tpu_map = tpu_map
def execute(self):
if self._name in self._tpu_map:
return self._tpu_map[self._name]
else:
raise KeyError('Resource %s was not found' % self._name)
class MockNodeClass(object):
def __init__(self, tpu_map):
self._tpu_map = tpu_map
def get(self, name):
return MockRequestClass(name, self._tpu_map)
def mock_request_compute_metadata(cls, *args, **kwargs):
del cls, kwargs # Unused.
if args[0] == 'project/project-id':
return 'test-project'
elif args[0] == 'instance/zone':
return 'projects/test-project/locations/us-central1-c'
elif args[0] == 'instance/network-interfaces/0/ip':
return '10.128.1.2'
return ''
def mock_is_running_in_gce():
return True
def mock_is_not_running_in_gce():
return False
def mock_running_in_gce_urlopen(cls, *args, **kwargs):
del cls, args, kwargs # Unused.
mock_response = mock.MagicMock()
mock_response.info.return_value = {'Metadata-Flavor': 'Google'}
return mock_response
def mock_not_running_in_gce_urlopen(cls, *args, **kwargs):
del cls, args, kwargs # Unused.
raise URLError(reason='Host does not exist.')
@test_util.run_all_in_graph_and_eager_modes
class TPUClusterResolverTest(test.TestCase):
def _verifyClusterSpecEquality(self, cluster_spec, expected_proto):
"""Verifies that the ClusterSpec generates the correct proto.
We are testing this four different ways to ensure that the ClusterSpec
returned by the TPUClusterResolver behaves identically to a normal
ClusterSpec when passed into the generic ClusterSpec libraries.
Args:
cluster_spec: ClusterSpec returned by the TPUClusterResolver
expected_proto: Expected protobuf
"""
self.assertProtoEquals(expected_proto, cluster_spec.as_cluster_def())
self.assertProtoEquals(
expected_proto,
server_lib.ClusterSpec(cluster_spec).as_cluster_def())
self.assertProtoEquals(expected_proto,
server_lib.ClusterSpec(
cluster_spec.as_cluster_def()).as_cluster_def())
self.assertProtoEquals(expected_proto,
server_lib.ClusterSpec(
cluster_spec.as_dict()).as_cluster_def())
def mock_service_client(self, tpu_map=None):
if tpu_map is None:
tpu_map = {}
mock_locations = mock.MagicMock()
mock_locations.nodes.return_value = MockNodeClass(tpu_map)
mock_project = mock.MagicMock()
mock_project.locations.return_value = mock_locations
mock_client = mock.MagicMock()
mock_client.projects.return_value = mock_project
return mock_client
@mock.patch.object(resolver, 'is_running_in_gce',
mock_is_running_in_gce)
def testCheckRunningInGceWithNoTpuName(self):
with self.assertRaisesRegexp(RuntimeError, '.*Google Cloud.*'):
resolver.TPUClusterResolver(tpu='')
@mock.patch.object(six.moves.urllib.request,
'urlopen',
mock_running_in_gce_urlopen)
def testIsRunningInGce(self):
self.assertTrue(resolver.is_running_in_gce())
@mock.patch.object(six.moves.urllib.request,
'urlopen',
mock_not_running_in_gce_urlopen)
def testIsNotRunningInGce(self):
self.assertFalse(resolver.is_running_in_gce())
@mock.patch.object(resolver.TPUClusterResolver,
'_request_compute_metadata', mock_request_compute_metadata)
def testRetrieveProjectAndZoneFromMetadata(self):
tpu_map = {
'projects/test-project/locations/us-central1-c/nodes/test-tpu-1': {
'ipAddress': '10.1.2.3',
'port': '8470',
'health': 'HEALTHY'
}
}
cluster_resolver = resolver.TPUClusterResolver(
project=None,
zone=None,
tpu=['test-tpu-1'],
credentials=None,
service=self.mock_service_client(tpu_map=tpu_map),
coordinator_name='coordinator')
actual_cluster_spec = cluster_resolver.cluster_spec()
expected_proto = """
job {
name: 'coordinator'
tasks { key: 0 value: '10.128.1.2:%s' }
}
job {
name: 'worker'
tasks { key: 0 value: '10.1.2.3:8470' }
}
""" % cluster_resolver._coordinator_port
self._verifyClusterSpecEquality(actual_cluster_spec, str(expected_proto))
self.assertEqual(cluster_resolver.master(), 'grpc://10.1.2.3:8470')
@mock.patch.object(resolver.TPUClusterResolver,
'_request_compute_metadata', mock_request_compute_metadata)
def testRetrieveProjectAndZoneFromMetadataNoCoordinator(self):
tpu_map = {
'projects/test-project/locations/us-central1-c/nodes/test-tpu-1': {
'ipAddress': '10.1.2.3',
'port': '8470',
'health': 'HEALTHY'
}
}
cluster_resolver = resolver.TPUClusterResolver(
project=None,
zone=None,
tpu=['test-tpu-1'],
coordinator_name=None,
credentials=None,
service=self.mock_service_client(tpu_map=tpu_map))
actual_cluster_spec = cluster_resolver.cluster_spec()
expected_proto = """
job { name: 'worker' tasks { key: 0 value: '10.1.2.3:8470' } }
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
self.assertEqual(cluster_resolver.master(), 'grpc://10.1.2.3:8470')
@mock.patch.object(resolver.TPUClusterResolver,
'_request_compute_metadata', mock_request_compute_metadata)
def testNotReadyCloudTpu(self):
tpu_map = {
'projects/test-project/locations/us-central1-c/nodes/test-tpu-1': {
'ipAddress': '10.1.2.3',
'port': '8470',
'state': 'CREATING'
}
}
cluster_resolver = resolver.TPUClusterResolver(
project=None,
zone=None,
tpu='test-tpu-1',
coordinator_name=None,
credentials=None,
service=self.mock_service_client(tpu_map=tpu_map))
with self.assertRaises(RuntimeError):
cluster_resolver.cluster_spec()
def testSimpleSuccessfulRetrieval(self):
tpu_map = {
'projects/test-project/locations/us-central1-c/nodes/test-tpu-1': {
'ipAddress': '10.1.2.3',
'port': '8470',
'health': 'HEALTHY'
}
}
cluster_resolver = resolver.TPUClusterResolver(
project='test-project',
zone='us-central1-c',
tpu=['test-tpu-1'],
coordinator_name='coordinator',
coordinator_address='10.128.1.5:10203',
credentials=None,
service=self.mock_service_client(tpu_map=tpu_map))
actual_cluster_spec = cluster_resolver.cluster_spec()
expected_proto = """
job { name: 'coordinator' tasks { key: 0 value: '10.128.1.5:10203' } }
job { name: 'worker' tasks { key: 0 value: '10.1.2.3:8470' } }
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
self.assertEqual(cluster_resolver.master(), 'grpc://10.1.2.3:8470')
def testNewNetworkEndpointFormat(self):
tpu_map = {
'projects/test-project/locations/us-central1-c/nodes/test-tpu-1': {
'health': 'HEALTHY',
'networkEndpoints': [{
'ipAddress': '10.2.3.4',
'port': 8470,
}]
}
}
cluster_resolver = resolver.TPUClusterResolver(
project='test-project',
zone='us-central1-c',
tpu='test-tpu-1',
coordinator_name='coordinator',
coordinator_address='10.128.1.5:10203',
credentials=None,
service=self.mock_service_client(tpu_map=tpu_map))
actual_cluster_spec = cluster_resolver.cluster_spec()
expected_proto = """
job { name: 'coordinator' tasks { key: 0 value: '10.128.1.5:10203' } }
job { name: 'worker' tasks { key: 0 value: '10.2.3.4:8470' } }
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
self.assertEqual('grpc://10.2.3.4:8470', cluster_resolver.master())
@mock.patch.object(resolver.TPUClusterResolver,
'_request_compute_metadata', mock_request_compute_metadata)
def testPodResolution(self):
tpu_map = {
'projects/test-project/locations/us-central1-c/nodes/test-tpu-1': {
'health':
'HEALTHY',
'networkEndpoints': [
{
'ipAddress': '10.2.3.4',
'port': 8470,
},
{
'ipAddress': '10.2.3.5',
'port': 8470,
},
{
'ipAddress': '10.2.3.6',
'port': 8470,
},
{
'ipAddress': '10.2.3.7',
'port': 8470,
},
]
}
}
cluster_resolver = resolver.TPUClusterResolver(
tpu='test-tpu-1',
credentials=None,
service=self.mock_service_client(tpu_map=tpu_map),
coordinator_name='coordinator')
actual_cluster_spec = cluster_resolver.cluster_spec()
expected_proto = """
job {
name: 'coordinator',
tasks { key: 0 value: '10.128.1.2:%s'}
}
job {
name: 'worker'
tasks { key: 0 value: '10.2.3.4:8470' }
tasks { key: 1 value: '10.2.3.5:8470' }
tasks { key: 2 value: '10.2.3.6:8470' }
tasks { key: 3 value: '10.2.3.7:8470' }
}
""" % cluster_resolver._coordinator_port
self._verifyClusterSpecEquality(actual_cluster_spec, str(expected_proto))
self.assertEqual(cluster_resolver.master(), 'grpc://10.2.3.4:8470')
def testPodResolutionNoCoordinator(self):
tpu_map = {
'projects/test-project/locations/us-central1-c/nodes/test-tpu-1': {
'health':
'HEALTHY',
'networkEndpoints': [
{
'ipAddress': '10.2.3.4',
'port': 8470,
},
{
'ipAddress': '10.2.3.5',
'port': 8470,
},
{
'ipAddress': '10.2.3.6',
'port': 8470,
},
{
'ipAddress': '10.2.3.7',
'port': 8470,
},
]
}
}
cluster_resolver = resolver.TPUClusterResolver(
project='test-project',
zone='us-central1-c',
tpu='test-tpu-1',
coordinator_name=None,
credentials=None,
service=self.mock_service_client(tpu_map=tpu_map))
actual_cluster_spec = cluster_resolver.cluster_spec()
expected_proto = """
job {
name: 'worker'
tasks { key: 0 value: '10.2.3.4:8470' }
tasks { key: 1 value: '10.2.3.5:8470' }
tasks { key: 2 value: '10.2.3.6:8470' }
tasks { key: 3 value: '10.2.3.7:8470' }
}
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
self.assertEqual(cluster_resolver.master(), 'grpc://10.2.3.4:8470')
def testGetMasterNoEntries(self):
tpu_map = {}
with self.assertRaises(ValueError):
resolver.TPUClusterResolver(
project='test-project',
zone='us-central1-c',
tpu=[],
coordinator_name=None,
credentials=None,
service=self.mock_service_client(tpu_map=tpu_map))
# TODO(saeta): Convert to parameterized test when included in OSS TF.
def verifyShouldResolve(self, tpu, should_resolve):
cluster_resolver = resolver.TPUClusterResolver(
project='test-project',
zone='us-central1-c',
tpu=tpu,
coordinator_name=None,
credentials=None,
service=self.mock_service_client(tpu_map={}))
self.assertEqual(should_resolve, cluster_resolver._should_resolve(),
"TPU: '%s'" % tpu)
@mock.patch.object(resolver, 'is_running_in_gce',
mock_is_not_running_in_gce)
def testShouldResolveNoName(self):
self.verifyShouldResolve('', False)
def testShouldResolveLocal(self):
self.verifyShouldResolve('local', False)
def testShouldResolveLocalhost(self):
self.verifyShouldResolve('localhost:12345', False)
def testShouldResolveGrpc(self):
self.verifyShouldResolve('grpc://10.1.2.3:8470', False)
def testShouldResolveBns(self):
self.verifyShouldResolve('/bns/foo/bar', False)
def testShouldResolveName(self):
self.verifyShouldResolve('mytpu', True)
def testShouldResolveList(self):
self.verifyShouldResolve(['myothertpu'], True)
def testShouldResolveGrpcPrefix(self):
self.verifyShouldResolve('grpctpu', True)
def testNoCallComputeMetadata(self):
cluster_resolver = resolver.TPUClusterResolver(tpu='/bns/foo/bar')
self.assertEqual('/bns/foo/bar', cluster_resolver.master())
self.assertEqual(None, cluster_resolver.cluster_spec())
def testLocalhostMaster(self):
cluster_resolver = resolver.TPUClusterResolver(tpu='localhost:12345')
self.assertEqual('localhost:12345', cluster_resolver.master())
def testGkeEnvironmentForDonut(self):
os.environ['KUBE_GOOGLE_CLOUD_TPU_ENDPOINTS'] = 'grpc://10.120.27.5:8470'
self.assertIn('KUBE_GOOGLE_CLOUD_TPU_ENDPOINTS', os.environ)
self.assertTrue(resolver.TPUClusterResolver._in_gke())
self.assertEqual(
compat.as_bytes('grpc://10.120.27.5:8470'),
compat.as_bytes(
resolver.TPUClusterResolver._gke_endpoints()))
cluster_resolver = resolver.TPUClusterResolver()
self.assertEqual(
compat.as_bytes('grpc://10.120.27.5:8470'),
compat.as_bytes(cluster_resolver.master()))
actual_cluster_spec = cluster_resolver.cluster_spec()
expected_proto = """
job {
name: 'worker'
tasks { key: 0 value: '10.120.27.5:8470' }
}
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
del os.environ['KUBE_GOOGLE_CLOUD_TPU_ENDPOINTS']
def testGkeEnvironmentForPod(self):
os.environ['KUBE_GOOGLE_CLOUD_TPU_ENDPOINTS'] = ('grpc://10.120.27.5:8470,'
'grpc://10.120.27.6:8470,'
'grpc://10.120.27.7:8470,'
'grpc://10.120.27.8:8470')
self.assertIn('KUBE_GOOGLE_CLOUD_TPU_ENDPOINTS', os.environ)
self.assertTrue(resolver.TPUClusterResolver._in_gke())
self.assertEqual(
compat.as_bytes('grpc://10.120.27.5:8470,'
'grpc://10.120.27.6:8470,'
'grpc://10.120.27.7:8470,'
'grpc://10.120.27.8:8470'),
compat.as_bytes(
resolver.TPUClusterResolver._gke_endpoints()))
cluster_resolver = resolver.TPUClusterResolver()
self.assertEqual(
compat.as_bytes('grpc://10.120.27.5:8470'),
compat.as_bytes(cluster_resolver.master()))
actual_cluster_spec = cluster_resolver.cluster_spec()
expected_proto = """
job {
name: 'worker'
tasks { key: 0 value: '10.120.27.5:8470' }
tasks { key: 1 value: '10.120.27.6:8470' }
tasks { key: 2 value: '10.120.27.7:8470' }
tasks { key: 3 value: '10.120.27.8:8470' }
}
"""
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
del os.environ['KUBE_GOOGLE_CLOUD_TPU_ENDPOINTS']
def testEnvironmentDiscoveryUrl(self):
os.environ['TPU_API_DISCOVERY_URL'] = 'https://{api}.internal/{apiVersion}'
self.assertEqual(
'https://{api}.internal/{apiVersion}',
(resolver.TPUClusterResolver._environment_discovery_url()))
def testEnvironmentAndRpcDetectionForGoogle(self):
cluster_resolver = resolver.TPUClusterResolver(tpu='/bns/ab/cd/ef')
self.assertEqual(cluster_resolver.environment, 'google')
self.assertEqual(cluster_resolver.rpc_layer, None)
def testEnvironmentAndRpcDetectionForGrpcString(self):
cluster_resolver = resolver.TPUClusterResolver(
tpu='grpc://10.1.2.3:8470')
self.assertEqual(cluster_resolver.environment, '')
self.assertEqual(cluster_resolver.rpc_layer, 'grpc')
self.assertEqual(cluster_resolver.master(), 'grpc://10.1.2.3:8470')
def testOverrideTaskTypeAndIndexAndGetMaster(self):
tpu_map = {
'projects/test-project/locations/us-central1-c/nodes/test-tpu-1': {
'health':
'HEALTHY',
'networkEndpoints': [
{
'ipAddress': '10.2.3.4',
'port': 8470,
},
{
'ipAddress': '10.2.3.5',
'port': 8470,
},
{
'ipAddress': '10.2.3.6',
'port': 8470,
},
{
'ipAddress': '10.2.3.7',
'port': 8470,
},
]
}
}
cluster_resolver = resolver.TPUClusterResolver(
project='test-project',
zone='us-central1-c',
tpu='test-tpu-1',
coordinator_name=None,
credentials=None,
service=self.mock_service_client(tpu_map=tpu_map))
self.assertEqual(cluster_resolver.master(), 'grpc://10.2.3.4:8470')
cluster_resolver.task_type = 'worker'
cluster_resolver.task_id = 3
self.assertEqual(cluster_resolver.master(), 'grpc://10.2.3.7:8470')
self.assertEqual(
cluster_resolver.master(
task_type='worker', task_id=2, rpc_layer='test'),
'test://10.2.3.6:8470')
def testGetDeviceDictAndCoresWithTPUs(self):
device_names = [
'/job:tpu_worker/task:0/device:TPU:0',
'/job:tpu_worker/task:1/device:TPU:1',
'/job:tpu_worker/task:2/device:TPU:0',
'/job:tpu_worker/task:3/device:TPU:1',
'/job:tpu_worker/task:0/device:TPU:4',
'/job:tpu_worker/task:1/device:TPU:5',
'/job:tpu_worker/task:2/device:TPU:4',
'/job:tpu_worker/task:3/device:TPU:5',
]
device_list = [
session._DeviceAttributes(
name, 'TPU', 1024, 0) for name in device_names
]
device_details = resolver.TPUClusterResolver._get_device_dict_and_cores(
device_list)
self.assertEqual(device_details.total_cores, 8)
self.assertEqual(device_details.device_map,
{'0': ['0', '4'],
'1': ['1', '5'],
'2': ['0', '4'],
'3': ['1', '5']})
def testGetDeviceDictAndCoresWithCPUsAndGPUs(self):
device_names = [
'/job:tpu_worker/task:0/device:CPU:0',
'/job:tpu_worker/task:1/device:CPU:0',
'/job:tpu_worker/task:2/device:CPU:0',
'/job:tpu_worker/task:3/device:CPU:0',
'/job:tpu_worker/task:0/device:GPU:1',
'/job:tpu_worker/task:1/device:GPU:1',
'/job:tpu_worker/task:2/device:GPU:1',
'/job:tpu_worker/task:3/device:GPU:1',
]
device_list = [
session._DeviceAttributes(
name, 'XLA', 1024, 0) for name in device_names
]
device_dict, num_cores =\
resolver.TPUClusterResolver._get_device_dict_and_cores(device_list)
self.assertEqual(num_cores, 0)
self.assertEqual(device_dict, {})
def testVerifySameCoreCount(self):
self.assertEqual(
resolver.TPUClusterResolver
._verify_and_return_same_core_count({0: [0, 1, 2, 3, 4, 5, 6, 7]}), 8)
self.assertEqual(
resolver.TPUClusterResolver
._verify_and_return_same_core_count({
0: [0, 1],
1: [2, 3]
}), 2)
with self.assertRaises(RuntimeError):
resolver.TPUClusterResolver._verify_and_return_same_core_count(
{
0: [0],
1: [1, 2]
})
@mock.patch.object(eager.context, 'list_devices')
@mock.patch.object(session.BaseSession, 'list_devices')
@mock.patch.object(resolver, 'is_running_in_gce',
mock_is_not_running_in_gce)
def testNumAcceleratorsSuccess(self, mock_list_devices,
mock_eager_list_devices):
device_names = [
'/job:tpu_worker/task:0/device:TPU:0',
'/job:tpu_worker/task:1/device:TPU:1',
'/job:tpu_worker/task:2/device:TPU:0',
'/job:tpu_worker/task:3/device:TPU:1',
'/job:tpu_worker/task:0/device:TPU:4',
'/job:tpu_worker/task:1/device:TPU:5',
'/job:tpu_worker/task:2/device:TPU:4',
'/job:tpu_worker/task:3/device:TPU:5',
]
device_list = [
session._DeviceAttributes(
name, 'TPU', 1024, 0) for name in device_names
]
mock_eager_list_devices.return_value = device_names
mock_list_devices.return_value = device_list
cluster_resolver = resolver.TPUClusterResolver(tpu='')
self.assertEqual(cluster_resolver.num_accelerators(), {'TPU': 2})
@mock.patch.object(eager.context, 'list_devices')
@mock.patch.object(session.BaseSession, 'list_devices')
@mock.patch.object(resolver, 'is_running_in_gce',
mock_is_not_running_in_gce)
def testNumAcceleratorsRetryFailure(self, mock_list_devices,
mock_eager_list_devices):
cluster_resolver = resolver.TPUClusterResolver(tpu='')
mock_list_devices.side_effect = errors.DeadlineExceededError(
None, None, 'timeout')
mock_eager_list_devices.side_effect = errors.DeadlineExceededError(
None, None, 'timeout')
with self.assertRaises(RuntimeError):
cluster_resolver.num_accelerators()
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/tpu_cluster_resolver_test.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for Cluster Resolvers."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python import eager
from tensorflow.python.client import session
from tensorflow.python.distribute.cluster_resolver import ClusterResolver
from tensorflow.python.distribute.cluster_resolver import SimpleClusterResolver
from tensorflow.python.distribute.cluster_resolver import UnionClusterResolver
from tensorflow.python.framework import test_util
from tensorflow.python.platform import test
from tensorflow.python.training import server_lib
mock = test.mock
class MockBaseClusterResolver(ClusterResolver):
def cluster_spec(self):
return None
def master(self, task_type=None, task_id=None, rpc_layer=None):
return ""
def environment(self):
return ""
@test_util.run_all_in_graph_and_eager_modes
class BaseClusterResolverTest(test.TestCase):
@mock.patch.object(eager.context, "list_devices")
@mock.patch.object(session.BaseSession, "list_devices")
def testNumAcceleratorsSuccess(self, mock_list_devices,
mock_eager_list_devices):
device_names = [
"/job:worker/task:0/device:GPU:0",
"/job:worker/task:0/device:GPU:1",
"/job:worker/task:0/device:GPU:2",
"/job:worker/task:0/device:GPU:3",
]
device_list = [
session._DeviceAttributes(name, "GPU", 1024, 0)
for name in device_names
]
mock_eager_list_devices.return_value = device_names
mock_list_devices.return_value = device_list
resolver = MockBaseClusterResolver()
self.assertEqual(resolver.num_accelerators(), {"GPU": 4})
@mock.patch.object(eager.context, "list_devices")
@mock.patch.object(session.BaseSession, "list_devices")
def testNumAcceleratorsMultiDeviceSuccess(self, mock_list_devices,
mock_eager_list_devices):
device_names = [
"/job:worker/task:0/device:TPU:0",
"/job:worker/task:0/device:TPU:1",
"/job:worker/task:0/device:TPU:2",
"/job:worker/task:0/device:TPU:3",
"/job:worker/task:0/device:GPU:0",
"/job:worker/task:0/device:GPU:1",
"/job:worker/task:0/device:GPU:2",
"/job:worker/task:0/device:GPU:3",
]
device_list = [
session._DeviceAttributes(name, name[26:29], 1024, 0)
for name in device_names
]
mock_eager_list_devices.return_value = device_names
mock_list_devices.return_value = device_list
resolver = MockBaseClusterResolver()
self.assertEqual(resolver.num_accelerators(), {"TPU": 4, "GPU": 4})
@mock.patch.object(eager.context, "list_devices")
@mock.patch.object(session.BaseSession, "list_devices")
def testNumAcceleratorsFilterTasks(self, mock_list_devices,
mock_eager_list_devices):
device_names = [
"/job:worker1/task:0/device:TPU:0",
"/job:worker1/task:0/device:TPU:1",
"/job:worker1/task:0/device:GPU:0",
"/job:worker1/task:0/device:GPU:1",
"/job:worker2/task:1/device:TPU:2",
"/job:worker2/task:2/device:TPU:3",
"/job:worker2/task:3/device:GPU:2",
"/job:worker2/task:4/device:GPU:3",
]
device_list = [
session._DeviceAttributes(name, name[27:30], 1024, 0)
for name in device_names
]
mock_eager_list_devices.return_value = device_names
mock_list_devices.return_value = device_list
resolver = MockBaseClusterResolver()
self.assertEqual(resolver.num_accelerators(task_type="worker1", task_id=0),
{"TPU": 2, "GPU": 2})
self.assertEqual(resolver.num_accelerators(task_type="worker2", task_id=3),
{"GPU": 1})
self.assertEqual(resolver.num_accelerators(task_type="worker2", task_id=4),
{"GPU": 1})
class UnionClusterResolverTest(test.TestCase):
# TODO(frankchn): Transform to parameterized test after it is included in the
# TF open source codebase.
def _verifyClusterSpecEquality(self, cluster_spec, expected_proto):
self.assertProtoEquals(expected_proto, cluster_spec.as_cluster_def())
self.assertProtoEquals(
expected_proto, server_lib.ClusterSpec(cluster_spec).as_cluster_def())
self.assertProtoEquals(
expected_proto,
server_lib.ClusterSpec(cluster_spec.as_cluster_def()).as_cluster_def())
self.assertProtoEquals(
expected_proto,
server_lib.ClusterSpec(cluster_spec.as_dict()).as_cluster_def())
def testSingleClusterResolver(self):
base_cluster_spec = server_lib.ClusterSpec({
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
})
simple_resolver = SimpleClusterResolver(base_cluster_spec)
union_resolver = UnionClusterResolver(simple_resolver)
expected_proto = """
job { name: 'ps' tasks { key: 0 value: 'ps0:2222' }
tasks { key: 1 value: 'ps1:2222' } }
job { name: 'worker' tasks { key: 0 value: 'worker0:2222' }
tasks { key: 1 value: 'worker1:2222' }
tasks { key: 2 value: 'worker2:2222' } }
"""
actual_cluster_spec = union_resolver.cluster_spec()
self._verifyClusterSpecEquality(actual_cluster_spec, expected_proto)
def testInitSimpleClusterResolver(self):
base_cluster_spec = server_lib.ClusterSpec({
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
})
simple_resolver = SimpleClusterResolver(base_cluster_spec, task_type="ps",
task_id=1, environment="cloud",
num_accelerators={"GPU": 8},
rpc_layer="grpc")
self.assertEqual(simple_resolver.task_type, "ps")
self.assertEqual(simple_resolver.task_id, 1)
self.assertEqual(simple_resolver.environment, "cloud")
self.assertEqual(simple_resolver.num_accelerators(), {"GPU": 8})
self.assertEqual(simple_resolver.rpc_layer, "grpc")
def testOverrideSimpleClusterResolver(self):
base_cluster_spec = server_lib.ClusterSpec({
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
})
simple_resolver = SimpleClusterResolver(base_cluster_spec, task_type="ps",
task_id=1, environment="cloud",
num_accelerators={"GPU": 8},
rpc_layer="grpc")
simple_resolver.task_type = "worker"
simple_resolver.task_id = 2
simple_resolver.rpc_layer = "http"
self.assertEqual(simple_resolver.task_type, "worker")
self.assertEqual(simple_resolver.task_id, 2)
self.assertEqual(simple_resolver.rpc_layer, "http")
def testSimpleOverrideMasterWithTaskIndexZero(self):
base_cluster_spec = server_lib.ClusterSpec({
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
})
simple_resolver = SimpleClusterResolver(base_cluster_spec)
actual_master = simple_resolver.master("worker", 0, rpc_layer="grpc")
self.assertEqual(actual_master, "grpc://worker0:2222")
def testSimpleOverrideMasterWithRpcLayer(self):
base_cluster_spec = server_lib.ClusterSpec({
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
})
simple_resolver = SimpleClusterResolver(base_cluster_spec)
actual_master = simple_resolver.master("worker", 2, rpc_layer="grpc")
self.assertEqual(actual_master, "grpc://worker2:2222")
def testSimpleOverrideMaster(self):
base_cluster_spec = server_lib.ClusterSpec({
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
})
simple_resolver = SimpleClusterResolver(base_cluster_spec)
actual_master = simple_resolver.master("worker", 2)
self.assertEqual(actual_master, "worker2:2222")
def testUnionClusterResolverGetProperties(self):
cluster_spec_1 = server_lib.ClusterSpec({
"ps": ["ps0:2222", "ps1:2222"],
"worker": ["worker0:2222", "worker1:2222", "worker2:2222"]
})
resolver1 = SimpleClusterResolver(cluster_spec_1, task_type="ps",
task_id=1, environment="cloud",
num_accelerators={"GPU": 8},
rpc_layer="grpc")
cluster_spec_2 = server_lib.ClusterSpec({
"ps": ["ps2:2222", "ps3:2222"],
"worker": ["worker3:2222", "worker4:2222", "worker5:2222"]
})
resolver2 = SimpleClusterResolver(cluster_spec_2, task_type="worker",
task_id=2, environment="local",
num_accelerators={"GPU": 16},
rpc_layer="http")
union_resolver = UnionClusterResolver(resolver1, resolver2)
self.assertEqual(union_resolver.task_type, "ps")
self.assertEqual(union_resolver.task_id, 1)
self.assertEqual(union_resolver.environment, "cloud")
self.assertEqual(union_resolver.num_accelerators(), {"GPU": 8})
self.assertEqual(union_resolver.rpc_layer, "grpc")
union_resolver.task_type = "worker"
union_resolver.task_id = 2
union_resolver.rpc_layer = "http"
self.assertEqual(union_resolver.task_type, "worker")
self.assertEqual(union_resolver.task_id, 2)
self.assertEqual(union_resolver.rpc_layer, "http")
def testTwoNonOverlappingJobMergedClusterResolver(self):
cluster_spec_1 = server_lib.ClusterSpec({
"ps": [
"ps0:2222",
"ps1:2222"
]
})
cluster_spec_2 = server_lib.ClusterSpec({
"worker": [
"worker0:2222",
"worker1:2222",
"worker2:2222"
]
})
cluster_resolver_1 = SimpleClusterResolver(cluster_spec_1)
cluster_resolver_2 = SimpleClusterResolver(cluster_spec_2)
union_cluster = UnionClusterResolver(cluster_resolver_1, cluster_resolver_2)
cluster_spec = union_cluster.cluster_spec()
expected_proto = """
job { name: 'ps' tasks { key: 0 value: 'ps0:2222' }
tasks { key: 1 value: 'ps1:2222' } }
job { name: 'worker' tasks { key: 0 value: 'worker0:2222' }
tasks { key: 1 value: 'worker1:2222' }
tasks { key: 2 value: 'worker2:2222' } }
"""
self._verifyClusterSpecEquality(cluster_spec, expected_proto)
def testMergedClusterResolverMaster(self):
cluster_spec_1 = server_lib.ClusterSpec({
"ps": [
"ps0:2222",
"ps1:2222"
]
})
cluster_spec_2 = server_lib.ClusterSpec({
"worker": [
"worker0:2222",
"worker1:2222",
"worker2:2222"
]
})
cluster_resolver_1 = SimpleClusterResolver(cluster_spec_1)
cluster_resolver_2 = SimpleClusterResolver(cluster_spec_2)
union_cluster = UnionClusterResolver(cluster_resolver_1, cluster_resolver_2)
unspecified_master = union_cluster.master()
self.assertEqual(unspecified_master, "")
specified_master = union_cluster.master("worker", 1)
self.assertEqual(specified_master, "worker1:2222")
rpc_master = union_cluster.master("worker", 1, rpc_layer="grpc")
self.assertEqual(rpc_master, "grpc://worker1:2222")
def testOverlappingJobMergedClusterResolver(self):
cluster_spec_1 = server_lib.ClusterSpec({
"worker": [
"worker4:2222",
"worker5:2222"
]
})
cluster_spec_2 = server_lib.ClusterSpec({
"worker": [
"worker0:2222",
"worker1:2222",
"worker2:2222"
]
})
cluster_resolver_1 = SimpleClusterResolver(cluster_spec_1)
cluster_resolver_2 = SimpleClusterResolver(cluster_spec_2)
union_cluster = UnionClusterResolver(cluster_resolver_1, cluster_resolver_2)
cluster_spec = union_cluster.cluster_spec()
expected_proto = """
job { name: 'worker' tasks { key: 0 value: 'worker4:2222' }
tasks { key: 1 value: 'worker5:2222' }
tasks { key: 2 value: 'worker0:2222' }
tasks { key: 3 value: 'worker1:2222' }
tasks { key: 4 value: 'worker2:2222' } }
"""
self._verifyClusterSpecEquality(cluster_spec, expected_proto)
def testOverlappingSparseJobMergedClusterResolverThrowError(self):
cluster_spec_1 = server_lib.ClusterSpec({
"worker": {
7: "worker4:2222",
9: "worker5:2222"
}
})
cluster_spec_2 = server_lib.ClusterSpec({
"worker": {
3: "worker0:2222",
6: "worker1:2222",
7: "worker2:2222"
}
})
cluster_resolver_1 = SimpleClusterResolver(cluster_spec_1)
cluster_resolver_2 = SimpleClusterResolver(cluster_spec_2)
union_cluster = UnionClusterResolver(cluster_resolver_1, cluster_resolver_2)
self.assertRaises(KeyError, union_cluster.cluster_spec)
def testOverlappingDictAndListThrowError(self):
cluster_spec_1 = server_lib.ClusterSpec({
"worker": [
"worker4:2222",
"worker5:2222"
]
})
cluster_spec_2 = server_lib.ClusterSpec({
"worker": {
1: "worker0:2222",
2: "worker1:2222",
3: "worker2:2222"
}
})
cluster_resolver_1 = SimpleClusterResolver(cluster_spec_1)
cluster_resolver_2 = SimpleClusterResolver(cluster_spec_2)
union_cluster = UnionClusterResolver(cluster_resolver_1, cluster_resolver_2)
self.assertRaises(KeyError, union_cluster.cluster_spec)
def testOverlappingJobNonOverlappingKey(self):
cluster_spec_1 = server_lib.ClusterSpec({
"worker": {
5: "worker4:2222",
9: "worker5:2222"
}
})
cluster_spec_2 = server_lib.ClusterSpec({
"worker": {
3: "worker0:2222",
6: "worker1:2222",
7: "worker2:2222"
}
})
cluster_resolver_1 = SimpleClusterResolver(cluster_spec_1)
cluster_resolver_2 = SimpleClusterResolver(cluster_spec_2)
union_cluster = UnionClusterResolver(cluster_resolver_1, cluster_resolver_2)
cluster_spec = union_cluster.cluster_spec()
expected_proto = """
job { name: 'worker' tasks { key: 3 value: 'worker0:2222' }
tasks { key: 5 value: 'worker4:2222' }
tasks { key: 6 value: 'worker1:2222' }
tasks { key: 7 value: 'worker2:2222' }
tasks { key: 9 value: 'worker5:2222' }}
"""
self._verifyClusterSpecEquality(cluster_spec, expected_proto)
def testMixedModeNonOverlappingKey(self):
cluster_spec_1 = server_lib.ClusterSpec({
"worker": [
"worker4:2222",
"worker5:2222"
]
})
cluster_spec_2 = server_lib.ClusterSpec({
"worker": {
3: "worker0:2222",
6: "worker1:2222",
7: "worker2:2222"
}
})
cluster_resolver_1 = SimpleClusterResolver(cluster_spec_1)
cluster_resolver_2 = SimpleClusterResolver(cluster_spec_2)
union_cluster = UnionClusterResolver(cluster_resolver_1, cluster_resolver_2)
cluster_spec = union_cluster.cluster_spec()
expected_proto = """
job { name: 'worker' tasks { key: 0 value: 'worker4:2222' }
tasks { key: 1 value: 'worker5:2222' }
tasks { key: 3 value: 'worker0:2222' }
tasks { key: 6 value: 'worker1:2222' }
tasks { key: 7 value: 'worker2:2222' }}
"""
self._verifyClusterSpecEquality(cluster_spec, expected_proto)
def testRetainSparseJobWithNoMerging(self):
base_cluster_spec = server_lib.ClusterSpec({
"worker": {
1: "worker0:2222",
3: "worker1:2222",
5: "worker2:2222"
}
})
base_cluster_resolver = SimpleClusterResolver(base_cluster_spec)
union_cluster = UnionClusterResolver(base_cluster_resolver)
cluster_spec = union_cluster.cluster_spec()
expected_proto = """
job { name: 'worker' tasks { key: 1 value: 'worker0:2222' }
tasks { key: 3 value: 'worker1:2222' }
tasks { key: 5 value: 'worker2:2222' } }
"""
self._verifyClusterSpecEquality(cluster_spec, expected_proto)
# TODO(saeta): Include tests for master resolution
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/distribute/cluster_resolver/cluster_resolver_test.py
|
tensorflow-master
|
tensorflow/python/lib/__init__.py
|
|
tensorflow-master
|
tensorflow/python/lib/core/__init__.py
|
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Test cases for the bfloat16 Python type."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import math
import numpy as np
# pylint: disable=unused-import,g-bad-import-order
from tensorflow.python import pywrap_tensorflow
from tensorflow.python.framework import dtypes
from tensorflow.python.platform import test
bfloat16 = pywrap_tensorflow.TF_bfloat16_type()
class Bfloat16Test(test.TestCase):
def float_values(self):
"""Returns values that should round trip exactly to float and back."""
epsilon = float.fromhex("1.0p-7")
return [
0.0, 1.0, -1, 0.5, -0.5, epsilon, 1.0 + epsilon, 1.0 - epsilon,
-1.0 - epsilon, -1.0 + epsilon, 3.5, 42.0, 255.0, 256.0,
float("inf"), float("-inf"), float("nan")]
def _assertFloatIdentical(self, v, w):
if math.isnan(v):
self.assertTrue(math.isnan(w))
else:
self.assertEqual(v, w)
def testRoundTripToFloat(self):
for v in self.float_values():
self._assertFloatIdentical(v, float(bfloat16(v)))
def testRoundTripToInt(self):
for v in [-256, -255, -34, -2, -1, 0, 1, 2, 10, 47, 128, 255, 256, 512]:
self.assertEqual(v, int(bfloat16(v)))
def testStr(self):
self.assertEqual("0", str(bfloat16(0.0)))
self.assertEqual("1", str(bfloat16(1.0)))
self.assertEqual("-3.5", str(bfloat16(-3.5)))
self.assertEqual("0.0078125", str(bfloat16(float.fromhex("1.0p-7"))))
self.assertEqual("inf", str(bfloat16(float("inf"))))
self.assertEqual("-inf", str(bfloat16(float("-inf"))))
self.assertEqual("nan", str(bfloat16(float("nan"))))
def testRepr(self):
self.assertEqual("bfloat16(0)", repr(bfloat16(0)))
self.assertEqual("bfloat16(1)", repr(bfloat16(1)))
self.assertEqual("bfloat16(-3.5)", repr(bfloat16(-3.5)))
self.assertEqual("bfloat16(0.0078125)",
repr(bfloat16(float.fromhex("1.0p-7"))))
self.assertEqual("bfloat16(inf)", repr(bfloat16(float("inf"))))
self.assertEqual("bfloat16(-inf)", repr(bfloat16(float("-inf"))))
self.assertEqual("bfloat16(nan)", repr(bfloat16(float("nan"))))
def testHash(self):
self.assertEqual(0, hash(bfloat16(0.0)))
self.assertEqual(0x3f80, hash(bfloat16(1.0)))
self.assertEqual(0x7fc0, hash(bfloat16(float("nan"))))
# Tests for Python operations
def testNegate(self):
for v in self.float_values():
self._assertFloatIdentical(-v, float(-bfloat16(v)))
def testAdd(self):
self._assertFloatIdentical(0, float(bfloat16(0) + bfloat16(0)))
self._assertFloatIdentical(1, float(bfloat16(1) + bfloat16(0)))
self._assertFloatIdentical(0, float(bfloat16(1) + bfloat16(-1)))
self._assertFloatIdentical(5.5, float(bfloat16(2) + bfloat16(3.5)))
self._assertFloatIdentical(1.25, float(bfloat16(3.5) + bfloat16(-2.25)))
self._assertFloatIdentical(float("inf"),
float(bfloat16(float("inf")) + bfloat16(-2.25)))
self._assertFloatIdentical(float("-inf"),
float(bfloat16(float("-inf")) + bfloat16(-2.25)))
self.assertTrue(math.isnan(float(bfloat16(3.5) + bfloat16(float("nan")))))
def testSub(self):
self._assertFloatIdentical(0, float(bfloat16(0) - bfloat16(0)))
self._assertFloatIdentical(1, float(bfloat16(1) - bfloat16(0)))
self._assertFloatIdentical(2, float(bfloat16(1) - bfloat16(-1)))
self._assertFloatIdentical(-1.5, float(bfloat16(2) - bfloat16(3.5)))
self._assertFloatIdentical(5.75, float(bfloat16(3.5) - bfloat16(-2.25)))
self._assertFloatIdentical(float("-inf"),
float(bfloat16(-2.25) - bfloat16(float("inf"))))
self._assertFloatIdentical(float("inf"),
float(bfloat16(-2.25) - bfloat16(float("-inf"))))
self.assertTrue(math.isnan(float(bfloat16(3.5) - bfloat16(float("nan")))))
def testMul(self):
self._assertFloatIdentical(0, float(bfloat16(0) * bfloat16(0)))
self._assertFloatIdentical(0, float(bfloat16(1) * bfloat16(0)))
self._assertFloatIdentical(-1, float(bfloat16(1) * bfloat16(-1)))
self._assertFloatIdentical(-7.875, float(bfloat16(3.5) * bfloat16(-2.25)))
self._assertFloatIdentical(float("-inf"),
float(bfloat16(float("inf")) * bfloat16(-2.25)))
self._assertFloatIdentical(float("inf"),
float(bfloat16(float("-inf")) * bfloat16(-2.25)))
self.assertTrue(math.isnan(float(bfloat16(3.5) * bfloat16(float("nan")))))
def testDiv(self):
self.assertTrue(math.isnan(float(bfloat16(0) / bfloat16(0))))
self._assertFloatIdentical(float("inf"), float(bfloat16(1) / bfloat16(0)))
self._assertFloatIdentical(-1, float(bfloat16(1) / bfloat16(-1)))
self._assertFloatIdentical(-1.75, float(bfloat16(3.5) / bfloat16(-2)))
self._assertFloatIdentical(float("-inf"),
float(bfloat16(float("inf")) / bfloat16(-2.25)))
self._assertFloatIdentical(float("inf"),
float(bfloat16(float("-inf")) / bfloat16(-2.25)))
self.assertTrue(math.isnan(float(bfloat16(3.5) / bfloat16(float("nan")))))
def testLess(self):
for v in self.float_values():
for w in self.float_values():
self.assertEqual(v < w, bfloat16(v) < bfloat16(w))
def testLessEqual(self):
for v in self.float_values():
for w in self.float_values():
self.assertEqual(v <= w, bfloat16(v) <= bfloat16(w))
def testGreater(self):
for v in self.float_values():
for w in self.float_values():
self.assertEqual(v > w, bfloat16(v) > bfloat16(w))
def testGreaterEqual(self):
for v in self.float_values():
for w in self.float_values():
self.assertEqual(v >= w, bfloat16(v) >= bfloat16(w))
def testEqual(self):
for v in self.float_values():
for w in self.float_values():
self.assertEqual(v == w, bfloat16(v) == bfloat16(w))
def testNotEqual(self):
for v in self.float_values():
for w in self.float_values():
self.assertEqual(v != w, bfloat16(v) != bfloat16(w))
def testNan(self):
a = np.isnan(bfloat16(float("nan")))
self.assertTrue(a)
np.testing.assert_allclose(np.array([1.0, a]), np.array([1.0, a]))
a = np.array(
[bfloat16(1.34375),
bfloat16(1.4375),
bfloat16(float("nan"))],
dtype=dtypes.bfloat16.as_numpy_dtype)
b = np.array(
[bfloat16(1.3359375),
bfloat16(1.4375),
bfloat16(float("nan"))],
dtype=dtypes.bfloat16.as_numpy_dtype)
np.testing.assert_allclose(
a, b, rtol=0.1, atol=0.1, equal_nan=True, err_msg="", verbose=True)
class Bfloat16NumPyTest(test.TestCase):
def testDtype(self):
self.assertEqual(bfloat16, np.dtype(bfloat16))
def testArray(self):
x = np.array([[1, 2, 3]], dtype=bfloat16)
self.assertEqual(bfloat16, x.dtype)
self.assertEqual("[[bfloat16(1) bfloat16(2) bfloat16(3)]]", str(x))
self.assertAllEqual(x, x)
self.assertAllClose(x, x)
self.assertTrue((x == x).all())
def testComparisons(self):
x = np.array([401408, 7, -32], dtype=np.float32)
bx = x.astype(bfloat16)
y = np.array([82432, 7, 0], dtype=np.float32)
by = y.astype(bfloat16)
self.assertAllEqual(x == y, bx == by)
self.assertAllEqual(x != y, bx != by)
self.assertAllEqual(x < y, bx < by)
self.assertAllEqual(x > y, bx > by)
self.assertAllEqual(x <= y, bx <= by)
self.assertAllEqual(x >= y, bx >= by)
def testEqual2(self):
a = np.array([401408], bfloat16)
b = np.array([82432], bfloat16)
self.assertFalse(a.__eq__(b))
def testCasts(self):
for dtype in [
np.float16, np.float32, np.float64, np.int32, np.int64,
np.complex64, np.complex128]:
x = np.array([[1, 2, 3]], dtype=dtype)
y = x.astype(bfloat16)
z = y.astype(dtype)
self.assertTrue(np.all(x == y))
self.assertEqual(bfloat16, y.dtype)
self.assertTrue(np.all(x == z))
self.assertEqual(dtype, z.dtype)
def testConformNumpyComplex(self):
for dtype in [np.complex64, np.complex128]:
x = np.array([1.1, 2.2 + 2.2j, 3.3], dtype=dtype)
y_np = x.astype(np.float32)
y_tf = x.astype(bfloat16)
self.assertAllClose(y_np, y_tf, atol=2e-2)
z_np = y_np.astype(dtype)
z_tf = y_tf.astype(dtype)
self.assertAllClose(z_np, z_tf, atol=2e-2)
def testAdd(self):
x = np.array([[1, 2, 3]], dtype=bfloat16)
y = np.array([[4, 5, 6]], dtype=bfloat16)
self.assertAllClose(np.array([[5, 7, 9]]), x + y)
def testLogSumExp(self):
x = np.array([[1, 2, 3]], dtype=np.float32)
y = np.array([[4, 5, 6]], dtype=np.float32)
self.assertAllClose(np.logaddexp(x, y),
np.logaddexp(x.astype(bfloat16), y.astype(bfloat16)),
atol=2e-2)
def testArange(self):
self.assertAllEqual(
np.arange(100, dtype=np.float32).astype(bfloat16),
np.arange(100, dtype=bfloat16))
self.assertAllEqual(
np.arange(-10.5, 7.8, 0.5, dtype=np.float32).astype(bfloat16),
np.arange(-10.5, 7.8, 0.5, dtype=bfloat16))
self.assertAllEqual(
np.arange(-0., -7., -0.25, dtype=np.float32).astype(bfloat16),
np.arange(-0., -7., -0.25, dtype=bfloat16))
self.assertAllEqual(
np.arange(-16384., 16384., 64., dtype=np.float32).astype(bfloat16),
np.arange(-16384., 16384., 64., dtype=bfloat16))
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/lib/core/bfloat16_test.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Python functions for directly manipulating TFRecord-formatted files."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
# go/tf-wildcard-import
# pylint: disable=wildcard-import
from tensorflow.python.lib.io.tf_record import *
# pylint: enable=wildcard-import
|
tensorflow-master
|
tensorflow/python/lib/io/python_io.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for tf_record.TFRecordWriter and tf_record.tf_record_iterator."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import gzip
import os
import random
import string
import zlib
import six
from tensorflow.python.framework import errors_impl
from tensorflow.python.lib.io import tf_record
from tensorflow.python.platform import test
from tensorflow.python.util import compat
prefix_path = "third_party/tensorflow/core/lib"
# pylint: disable=invalid-name
TFRecordCompressionType = tf_record.TFRecordCompressionType
# pylint: enable=invalid-name
# Edgar Allan Poe's 'Eldorado'
_TEXT = b"""Gaily bedight,
A gallant knight,
In sunshine and in shadow,
Had journeyed long,
Singing a song,
In search of Eldorado.
But he grew old
This knight so bold
And o'er his heart a shadow
Fell as he found
No spot of ground
That looked like Eldorado.
And, as his strength
Failed him at length,
He met a pilgrim shadow
'Shadow,' said he,
'Where can it be
This land of Eldorado?'
'Over the Mountains
Of the Moon'
Down the Valley of the Shadow,
Ride, boldly ride,'
The shade replied,
'If you seek for Eldorado!'
"""
class TFCompressionTestCase(test.TestCase):
def setUp(self):
super(TFCompressionTestCase, self).setUp()
self._num_files = 2
self._num_records = 7
def _Record(self, f, r):
return compat.as_bytes("Record %d of file %d" % (r, f))
def _CreateFiles(self, options=None, prefix=""):
filenames = []
for i in range(self._num_files):
name = prefix + "tfrecord.%d.txt" % i
records = [self._Record(i, j) for j in range(self._num_records)]
fn = self._WriteRecordsToFile(records, name, options)
filenames.append(fn)
return filenames
def _WriteRecordsToFile(self, records, name="tfrecord", options=None):
fn = os.path.join(self.get_temp_dir(), name)
with tf_record.TFRecordWriter(fn, options=options) as writer:
for r in records:
writer.write(r)
return fn
def _ZlibCompressFile(self, infile, name="tfrecord.z"):
# zlib compress the file and write compressed contents to file.
with open(infile, "rb") as f:
cdata = zlib.compress(f.read())
zfn = os.path.join(self.get_temp_dir(), name)
with open(zfn, "wb") as f:
f.write(cdata)
return zfn
def _GzipCompressFile(self, infile, name="tfrecord.gz"):
# gzip compress the file and write compressed contents to file.
with open(infile, "rb") as f:
cdata = f.read()
gzfn = os.path.join(self.get_temp_dir(), name)
with gzip.GzipFile(gzfn, "wb") as f:
f.write(cdata)
return gzfn
def _ZlibDecompressFile(self, infile, name="tfrecord"):
with open(infile, "rb") as f:
cdata = zlib.decompress(f.read())
fn = os.path.join(self.get_temp_dir(), name)
with open(fn, "wb") as f:
f.write(cdata)
return fn
def _GzipDecompressFile(self, infile, name="tfrecord"):
with gzip.GzipFile(infile, "rb") as f:
cdata = f.read()
fn = os.path.join(self.get_temp_dir(), name)
with open(fn, "wb") as f:
f.write(cdata)
return fn
class TFRecordWriterTest(TFCompressionTestCase):
def _AssertFilesEqual(self, a, b, equal):
for an, bn in zip(a, b):
with open(an, "rb") as af, open(bn, "rb") as bf:
if equal:
self.assertEqual(af.read(), bf.read())
else:
self.assertNotEqual(af.read(), bf.read())
def _CompressionSizeDelta(self, records, options_a, options_b):
"""Validate compression with options_a and options_b and return size delta.
Compress records with options_a and options_b. Uncompress both compressed
files and assert that the contents match the original records. Finally
calculate how much smaller the file compressed with options_a was than the
file compressed with options_b.
Args:
records: The records to compress
options_a: First set of options to compress with, the baseline for size.
options_b: Second set of options to compress with.
Returns:
The difference in file size when using options_a vs options_b. A positive
value means options_a was a better compression than options_b. A negative
value means options_b had better compression than options_a.
"""
fn_a = self._WriteRecordsToFile(records, "tfrecord_a", options=options_a)
test_a = list(tf_record.tf_record_iterator(fn_a, options=options_a))
self.assertEqual(records, test_a, options_a)
fn_b = self._WriteRecordsToFile(records, "tfrecord_b", options=options_b)
test_b = list(tf_record.tf_record_iterator(fn_b, options=options_b))
self.assertEqual(records, test_b, options_b)
# Negative number => better compression.
return os.path.getsize(fn_a) - os.path.getsize(fn_b)
def testWriteReadZLibFiles(self):
# Write uncompressed then compress manually.
options = tf_record.TFRecordOptions(TFRecordCompressionType.NONE)
files = self._CreateFiles(options, prefix="uncompressed")
zlib_files = [
self._ZlibCompressFile(fn, "tfrecord_%s.z" % i)
for i, fn in enumerate(files)
]
self._AssertFilesEqual(files, zlib_files, False)
# Now write compressd and verify same.
options = tf_record.TFRecordOptions(TFRecordCompressionType.ZLIB)
compressed_files = self._CreateFiles(options, prefix="compressed")
self._AssertFilesEqual(compressed_files, zlib_files, True)
# Decompress compress and verify same.
uncompressed_files = [
self._ZlibDecompressFile(fn, "tfrecord_%s.z" % i)
for i, fn in enumerate(compressed_files)
]
self._AssertFilesEqual(uncompressed_files, files, True)
def testWriteReadGzipFiles(self):
# Write uncompressed then compress manually.
options = tf_record.TFRecordOptions(TFRecordCompressionType.NONE)
files = self._CreateFiles(options, prefix="uncompressed")
gzip_files = [
self._GzipCompressFile(fn, "tfrecord_%s.gz" % i)
for i, fn in enumerate(files)
]
self._AssertFilesEqual(files, gzip_files, False)
# Now write compressd and verify same.
options = tf_record.TFRecordOptions(TFRecordCompressionType.GZIP)
compressed_files = self._CreateFiles(options, prefix="compressed")
# Note: Gzips written by TFRecordWriter add 'tfrecord_0' so
# compressed_files can't be compared with gzip_files
# Decompress compress and verify same.
uncompressed_files = [
self._GzipDecompressFile(fn, "tfrecord_%s.gz" % i)
for i, fn in enumerate(compressed_files)
]
self._AssertFilesEqual(uncompressed_files, files, True)
def testNoCompressionType(self):
self.assertEqual(
"",
tf_record.TFRecordOptions.get_compression_type_string(
tf_record.TFRecordOptions()))
self.assertEqual(
"",
tf_record.TFRecordOptions.get_compression_type_string(
tf_record.TFRecordOptions("")))
with self.assertRaises(ValueError):
tf_record.TFRecordOptions(5)
with self.assertRaises(ValueError):
tf_record.TFRecordOptions("BZ2")
def testZlibCompressionType(self):
zlib_t = tf_record.TFRecordCompressionType.ZLIB
self.assertEqual(
"ZLIB",
tf_record.TFRecordOptions.get_compression_type_string(
tf_record.TFRecordOptions("ZLIB")))
self.assertEqual(
"ZLIB",
tf_record.TFRecordOptions.get_compression_type_string(
tf_record.TFRecordOptions(zlib_t)))
self.assertEqual(
"ZLIB",
tf_record.TFRecordOptions.get_compression_type_string(
tf_record.TFRecordOptions(tf_record.TFRecordOptions(zlib_t))))
def testCompressionOptions(self):
# Create record with mix of random and repeated data to test compression on.
rnd = random.Random(123)
random_record = compat.as_bytes(
"".join(rnd.choice(string.digits) for _ in range(10000)))
repeated_record = compat.as_bytes(_TEXT)
for _ in range(10000):
start_i = rnd.randint(0, len(_TEXT))
length = rnd.randint(10, 200)
repeated_record += _TEXT[start_i:start_i + length]
records = [random_record, repeated_record, random_record]
tests = [
("compression_level", 2, -1), # Lower compression is worse.
("compression_level", 6, 0), # Default compression_level is equal.
("flush_mode", zlib.Z_FULL_FLUSH, 1), # A few less bytes.
("flush_mode", zlib.Z_NO_FLUSH, 0), # NO_FLUSH is the default.
("input_buffer_size", 4096, 0), # Increases time not size.
("output_buffer_size", 4096, 0), # Increases time not size.
("window_bits", 8, -1), # Smaller than default window increases size.
("compression_strategy", zlib.Z_HUFFMAN_ONLY, -1), # Worse.
("compression_strategy", zlib.Z_FILTERED, -1), # Worse.
]
compression_type = tf_record.TFRecordCompressionType.ZLIB
options_a = tf_record.TFRecordOptions(compression_type)
for prop, value, delta_sign in tests:
options_b = tf_record.TFRecordOptions(
compression_type=compression_type, **{prop: value})
delta = self._CompressionSizeDelta(records, options_a, options_b)
self.assertTrue(
delta == 0 if delta_sign == 0 else delta // delta_sign > 0,
"Setting {} = {}, file was {} smaller didn't match sign of {}".format(
prop, value, delta, delta_sign))
class TFRecordWriterZlibTest(TFCompressionTestCase):
def testZLibFlushRecord(self):
original = [b"small record"]
fn = self._WriteRecordsToFile(original, "small_record")
with open(fn, "rb") as h:
buff = h.read()
# creating more blocks and trailing blocks shouldn't break reads
compressor = zlib.compressobj(9, zlib.DEFLATED, zlib.MAX_WBITS)
output = b""
for c in buff:
if isinstance(c, int):
c = six.int2byte(c)
output += compressor.compress(c)
output += compressor.flush(zlib.Z_FULL_FLUSH)
output += compressor.flush(zlib.Z_FULL_FLUSH)
output += compressor.flush(zlib.Z_FULL_FLUSH)
output += compressor.flush(zlib.Z_FINISH)
# overwrite the original file with the compressed data
with open(fn, "wb") as h:
h.write(output)
options = tf_record.TFRecordOptions(TFRecordCompressionType.ZLIB)
actual = list(tf_record.tf_record_iterator(fn, options=options))
self.assertEqual(actual, original)
def testZlibReadWrite(self):
"""Verify that files produced are zlib compatible."""
original = [b"foo", b"bar"]
fn = self._WriteRecordsToFile(original, "zlib_read_write.tfrecord")
zfn = self._ZlibCompressFile(fn, "zlib_read_write.tfrecord.z")
# read the compressed contents and verify.
options = tf_record.TFRecordOptions(TFRecordCompressionType.ZLIB)
actual = list(tf_record.tf_record_iterator(zfn, options=options))
self.assertEqual(actual, original)
def testZlibReadWriteLarge(self):
"""Verify that writing large contents also works."""
# Make it large (about 5MB)
original = [_TEXT * 10240]
fn = self._WriteRecordsToFile(original, "zlib_read_write_large.tfrecord")
zfn = self._ZlibCompressFile(fn, "zlib_read_write_large.tfrecord.z")
options = tf_record.TFRecordOptions(TFRecordCompressionType.ZLIB)
actual = list(tf_record.tf_record_iterator(zfn, options=options))
self.assertEqual(actual, original)
def testGzipReadWrite(self):
"""Verify that files produced are gzip compatible."""
original = [b"foo", b"bar"]
fn = self._WriteRecordsToFile(original, "gzip_read_write.tfrecord")
gzfn = self._GzipCompressFile(fn, "tfrecord.gz")
options = tf_record.TFRecordOptions(TFRecordCompressionType.GZIP)
actual = list(tf_record.tf_record_iterator(gzfn, options=options))
self.assertEqual(actual, original)
class TFRecordIteratorTest(TFCompressionTestCase):
def setUp(self):
super(TFRecordIteratorTest, self).setUp()
self._num_records = 7
def testIterator(self):
records = [self._Record(0, i) for i in range(self._num_records)]
options = tf_record.TFRecordOptions(TFRecordCompressionType.ZLIB)
fn = self._WriteRecordsToFile(records, "compressed_records", options)
reader = tf_record.tf_record_iterator(fn, options)
for expected in records:
record = next(reader)
self.assertAllEqual(expected, record)
with self.assertRaises(StopIteration):
record = next(reader)
def testWriteZlibRead(self):
"""Verify compression with TFRecordWriter is zlib library compatible."""
original = [b"foo", b"bar"]
options = tf_record.TFRecordOptions(TFRecordCompressionType.ZLIB)
fn = self._WriteRecordsToFile(original, "write_zlib_read.tfrecord.z",
options)
zfn = self._ZlibDecompressFile(fn, "write_zlib_read.tfrecord")
actual = list(tf_record.tf_record_iterator(zfn))
self.assertEqual(actual, original)
def testWriteZlibReadLarge(self):
"""Verify compression for large records is zlib library compatible."""
# Make it large (about 5MB)
original = [_TEXT * 10240]
options = tf_record.TFRecordOptions(TFRecordCompressionType.ZLIB)
fn = self._WriteRecordsToFile(original, "write_zlib_read_large.tfrecord.z",
options)
zfn = self._ZlibDecompressFile(fn, "write_zlib_read_large.tfrecord")
actual = list(tf_record.tf_record_iterator(zfn))
self.assertEqual(actual, original)
def testWriteGzipRead(self):
original = [b"foo", b"bar"]
options = tf_record.TFRecordOptions(TFRecordCompressionType.GZIP)
fn = self._WriteRecordsToFile(original, "write_gzip_read.tfrecord.gz",
options)
gzfn = self._GzipDecompressFile(fn, "write_gzip_read.tfrecord")
actual = list(tf_record.tf_record_iterator(gzfn))
self.assertEqual(actual, original)
def testBadFile(self):
"""Verify that tf_record_iterator throws an exception on bad TFRecords."""
fn = os.path.join(self.get_temp_dir(), "bad_file")
with tf_record.TFRecordWriter(fn) as writer:
writer.write(b"123")
fn_truncated = os.path.join(self.get_temp_dir(), "bad_file_truncated")
with open(fn, "rb") as f:
with open(fn_truncated, "wb") as f2:
# DataLossError requires that we've written the header, so this must
# be at least 12 bytes.
f2.write(f.read(14))
with self.assertRaises(errors_impl.DataLossError):
for _ in tf_record.tf_record_iterator(fn_truncated):
pass
class TFRecordWriterCloseAndFlushTests(test.TestCase):
def setUp(self, compression_type=TFRecordCompressionType.NONE):
super(TFRecordWriterCloseAndFlushTests, self).setUp()
self._fn = os.path.join(self.get_temp_dir(), "tf_record_writer_test.txt")
self._options = tf_record.TFRecordOptions(compression_type)
self._writer = tf_record.TFRecordWriter(self._fn, self._options)
self._num_records = 20
def _Record(self, r):
return compat.as_bytes("Record %d" % r)
def testWriteAndLeaveOpen(self):
records = list(map(self._Record, range(self._num_records)))
for record in records:
self._writer.write(record)
# Verify no segfault if writer isn't explicitly closed.
def testWriteAndRead(self):
records = list(map(self._Record, range(self._num_records)))
for record in records:
self._writer.write(record)
self._writer.close()
actual = list(tf_record.tf_record_iterator(self._fn, self._options))
self.assertListEqual(actual, records)
def testDoubleClose(self):
self._writer.write(self._Record(0))
self._writer.close()
self._writer.close()
def testFlushAfterCloseIsError(self):
self._writer.write(self._Record(0))
self._writer.close()
with self.assertRaises(errors_impl.FailedPreconditionError):
self._writer.flush()
def testWriteAfterCloseIsError(self):
self._writer.write(self._Record(0))
self._writer.close()
with self.assertRaises(errors_impl.FailedPreconditionError):
self._writer.write(self._Record(1))
class TFRecordWriterCloseAndFlushGzipTests(TFRecordWriterCloseAndFlushTests):
def setUp(self):
super(TFRecordWriterCloseAndFlushGzipTests,
self).setUp(TFRecordCompressionType.GZIP)
class TFRecordWriterCloseAndFlushZlibTests(TFRecordWriterCloseAndFlushTests):
def setUp(self):
super(TFRecordWriterCloseAndFlushZlibTests,
self).setUp(TFRecordCompressionType.ZLIB)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/lib/io/tf_record_test.py
|
tensorflow-master
|
tensorflow/python/lib/io/__init__.py
|
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""For reading and writing TFRecords files."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python import pywrap_tensorflow
from tensorflow.python.framework import errors
from tensorflow.python.util import compat
from tensorflow.python.util import deprecation
from tensorflow.python.util.tf_export import tf_export
@tf_export(
v1=["io.TFRecordCompressionType", "python_io.TFRecordCompressionType"])
@deprecation.deprecated_endpoints("io.TFRecordCompressionType",
"python_io.TFRecordCompressionType")
class TFRecordCompressionType(object):
"""The type of compression for the record."""
NONE = 0
ZLIB = 1
GZIP = 2
@tf_export(
"io.TFRecordOptions",
v1=["io.TFRecordOptions", "python_io.TFRecordOptions"])
@deprecation.deprecated_endpoints("python_io.TFRecordOptions")
class TFRecordOptions(object):
"""Options used for manipulating TFRecord files."""
compression_type_map = {
TFRecordCompressionType.ZLIB: "ZLIB",
TFRecordCompressionType.GZIP: "GZIP",
TFRecordCompressionType.NONE: ""
}
def __init__(self,
compression_type=None,
flush_mode=None,
input_buffer_size=None,
output_buffer_size=None,
window_bits=None,
compression_level=None,
compression_method=None,
mem_level=None,
compression_strategy=None):
# pylint: disable=line-too-long
"""Creates a `TFRecordOptions` instance.
Options only effect TFRecordWriter when compression_type is not `None`.
Documentation, details, and defaults can be found in
[`zlib_compression_options.h`](https://www.tensorflow.org/code/tensorflow/core/lib/io/zlib_compression_options.h)
and in the [zlib manual](http://www.zlib.net/manual.html).
Leaving an option as `None` allows C++ to set a reasonable default.
Args:
compression_type: `"GZIP"`, `"ZLIB"`, or `""` (no compression).
flush_mode: flush mode or `None`, Default: Z_NO_FLUSH.
input_buffer_size: int or `None`.
output_buffer_size: int or `None`.
window_bits: int or `None`.
compression_level: 0 to 9, or `None`.
compression_method: compression method or `None`.
mem_level: 1 to 9, or `None`.
compression_strategy: strategy or `None`. Default: Z_DEFAULT_STRATEGY.
Returns:
A `TFRecordOptions` object.
Raises:
ValueError: If compression_type is invalid.
"""
# pylint: enable=line-too-long
# Check compression_type is valid, but for backwards compatibility don't
# immediately convert to a string.
self.get_compression_type_string(compression_type)
self.compression_type = compression_type
self.flush_mode = flush_mode
self.input_buffer_size = input_buffer_size
self.output_buffer_size = output_buffer_size
self.window_bits = window_bits
self.compression_level = compression_level
self.compression_method = compression_method
self.mem_level = mem_level
self.compression_strategy = compression_strategy
@classmethod
def get_compression_type_string(cls, options):
"""Convert various option types to a unified string.
Args:
options: `TFRecordOption`, `TFRecordCompressionType`, or string.
Returns:
Compression type as string (e.g. `'ZLIB'`, `'GZIP'`, or `''`).
Raises:
ValueError: If compression_type is invalid.
"""
if not options:
return ""
elif isinstance(options, TFRecordOptions):
return cls.get_compression_type_string(options.compression_type)
elif isinstance(options, TFRecordCompressionType):
return cls.compression_type_map[options]
elif options in TFRecordOptions.compression_type_map:
return cls.compression_type_map[options]
elif options in TFRecordOptions.compression_type_map.values():
return options
else:
raise ValueError('Not a valid compression_type: "{}"'.format(options))
def _as_record_writer_options(self):
"""Convert to RecordWriterOptions for use with PyRecordWriter."""
options = pywrap_tensorflow.RecordWriterOptions_CreateRecordWriterOptions(
compat.as_bytes(
self.get_compression_type_string(self.compression_type)))
if self.flush_mode is not None:
options.zlib_options.flush_mode = self.flush_mode
if self.input_buffer_size is not None:
options.zlib_options.input_buffer_size = self.input_buffer_size
if self.output_buffer_size is not None:
options.zlib_options.output_buffer_size = self.output_buffer_size
if self.window_bits is not None:
options.zlib_options.window_bits = self.window_bits
if self.compression_level is not None:
options.zlib_options.compression_level = self.compression_level
if self.compression_method is not None:
options.zlib_options.compression_method = self.compression_method
if self.mem_level is not None:
options.zlib_options.mem_level = self.mem_level
if self.compression_strategy is not None:
options.zlib_options.compression_strategy = self.compression_strategy
return options
@tf_export(v1=["io.tf_record_iterator", "python_io.tf_record_iterator"])
@deprecation.deprecated(
date=None,
instructions=("Use eager execution and: \n"
"`tf.data.TFRecordDataset(path)`"))
def tf_record_iterator(path, options=None):
"""An iterator that read the records from a TFRecords file.
Args:
path: The path to the TFRecords file.
options: (optional) A TFRecordOptions object.
Yields:
Strings.
Raises:
IOError: If `path` cannot be opened for reading.
"""
compression_type = TFRecordOptions.get_compression_type_string(options)
with errors.raise_exception_on_not_ok_status() as status:
reader = pywrap_tensorflow.PyRecordReader_New(
compat.as_bytes(path), 0, compat.as_bytes(compression_type), status)
if reader is None:
raise IOError("Could not open %s." % path)
try:
while True:
try:
reader.GetNext()
except errors.OutOfRangeError:
break
yield reader.record()
finally:
reader.Close()
@tf_export(
"io.TFRecordWriter", v1=["io.TFRecordWriter", "python_io.TFRecordWriter"])
@deprecation.deprecated_endpoints("python_io.TFRecordWriter")
class TFRecordWriter(object):
"""A class to write records to a TFRecords file.
This class implements `__enter__` and `__exit__`, and can be used
in `with` blocks like a normal file.
"""
# TODO(josh11b): Support appending?
def __init__(self, path, options=None):
"""Opens file `path` and creates a `TFRecordWriter` writing to it.
Args:
path: The path to the TFRecords file.
options: (optional) String specifying compression type,
`TFRecordCompressionType`, or `TFRecordOptions` object.
Raises:
IOError: If `path` cannot be opened for writing.
ValueError: If valid compression_type can't be determined from `options`.
"""
if not isinstance(options, TFRecordOptions):
options = TFRecordOptions(compression_type=options)
with errors.raise_exception_on_not_ok_status() as status:
# pylint: disable=protected-access
self._writer = pywrap_tensorflow.PyRecordWriter_New(
compat.as_bytes(path), options._as_record_writer_options(), status)
# pylint: enable=protected-access
def __enter__(self):
"""Enter a `with` block."""
return self
def __exit__(self, unused_type, unused_value, unused_traceback):
"""Exit a `with` block, closing the file."""
self.close()
def write(self, record):
"""Write a string record to the file.
Args:
record: str
"""
with errors.raise_exception_on_not_ok_status() as status:
self._writer.WriteRecord(record, status)
def flush(self):
"""Flush the file."""
with errors.raise_exception_on_not_ok_status() as status:
self._writer.Flush(status)
def close(self):
"""Close the file."""
with errors.raise_exception_on_not_ok_status() as status:
self._writer.Close(status)
|
tensorflow-master
|
tensorflow/python/lib/io/tf_record.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""File IO methods that wrap the C++ FileSystem API.
The C++ FileSystem API is SWIG wrapped in file_io.i. These functions call those
to accomplish basic File IO operations.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import binascii
import os
import uuid
import six
from tensorflow.python import pywrap_tensorflow
from tensorflow.python.framework import c_api_util
from tensorflow.python.framework import errors
from tensorflow.python.util import compat
from tensorflow.python.util import deprecation
from tensorflow.python.util.tf_export import tf_export
# A good default block size depends on the system in question.
# A somewhat conservative default chosen here.
_DEFAULT_BLOCK_SIZE = 16 * 1024 * 1024
class FileIO(object):
"""FileIO class that exposes methods to read / write to / from files.
The constructor takes the following arguments:
name: name of the file
mode: one of 'r', 'w', 'a', 'r+', 'w+', 'a+'. Append 'b' for bytes mode.
Can be used as an iterator to iterate over lines in the file.
The default buffer size used for the BufferedInputStream used for reading
the file line by line is 1024 * 512 bytes.
"""
def __init__(self, name, mode):
self.__name = name
self.__mode = mode
self._read_buf = None
self._writable_file = None
self._binary_mode = "b" in mode
mode = mode.replace("b", "")
if mode not in ("r", "w", "a", "r+", "w+", "a+"):
raise errors.InvalidArgumentError(
None, None, "mode is not 'r' or 'w' or 'a' or 'r+' or 'w+' or 'a+'")
self._read_check_passed = mode in ("r", "r+", "a+", "w+")
self._write_check_passed = mode in ("a", "w", "r+", "a+", "w+")
@property
def name(self):
"""Returns the file name."""
return self.__name
@property
def mode(self):
"""Returns the mode in which the file was opened."""
return self.__mode
def _preread_check(self):
if not self._read_buf:
if not self._read_check_passed:
raise errors.PermissionDeniedError(None, None,
"File isn't open for reading")
self._read_buf = pywrap_tensorflow.CreateBufferedInputStream(
compat.as_bytes(self.__name), 1024 * 512)
def _prewrite_check(self):
if not self._writable_file:
if not self._write_check_passed:
raise errors.PermissionDeniedError(None, None,
"File isn't open for writing")
self._writable_file = pywrap_tensorflow.CreateWritableFile(
compat.as_bytes(self.__name), compat.as_bytes(self.__mode))
def _prepare_value(self, val):
if self._binary_mode:
return compat.as_bytes(val)
else:
return compat.as_str_any(val)
def size(self):
"""Returns the size of the file."""
return stat(self.__name).length
def write(self, file_content):
"""Writes file_content to the file. Appends to the end of the file."""
self._prewrite_check()
pywrap_tensorflow.AppendToFile(
compat.as_bytes(file_content), self._writable_file)
def read(self, n=-1):
"""Returns the contents of a file as a string.
Starts reading from current position in file.
Args:
n: Read 'n' bytes if n != -1. If n = -1, reads to end of file.
Returns:
'n' bytes of the file (or whole file) in bytes mode or 'n' bytes of the
string if in string (regular) mode.
"""
self._preread_check()
if n == -1:
length = self.size() - self.tell()
else:
length = n
return self._prepare_value(
pywrap_tensorflow.ReadFromStream(self._read_buf, length))
@deprecation.deprecated_args(
None, "position is deprecated in favor of the offset argument.",
"position")
def seek(self, offset=None, whence=0, position=None):
# TODO(jhseu): Delete later. Used to omit `position` from docs.
# pylint: disable=g-doc-args
"""Seeks to the offset in the file.
Args:
offset: The byte count relative to the whence argument.
whence: Valid values for whence are:
0: start of the file (default)
1: relative to the current position of the file
2: relative to the end of file. offset is usually negative.
"""
# pylint: enable=g-doc-args
self._preread_check()
# We needed to make offset a keyword argument for backwards-compatibility.
# This check exists so that we can convert back to having offset be a
# positional argument.
# TODO(jhseu): Make `offset` a positional argument after `position` is
# deleted.
if offset is None and position is None:
raise TypeError("seek(): offset argument required")
if offset is not None and position is not None:
raise TypeError("seek(): offset and position may not be set "
"simultaneously.")
if position is not None:
offset = position
with errors.raise_exception_on_not_ok_status() as status:
if whence == 0:
pass
elif whence == 1:
offset += self.tell()
elif whence == 2:
offset += self.size()
else:
raise errors.InvalidArgumentError(
None, None,
"Invalid whence argument: {}. Valid values are 0, 1, or 2.".format(
whence))
ret_status = self._read_buf.Seek(offset)
pywrap_tensorflow.Set_TF_Status_from_Status(status, ret_status)
def readline(self):
r"""Reads the next line from the file. Leaves the '\n' at the end."""
self._preread_check()
return self._prepare_value(self._read_buf.ReadLineAsString())
def readlines(self):
"""Returns all lines from the file in a list."""
self._preread_check()
lines = []
while True:
s = self.readline()
if not s:
break
lines.append(s)
return lines
def tell(self):
"""Returns the current position in the file."""
if self._read_check_passed:
self._preread_check()
return self._read_buf.Tell()
else:
self._prewrite_check()
return pywrap_tensorflow.TellFile(self._writable_file)
def __enter__(self):
"""Make usable with "with" statement."""
return self
def __exit__(self, unused_type, unused_value, unused_traceback):
"""Make usable with "with" statement."""
self.close()
def __iter__(self):
return self
def next(self):
retval = self.readline()
if not retval:
raise StopIteration()
return retval
def __next__(self):
return self.next()
def flush(self):
"""Flushes the Writable file.
This only ensures that the data has made its way out of the process without
any guarantees on whether it's written to disk. This means that the
data would survive an application crash but not necessarily an OS crash.
"""
if self._writable_file:
with errors.raise_exception_on_not_ok_status() as status:
ret_status = self._writable_file.Flush()
pywrap_tensorflow.Set_TF_Status_from_Status(status, ret_status)
def close(self):
"""Closes FileIO. Should be called for the WritableFile to be flushed."""
self._read_buf = None
if self._writable_file:
with errors.raise_exception_on_not_ok_status() as status:
ret_status = self._writable_file.Close()
pywrap_tensorflow.Set_TF_Status_from_Status(status, ret_status)
self._writable_file = None
def seekable(self):
"""Returns True as FileIO supports random access ops of seek()/tell()"""
return True
@tf_export(v1=["gfile.Exists"])
def file_exists(filename):
"""Determines whether a path exists or not.
Args:
filename: string, a path
Returns:
True if the path exists, whether it's a file or a directory.
False if the path does not exist and there are no filesystem errors.
Raises:
errors.OpError: Propagates any errors reported by the FileSystem API.
"""
return file_exists_v2(filename)
@tf_export("io.gfile.exists")
def file_exists_v2(path):
"""Determines whether a path exists or not.
Args:
path: string, a path
Returns:
True if the path exists, whether it's a file or a directory.
False if the path does not exist and there are no filesystem errors.
Raises:
errors.OpError: Propagates any errors reported by the FileSystem API.
"""
try:
pywrap_tensorflow.FileExists(compat.as_bytes(path))
except errors.NotFoundError:
return False
return True
@tf_export(v1=["gfile.Remove"])
def delete_file(filename):
"""Deletes the file located at 'filename'.
Args:
filename: string, a filename
Raises:
errors.OpError: Propagates any errors reported by the FileSystem API. E.g.,
NotFoundError if the file does not exist.
"""
delete_file_v2(filename)
@tf_export("io.gfile.remove")
def delete_file_v2(path):
"""Deletes the path located at 'path'.
Args:
path: string, a path
Raises:
errors.OpError: Propagates any errors reported by the FileSystem API. E.g.,
NotFoundError if the path does not exist.
"""
pywrap_tensorflow.DeleteFile(compat.as_bytes(path))
def read_file_to_string(filename, binary_mode=False):
"""Reads the entire contents of a file to a string.
Args:
filename: string, path to a file
binary_mode: whether to open the file in binary mode or not. This changes
the type of the object returned.
Returns:
contents of the file as a string or bytes.
Raises:
errors.OpError: Raises variety of errors that are subtypes e.g.
NotFoundError etc.
"""
if binary_mode:
f = FileIO(filename, mode="rb")
else:
f = FileIO(filename, mode="r")
return f.read()
def write_string_to_file(filename, file_content):
"""Writes a string to a given file.
Args:
filename: string, path to a file
file_content: string, contents that need to be written to the file
Raises:
errors.OpError: If there are errors during the operation.
"""
with FileIO(filename, mode="w") as f:
f.write(file_content)
@tf_export(v1=["gfile.Glob"])
def get_matching_files(filename):
"""Returns a list of files that match the given pattern(s).
Args:
filename: string or iterable of strings. The glob pattern(s).
Returns:
A list of strings containing filenames that match the given pattern(s).
Raises:
errors.OpError: If there are filesystem / directory listing errors.
"""
return get_matching_files_v2(filename)
@tf_export("io.gfile.glob")
def get_matching_files_v2(pattern):
"""Returns a list of files that match the given pattern(s).
Args:
pattern: string or iterable of strings. The glob pattern(s).
Returns:
A list of strings containing filenames that match the given pattern(s).
Raises:
errors.OpError: If there are filesystem / directory listing errors.
"""
if isinstance(pattern, six.string_types):
return [
# Convert the filenames to string from bytes.
compat.as_str_any(matching_filename)
for matching_filename in pywrap_tensorflow.GetMatchingFiles(
compat.as_bytes(pattern))
]
else:
return [
# Convert the filenames to string from bytes.
compat.as_str_any(matching_filename) # pylint: disable=g-complex-comprehension
for single_filename in pattern
for matching_filename in pywrap_tensorflow.GetMatchingFiles(
compat.as_bytes(single_filename))
]
@tf_export(v1=["gfile.MkDir"])
def create_dir(dirname):
"""Creates a directory with the name 'dirname'.
Args:
dirname: string, name of the directory to be created
Notes: The parent directories need to exist. Use recursive_create_dir instead
if there is the possibility that the parent dirs don't exist.
Raises:
errors.OpError: If the operation fails.
"""
create_dir_v2(dirname)
@tf_export("io.gfile.mkdir")
def create_dir_v2(path):
"""Creates a directory with the name given by 'path'.
Args:
path: string, name of the directory to be created
Notes: The parent directories need to exist. Use recursive_create_dir instead
if there is the possibility that the parent dirs don't exist.
Raises:
errors.OpError: If the operation fails.
"""
pywrap_tensorflow.CreateDir(compat.as_bytes(path))
@tf_export(v1=["gfile.MakeDirs"])
def recursive_create_dir(dirname):
"""Creates a directory and all parent/intermediate directories.
It succeeds if dirname already exists and is writable.
Args:
dirname: string, name of the directory to be created
Raises:
errors.OpError: If the operation fails.
"""
recursive_create_dir_v2(dirname)
@tf_export("io.gfile.makedirs")
def recursive_create_dir_v2(path):
"""Creates a directory and all parent/intermediate directories.
It succeeds if path already exists and is writable.
Args:
path: string, name of the directory to be created
Raises:
errors.OpError: If the operation fails.
"""
pywrap_tensorflow.RecursivelyCreateDir(compat.as_bytes(path))
@tf_export(v1=["gfile.Copy"])
def copy(oldpath, newpath, overwrite=False):
"""Copies data from `oldpath` to `newpath`.
Args:
oldpath: string, name of the file who's contents need to be copied
newpath: string, name of the file to which to copy to
overwrite: boolean, if false it's an error for `newpath` to be occupied by
an existing file.
Raises:
errors.OpError: If the operation fails.
"""
copy_v2(oldpath, newpath, overwrite)
@tf_export("io.gfile.copy")
def copy_v2(src, dst, overwrite=False):
"""Copies data from `src` to `dst`.
Args:
src: string, name of the file whose contents need to be copied
dst: string, name of the file to which to copy to
overwrite: boolean, if false it's an error for `dst` to be occupied by an
existing file.
Raises:
errors.OpError: If the operation fails.
"""
pywrap_tensorflow.CopyFile(
compat.as_bytes(src), compat.as_bytes(dst), overwrite)
@tf_export(v1=["gfile.Rename"])
def rename(oldname, newname, overwrite=False):
"""Rename or move a file / directory.
Args:
oldname: string, pathname for a file
newname: string, pathname to which the file needs to be moved
overwrite: boolean, if false it's an error for `newname` to be occupied by
an existing file.
Raises:
errors.OpError: If the operation fails.
"""
rename_v2(oldname, newname, overwrite)
@tf_export("io.gfile.rename")
def rename_v2(src, dst, overwrite=False):
"""Rename or move a file / directory.
Args:
src: string, pathname for a file
dst: string, pathname to which the file needs to be moved
overwrite: boolean, if false it's an error for `dst` to be occupied by an
existing file.
Raises:
errors.OpError: If the operation fails.
"""
pywrap_tensorflow.RenameFile(
compat.as_bytes(src), compat.as_bytes(dst), overwrite)
def atomic_write_string_to_file(filename, contents, overwrite=True):
"""Writes to `filename` atomically.
This means that when `filename` appears in the filesystem, it will contain
all of `contents`. With write_string_to_file, it is possible for the file
to appear in the filesystem with `contents` only partially written.
Accomplished by writing to a temp file and then renaming it.
Args:
filename: string, pathname for a file
contents: string, contents that need to be written to the file
overwrite: boolean, if false it's an error for `filename` to be occupied by
an existing file.
"""
temp_pathname = filename + ".tmp" + uuid.uuid4().hex
write_string_to_file(temp_pathname, contents)
try:
rename(temp_pathname, filename, overwrite)
except errors.OpError:
delete_file(temp_pathname)
raise
@tf_export(v1=["gfile.DeleteRecursively"])
def delete_recursively(dirname):
"""Deletes everything under dirname recursively.
Args:
dirname: string, a path to a directory
Raises:
errors.OpError: If the operation fails.
"""
delete_recursively_v2(dirname)
@tf_export("io.gfile.rmtree")
def delete_recursively_v2(path):
"""Deletes everything under path recursively.
Args:
path: string, a path
Raises:
errors.OpError: If the operation fails.
"""
pywrap_tensorflow.DeleteRecursively(compat.as_bytes(path))
@tf_export(v1=["gfile.IsDirectory"])
def is_directory(dirname):
"""Returns whether the path is a directory or not.
Args:
dirname: string, path to a potential directory
Returns:
True, if the path is a directory; False otherwise
"""
return is_directory_v2(dirname)
@tf_export("io.gfile.isdir")
def is_directory_v2(path):
"""Returns whether the path is a directory or not.
Args:
path: string, path to a potential directory
Returns:
True, if the path is a directory; False otherwise
"""
status = c_api_util.ScopedTFStatus()
return pywrap_tensorflow.IsDirectory(compat.as_bytes(path), status)
@tf_export(v1=["gfile.ListDirectory"])
def list_directory(dirname):
"""Returns a list of entries contained within a directory.
The list is in arbitrary order. It does not contain the special entries "."
and "..".
Args:
dirname: string, path to a directory
Returns:
[filename1, filename2, ... filenameN] as strings
Raises:
errors.NotFoundError if directory doesn't exist
"""
return list_directory_v2(dirname)
@tf_export("io.gfile.listdir")
def list_directory_v2(path):
"""Returns a list of entries contained within a directory.
The list is in arbitrary order. It does not contain the special entries "."
and "..".
Args:
path: string, path to a directory
Returns:
[filename1, filename2, ... filenameN] as strings
Raises:
errors.NotFoundError if directory doesn't exist
"""
if not is_directory(path):
raise errors.NotFoundError(
node_def=None,
op=None,
message="Could not find directory {}".format(path))
# Convert each element to string, since the return values of the
# vector of string should be interpreted as strings, not bytes.
return [
compat.as_str_any(filename)
for filename in pywrap_tensorflow.GetChildren(compat.as_bytes(path))
]
@tf_export(v1=["gfile.Walk"])
def walk(top, in_order=True):
"""Recursive directory tree generator for directories.
Args:
top: string, a Directory name
in_order: bool, Traverse in order if True, post order if False. Errors that
happen while listing directories are ignored.
Yields:
Each yield is a 3-tuple: the pathname of a directory, followed by lists of
all its subdirectories and leaf files.
(dirname, [subdirname, subdirname, ...], [filename, filename, ...])
as strings
"""
return walk_v2(top, in_order)
@tf_export("io.gfile.walk")
def walk_v2(top, topdown=True, onerror=None):
"""Recursive directory tree generator for directories.
Args:
top: string, a Directory name
topdown: bool, Traverse pre order if True, post order if False.
onerror: optional handler for errors. Should be a function, it will be
called with the error as argument. Rethrowing the error aborts the walk.
Errors that happen while listing directories are ignored.
Yields:
Each yield is a 3-tuple: the pathname of a directory, followed by lists of
all its subdirectories and leaf files.
(dirname, [subdirname, subdirname, ...], [filename, filename, ...])
as strings
"""
top = compat.as_str_any(top)
try:
listing = list_directory(top)
except errors.NotFoundError as err:
if onerror:
onerror(err)
else:
return
files = []
subdirs = []
for item in listing:
full_path = os.path.join(top, item)
if is_directory(full_path):
subdirs.append(item)
else:
files.append(item)
here = (top, subdirs, files)
if topdown:
yield here
for subdir in subdirs:
for subitem in walk_v2(os.path.join(top, subdir), topdown, onerror=onerror):
yield subitem
if not topdown:
yield here
@tf_export(v1=["gfile.Stat"])
def stat(filename):
"""Returns file statistics for a given path.
Args:
filename: string, path to a file
Returns:
FileStatistics struct that contains information about the path
Raises:
errors.OpError: If the operation fails.
"""
return stat_v2(filename)
@tf_export("io.gfile.stat")
def stat_v2(path):
"""Returns file statistics for a given path.
Args:
path: string, path to a file
Returns:
FileStatistics struct that contains information about the path
Raises:
errors.OpError: If the operation fails.
"""
file_statistics = pywrap_tensorflow.FileStatistics()
pywrap_tensorflow.Stat(compat.as_bytes(path), file_statistics)
return file_statistics
def filecmp(filename_a, filename_b):
"""Compare two files, returning True if they are the same, False otherwise.
We check size first and return False quickly if the files are different sizes.
If they are the same size, we continue to generating a crc for the whole file.
You might wonder: why not use Python's filecmp.cmp() instead? The answer is
that the builtin library is not robust to the many different filesystems
TensorFlow runs on, and so we here perform a similar comparison with
the more robust FileIO.
Args:
filename_a: string path to the first file.
filename_b: string path to the second file.
Returns:
True if the files are the same, False otherwise.
"""
size_a = FileIO(filename_a, "rb").size()
size_b = FileIO(filename_b, "rb").size()
if size_a != size_b:
return False
# Size is the same. Do a full check.
crc_a = file_crc32(filename_a)
crc_b = file_crc32(filename_b)
return crc_a == crc_b
def file_crc32(filename, block_size=_DEFAULT_BLOCK_SIZE):
"""Get the crc32 of the passed file.
The crc32 of a file can be used for error checking; two files with the same
crc32 are considered equivalent. Note that the entire file must be read
to produce the crc32.
Args:
filename: string, path to a file
block_size: Integer, process the files by reading blocks of `block_size`
bytes. Use -1 to read the file as once.
Returns:
hexadecimal as string, the crc32 of the passed file.
"""
crc = 0
with FileIO(filename, mode="rb") as f:
chunk = f.read(n=block_size)
while chunk:
crc = binascii.crc32(chunk, crc)
chunk = f.read(n=block_size)
return hex(crc & 0xFFFFFFFF)
|
tensorflow-master
|
tensorflow/python/lib/io/file_io.py
|
# This Python file uses the following encoding: utf-8
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# =============================================================================
"""Testing File IO operations in file_io.py."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os.path
import numpy as np
from tensorflow.python.framework import errors
from tensorflow.python.lib.io import file_io
from tensorflow.python.platform import gfile
from tensorflow.python.platform import test
class FileIoTest(test.TestCase):
def setUp(self):
self._base_dir = os.path.join(self.get_temp_dir(), "base_dir")
file_io.create_dir(self._base_dir)
def tearDown(self):
file_io.delete_recursively(self._base_dir)
def testEmptyFilename(self):
f = file_io.FileIO("", mode="r")
with self.assertRaises(errors.NotFoundError):
_ = f.read()
def testFileDoesntExist(self):
file_path = os.path.join(self._base_dir, "temp_file")
self.assertFalse(file_io.file_exists(file_path))
with self.assertRaises(errors.NotFoundError):
_ = file_io.read_file_to_string(file_path)
def testWriteToString(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.write_string_to_file(file_path, "testing")
self.assertTrue(file_io.file_exists(file_path))
file_contents = file_io.read_file_to_string(file_path)
self.assertEqual("testing", file_contents)
def testAtomicWriteStringToFile(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.atomic_write_string_to_file(file_path, "testing")
self.assertTrue(file_io.file_exists(file_path))
file_contents = file_io.read_file_to_string(file_path)
self.assertEqual("testing", file_contents)
def testAtomicWriteStringToFileOverwriteFalse(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.atomic_write_string_to_file(file_path, "old", overwrite=False)
with self.assertRaises(errors.AlreadyExistsError):
file_io.atomic_write_string_to_file(file_path, "new", overwrite=False)
file_contents = file_io.read_file_to_string(file_path)
self.assertEqual("old", file_contents)
file_io.delete_file(file_path)
file_io.atomic_write_string_to_file(file_path, "new", overwrite=False)
file_contents = file_io.read_file_to_string(file_path)
self.assertEqual("new", file_contents)
def testReadBinaryMode(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.write_string_to_file(file_path, "testing")
with file_io.FileIO(file_path, mode="rb") as f:
self.assertEqual(b"testing", f.read())
def testWriteBinaryMode(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.FileIO(file_path, "wb").write("testing")
with file_io.FileIO(file_path, mode="r") as f:
self.assertEqual("testing", f.read())
def testAppend(self):
file_path = os.path.join(self._base_dir, "temp_file")
with file_io.FileIO(file_path, mode="w") as f:
f.write("begin\n")
with file_io.FileIO(file_path, mode="a") as f:
f.write("a1\n")
with file_io.FileIO(file_path, mode="a") as f:
f.write("a2\n")
with file_io.FileIO(file_path, mode="r") as f:
file_contents = f.read()
self.assertEqual("begin\na1\na2\n", file_contents)
def testMultipleFiles(self):
file_prefix = os.path.join(self._base_dir, "temp_file")
for i in range(5000):
f = file_io.FileIO(file_prefix + str(i), mode="w+")
f.write("testing")
f.flush()
self.assertEqual("testing", f.read())
f.close()
def testMultipleWrites(self):
file_path = os.path.join(self._base_dir, "temp_file")
with file_io.FileIO(file_path, mode="w") as f:
f.write("line1\n")
f.write("line2")
file_contents = file_io.read_file_to_string(file_path)
self.assertEqual("line1\nline2", file_contents)
def testFileWriteBadMode(self):
file_path = os.path.join(self._base_dir, "temp_file")
with self.assertRaises(errors.PermissionDeniedError):
file_io.FileIO(file_path, mode="r").write("testing")
def testFileReadBadMode(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.FileIO(file_path, mode="w").write("testing")
self.assertTrue(file_io.file_exists(file_path))
with self.assertRaises(errors.PermissionDeniedError):
file_io.FileIO(file_path, mode="w").read()
def testFileDelete(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.FileIO(file_path, mode="w").write("testing")
file_io.delete_file(file_path)
self.assertFalse(file_io.file_exists(file_path))
def testFileDeleteFail(self):
file_path = os.path.join(self._base_dir, "temp_file")
with self.assertRaises(errors.NotFoundError):
file_io.delete_file(file_path)
def testGetMatchingFiles(self):
dir_path = os.path.join(self._base_dir, "temp_dir")
file_io.create_dir(dir_path)
files = ["file1.txt", "file2.txt", "file3.txt", "file*.txt"]
for name in files:
file_path = os.path.join(dir_path, name)
file_io.FileIO(file_path, mode="w").write("testing")
expected_match = [os.path.join(dir_path, name) for name in files]
self.assertItemsEqual(
file_io.get_matching_files(os.path.join(dir_path, "file*.txt")),
expected_match)
self.assertItemsEqual(file_io.get_matching_files(tuple()), [])
files_subset = [
os.path.join(dir_path, files[0]), os.path.join(dir_path, files[2])
]
self.assertItemsEqual(
file_io.get_matching_files(files_subset), files_subset)
file_io.delete_recursively(dir_path)
self.assertFalse(file_io.file_exists(os.path.join(dir_path, "file3.txt")))
def testCreateRecursiveDir(self):
dir_path = os.path.join(self._base_dir, "temp_dir/temp_dir1/temp_dir2")
file_io.recursive_create_dir(dir_path)
file_io.recursive_create_dir(dir_path) # repeat creation
file_path = os.path.join(dir_path, "temp_file")
file_io.FileIO(file_path, mode="w").write("testing")
self.assertTrue(file_io.file_exists(file_path))
file_io.delete_recursively(os.path.join(self._base_dir, "temp_dir"))
self.assertFalse(file_io.file_exists(file_path))
def testCopy(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.FileIO(file_path, mode="w").write("testing")
copy_path = os.path.join(self._base_dir, "copy_file")
file_io.copy(file_path, copy_path)
self.assertTrue(file_io.file_exists(copy_path))
f = file_io.FileIO(file_path, mode="r")
self.assertEqual("testing", f.read())
self.assertEqual(7, f.tell())
def testCopyOverwrite(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.FileIO(file_path, mode="w").write("testing")
copy_path = os.path.join(self._base_dir, "copy_file")
file_io.FileIO(copy_path, mode="w").write("copy")
file_io.copy(file_path, copy_path, overwrite=True)
self.assertTrue(file_io.file_exists(copy_path))
self.assertEqual("testing", file_io.FileIO(file_path, mode="r").read())
def testCopyOverwriteFalse(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.FileIO(file_path, mode="w").write("testing")
copy_path = os.path.join(self._base_dir, "copy_file")
file_io.FileIO(copy_path, mode="w").write("copy")
with self.assertRaises(errors.AlreadyExistsError):
file_io.copy(file_path, copy_path, overwrite=False)
def testRename(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.FileIO(file_path, mode="w").write("testing")
rename_path = os.path.join(self._base_dir, "rename_file")
file_io.rename(file_path, rename_path)
self.assertTrue(file_io.file_exists(rename_path))
self.assertFalse(file_io.file_exists(file_path))
def testRenameOverwrite(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.FileIO(file_path, mode="w").write("testing")
rename_path = os.path.join(self._base_dir, "rename_file")
file_io.FileIO(rename_path, mode="w").write("rename")
file_io.rename(file_path, rename_path, overwrite=True)
self.assertTrue(file_io.file_exists(rename_path))
self.assertFalse(file_io.file_exists(file_path))
def testRenameOverwriteFalse(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.FileIO(file_path, mode="w").write("testing")
rename_path = os.path.join(self._base_dir, "rename_file")
file_io.FileIO(rename_path, mode="w").write("rename")
with self.assertRaises(errors.AlreadyExistsError):
file_io.rename(file_path, rename_path, overwrite=False)
self.assertTrue(file_io.file_exists(rename_path))
self.assertTrue(file_io.file_exists(file_path))
def testDeleteRecursivelyFail(self):
fake_dir_path = os.path.join(self._base_dir, "temp_dir")
with self.assertRaises(errors.NotFoundError):
file_io.delete_recursively(fake_dir_path)
def testIsDirectory(self):
dir_path = os.path.join(self._base_dir, "test_dir")
# Failure for a non-existing dir.
self.assertFalse(file_io.is_directory(dir_path))
file_io.create_dir(dir_path)
self.assertTrue(file_io.is_directory(dir_path))
file_path = os.path.join(dir_path, "test_file")
file_io.FileIO(file_path, mode="w").write("test")
# False for a file.
self.assertFalse(file_io.is_directory(file_path))
# Test that the value returned from `stat()` has `is_directory` set.
file_statistics = file_io.stat(dir_path)
self.assertTrue(file_statistics.is_directory)
def testListDirectory(self):
dir_path = os.path.join(self._base_dir, "test_dir")
file_io.create_dir(dir_path)
files = ["file1.txt", "file2.txt", "file3.txt"]
for name in files:
file_path = os.path.join(dir_path, name)
file_io.FileIO(file_path, mode="w").write("testing")
subdir_path = os.path.join(dir_path, "sub_dir")
file_io.create_dir(subdir_path)
subdir_file_path = os.path.join(subdir_path, "file4.txt")
file_io.FileIO(subdir_file_path, mode="w").write("testing")
dir_list = file_io.list_directory(dir_path)
self.assertItemsEqual(files + ["sub_dir"], dir_list)
def testListDirectoryFailure(self):
dir_path = os.path.join(self._base_dir, "test_dir")
with self.assertRaises(errors.NotFoundError):
file_io.list_directory(dir_path)
def _setupWalkDirectories(self, dir_path):
# Creating a file structure as follows
# test_dir -> file: file1.txt; dirs: subdir1_1, subdir1_2, subdir1_3
# subdir1_1 -> file: file3.txt
# subdir1_2 -> dir: subdir2
file_io.create_dir(dir_path)
file_io.FileIO(
os.path.join(dir_path, "file1.txt"), mode="w").write("testing")
sub_dirs1 = ["subdir1_1", "subdir1_2", "subdir1_3"]
for name in sub_dirs1:
file_io.create_dir(os.path.join(dir_path, name))
file_io.FileIO(
os.path.join(dir_path, "subdir1_1/file2.txt"),
mode="w").write("testing")
file_io.create_dir(os.path.join(dir_path, "subdir1_2/subdir2"))
def testWalkInOrder(self):
dir_path = os.path.join(self._base_dir, "test_dir")
self._setupWalkDirectories(dir_path)
# Now test the walk (in_order = True)
all_dirs = []
all_subdirs = []
all_files = []
for (w_dir, w_subdirs, w_files) in file_io.walk(dir_path, in_order=True):
all_dirs.append(w_dir)
all_subdirs.append(w_subdirs)
all_files.append(w_files)
self.assertItemsEqual(all_dirs, [dir_path] + [
os.path.join(dir_path, item)
for item in
["subdir1_1", "subdir1_2", "subdir1_2/subdir2", "subdir1_3"]
])
self.assertEqual(dir_path, all_dirs[0])
self.assertLess(
all_dirs.index(os.path.join(dir_path, "subdir1_2")),
all_dirs.index(os.path.join(dir_path, "subdir1_2/subdir2")))
self.assertItemsEqual(all_subdirs[1:5], [[], ["subdir2"], [], []])
self.assertItemsEqual(all_subdirs[0],
["subdir1_1", "subdir1_2", "subdir1_3"])
self.assertItemsEqual(all_files, [["file1.txt"], ["file2.txt"], [], [], []])
self.assertLess(
all_files.index(["file1.txt"]), all_files.index(["file2.txt"]))
def testWalkPostOrder(self):
dir_path = os.path.join(self._base_dir, "test_dir")
self._setupWalkDirectories(dir_path)
# Now test the walk (in_order = False)
all_dirs = []
all_subdirs = []
all_files = []
for (w_dir, w_subdirs, w_files) in file_io.walk(dir_path, in_order=False):
all_dirs.append(w_dir)
all_subdirs.append(w_subdirs)
all_files.append(w_files)
self.assertItemsEqual(all_dirs, [
os.path.join(dir_path, item)
for item in
["subdir1_1", "subdir1_2/subdir2", "subdir1_2", "subdir1_3"]
] + [dir_path])
self.assertEqual(dir_path, all_dirs[4])
self.assertLess(
all_dirs.index(os.path.join(dir_path, "subdir1_2/subdir2")),
all_dirs.index(os.path.join(dir_path, "subdir1_2")))
self.assertItemsEqual(all_subdirs[0:4], [[], [], ["subdir2"], []])
self.assertItemsEqual(all_subdirs[4],
["subdir1_1", "subdir1_2", "subdir1_3"])
self.assertItemsEqual(all_files, [["file2.txt"], [], [], [], ["file1.txt"]])
self.assertLess(
all_files.index(["file2.txt"]), all_files.index(["file1.txt"]))
def testWalkFailure(self):
dir_path = os.path.join(self._base_dir, "test_dir")
# Try walking a directory that wasn't created.
all_dirs = []
all_subdirs = []
all_files = []
for (w_dir, w_subdirs, w_files) in file_io.walk(dir_path, in_order=False):
all_dirs.append(w_dir)
all_subdirs.append(w_subdirs)
all_files.append(w_files)
self.assertItemsEqual(all_dirs, [])
self.assertItemsEqual(all_subdirs, [])
self.assertItemsEqual(all_files, [])
def testStat(self):
file_path = os.path.join(self._base_dir, "temp_file")
file_io.FileIO(file_path, mode="w").write("testing")
file_statistics = file_io.stat(file_path)
os_statistics = os.stat(file_path)
self.assertEqual(7, file_statistics.length)
self.assertEqual(
int(os_statistics.st_mtime), int(file_statistics.mtime_nsec / 1e9))
self.assertFalse(file_statistics.is_directory)
def testReadLine(self):
file_path = os.path.join(self._base_dir, "temp_file")
with file_io.FileIO(file_path, mode="r+") as f:
f.write("testing1\ntesting2\ntesting3\n\ntesting5")
self.assertEqual(36, f.size())
self.assertEqual("testing1\n", f.readline())
self.assertEqual("testing2\n", f.readline())
self.assertEqual("testing3\n", f.readline())
self.assertEqual("\n", f.readline())
self.assertEqual("testing5", f.readline())
self.assertEqual("", f.readline())
def testRead(self):
file_path = os.path.join(self._base_dir, "temp_file")
with file_io.FileIO(file_path, mode="r+") as f:
f.write("testing1\ntesting2\ntesting3\n\ntesting5")
self.assertEqual(36, f.size())
self.assertEqual("testing1\n", f.read(9))
self.assertEqual("testing2\n", f.read(9))
self.assertEqual("t", f.read(1))
self.assertEqual("esting3\n\ntesting5", f.read())
def testTell(self):
file_path = os.path.join(self._base_dir, "temp_file")
with file_io.FileIO(file_path, mode="r+") as f:
f.write("testing1\ntesting2\ntesting3\n\ntesting5")
self.assertEqual(0, f.tell())
self.assertEqual("testing1\n", f.readline())
self.assertEqual(9, f.tell())
self.assertEqual("testing2\n", f.readline())
self.assertEqual(18, f.tell())
self.assertEqual("testing3\n", f.readline())
self.assertEqual(27, f.tell())
self.assertEqual("\n", f.readline())
self.assertEqual(28, f.tell())
self.assertEqual("testing5", f.readline())
self.assertEqual(36, f.tell())
self.assertEqual("", f.readline())
self.assertEqual(36, f.tell())
def testSeek(self):
file_path = os.path.join(self._base_dir, "temp_file")
with file_io.FileIO(file_path, mode="r+") as f:
f.write("testing1\ntesting2\ntesting3\n\ntesting5")
self.assertEqual("testing1\n", f.readline())
self.assertEqual(9, f.tell())
# Seek to 18
f.seek(18)
self.assertEqual(18, f.tell())
self.assertEqual("testing3\n", f.readline())
# Seek back to 9
f.seek(9)
self.assertEqual(9, f.tell())
self.assertEqual("testing2\n", f.readline())
f.seek(0)
self.assertEqual(0, f.tell())
self.assertEqual("testing1\n", f.readline())
with self.assertRaises(errors.InvalidArgumentError):
f.seek(-1)
with self.assertRaises(TypeError):
f.seek()
# TODO(jhseu): Delete after position deprecation.
with self.assertRaises(TypeError):
f.seek(offset=0, position=0)
f.seek(position=9)
self.assertEqual(9, f.tell())
self.assertEqual("testing2\n", f.readline())
def testSeekFromWhat(self):
file_path = os.path.join(self._base_dir, "temp_file")
with file_io.FileIO(file_path, mode="r+") as f:
f.write("testing1\ntesting2\ntesting3\n\ntesting5")
self.assertEqual("testing1\n", f.readline())
self.assertEqual(9, f.tell())
# Seek to 18
f.seek(9, 1)
self.assertEqual(18, f.tell())
self.assertEqual("testing3\n", f.readline())
# Seek back to 9
f.seek(9, 0)
self.assertEqual(9, f.tell())
self.assertEqual("testing2\n", f.readline())
f.seek(-f.size(), 2)
self.assertEqual(0, f.tell())
self.assertEqual("testing1\n", f.readline())
with self.assertRaises(errors.InvalidArgumentError):
f.seek(0, 3)
def testReadingIterator(self):
file_path = os.path.join(self._base_dir, "temp_file")
data = ["testing1\n", "testing2\n", "testing3\n", "\n", "testing5"]
with file_io.FileIO(file_path, mode="r+") as f:
f.write("".join(data))
actual_data = []
for line in f:
actual_data.append(line)
self.assertSequenceEqual(actual_data, data)
def testReadlines(self):
file_path = os.path.join(self._base_dir, "temp_file")
data = ["testing1\n", "testing2\n", "testing3\n", "\n", "testing5"]
f = file_io.FileIO(file_path, mode="r+")
f.write("".join(data))
f.flush()
lines = f.readlines()
self.assertSequenceEqual(lines, data)
def testUTF8StringPath(self):
file_path = os.path.join(self._base_dir, "UTF8测试_file")
file_io.write_string_to_file(file_path, "testing")
with file_io.FileIO(file_path, mode="rb") as f:
self.assertEqual(b"testing", f.read())
def testEof(self):
"""Test that reading past EOF does not raise an exception."""
file_path = os.path.join(self._base_dir, "temp_file")
f = file_io.FileIO(file_path, mode="r+")
content = "testing"
f.write(content)
f.flush()
self.assertEqual(content, f.read(len(content) + 1))
def testUTF8StringPathExists(self):
file_path = os.path.join(self._base_dir, "UTF8测试_file_exist")
file_io.write_string_to_file(file_path, "testing")
v = file_io.file_exists(file_path)
self.assertEqual(v, True)
def testFilecmp(self):
file1 = os.path.join(self._base_dir, "file1")
file_io.write_string_to_file(file1, "This is a sentence\n" * 100)
file2 = os.path.join(self._base_dir, "file2")
file_io.write_string_to_file(file2, "This is another sentence\n" * 100)
file3 = os.path.join(self._base_dir, "file3")
file_io.write_string_to_file(file3, u"This is another sentence\n" * 100)
self.assertFalse(file_io.filecmp(file1, file2))
self.assertTrue(file_io.filecmp(file2, file3))
def testFilecmpSameSize(self):
file1 = os.path.join(self._base_dir, "file1")
file_io.write_string_to_file(file1, "This is a sentence\n" * 100)
file2 = os.path.join(self._base_dir, "file2")
file_io.write_string_to_file(file2, "This is b sentence\n" * 100)
file3 = os.path.join(self._base_dir, "file3")
file_io.write_string_to_file(file3, u"This is b sentence\n" * 100)
self.assertFalse(file_io.filecmp(file1, file2))
self.assertTrue(file_io.filecmp(file2, file3))
def testFilecmpBinary(self):
file1 = os.path.join(self._base_dir, "file1")
file_io.FileIO(file1, "wb").write("testing\n\na")
file2 = os.path.join(self._base_dir, "file2")
file_io.FileIO(file2, "wb").write("testing\n\nb")
file3 = os.path.join(self._base_dir, "file3")
file_io.FileIO(file3, "wb").write("testing\n\nb")
file4 = os.path.join(self._base_dir, "file4")
file_io.FileIO(file4, "wb").write("testing\n\ntesting")
self.assertFalse(file_io.filecmp(file1, file2))
self.assertFalse(file_io.filecmp(file1, file4))
self.assertTrue(file_io.filecmp(file2, file3))
def testFileCrc32(self):
file1 = os.path.join(self._base_dir, "file1")
file_io.write_string_to_file(file1, "This is a sentence\n" * 100)
crc1 = file_io.file_crc32(file1)
file2 = os.path.join(self._base_dir, "file2")
file_io.write_string_to_file(file2, "This is another sentence\n" * 100)
crc2 = file_io.file_crc32(file2)
file3 = os.path.join(self._base_dir, "file3")
file_io.write_string_to_file(file3, "This is another sentence\n" * 100)
crc3 = file_io.file_crc32(file3)
self.assertTrue(crc1 != crc2)
self.assertEqual(crc2, crc3)
def testFileCrc32WithBytes(self):
file1 = os.path.join(self._base_dir, "file1")
file_io.write_string_to_file(file1, "This is a sentence\n" * 100)
crc1 = file_io.file_crc32(file1, block_size=24)
file2 = os.path.join(self._base_dir, "file2")
file_io.write_string_to_file(file2, "This is another sentence\n" * 100)
crc2 = file_io.file_crc32(file2, block_size=24)
file3 = os.path.join(self._base_dir, "file3")
file_io.write_string_to_file(file3, "This is another sentence\n" * 100)
crc3 = file_io.file_crc32(file3, block_size=-1)
self.assertTrue(crc1 != crc2)
self.assertEqual(crc2, crc3)
def testFileCrc32Binary(self):
file1 = os.path.join(self._base_dir, "file1")
file_io.FileIO(file1, "wb").write("testing\n\n")
crc1 = file_io.file_crc32(file1)
file2 = os.path.join(self._base_dir, "file2")
file_io.FileIO(file2, "wb").write("testing\n\n\n")
crc2 = file_io.file_crc32(file2)
file3 = os.path.join(self._base_dir, "file3")
file_io.FileIO(file3, "wb").write("testing\n\n\n")
crc3 = file_io.file_crc32(file3)
self.assertTrue(crc1 != crc2)
self.assertEqual(crc2, crc3)
def testMatchingFilesPermission(self):
# Create top level directory test_dir.
dir_path = os.path.join(self._base_dir, "test_dir")
file_io.create_dir(dir_path)
# Create second level directories `noread` and `any`.
noread_path = os.path.join(dir_path, "noread")
file_io.create_dir(noread_path)
any_path = os.path.join(dir_path, "any")
file_io.create_dir(any_path)
files = ["file1.txt", "file2.txt", "file3.txt"]
for name in files:
file_path = os.path.join(any_path, name)
file_io.FileIO(file_path, mode="w").write("testing")
file_path = os.path.join(noread_path, "file4.txt")
file_io.FileIO(file_path, mode="w").write("testing")
# Change noread to noread access.
os.chmod(noread_path, 0)
expected_match = [os.path.join(any_path, name) for name in files]
self.assertItemsEqual(
file_io.get_matching_files(os.path.join(dir_path, "*", "file*.txt")),
expected_match)
# Change noread back so that it could be cleaned during tearDown.
os.chmod(noread_path, 0o777)
def testFileSeekableWithZip(self):
# Note: Test case for GitHub issue 27276, issue only exposed in python 3.7+.
filename = os.path.join(self._base_dir, "a.npz")
np.savez_compressed(filename, {"a": 1, "b": 2})
with gfile.GFile(filename, "rb") as f:
info = np.load(f, allow_pickle=True)
_ = [i for i in info.items()]
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/lib/io/file_io_test.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Operations for working with string Tensors."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.compat import compat
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import sparse_tensor
from tensorflow.python.framework import tensor_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gen_parsing_ops
from tensorflow.python.ops import gen_string_ops
from tensorflow.python.ops import math_ops
# go/tf-wildcard-import
# pylint: disable=wildcard-import
# pylint: disable=g-bad-import-order
from tensorflow.python.ops.gen_string_ops import *
from tensorflow.python.util import compat as util_compat
from tensorflow.python.util import deprecation
from tensorflow.python.util import dispatch
from tensorflow.python.util.tf_export import tf_export
# pylint: enable=g-bad-import-order
# pylint: enable=wildcard-import
# pylint: disable=redefined-builtin
@tf_export("strings.regex_full_match")
@dispatch.add_dispatch_support
def regex_full_match(input, pattern, name=None):
r"""Match elements of `input` with regex `pattern`.
Args:
input: string `Tensor`, the source strings to process.
pattern: string or scalar string `Tensor`, regular expression to use,
see more details at https://github.com/google/re2/wiki/Syntax
name: Name of the op.
Returns:
bool `Tensor` of the same shape as `input` with match results.
"""
# TODO(b/112455102): Remove compat.forward_compatible once past the horizon.
if not compat.forward_compatible(2018, 11, 10):
return gen_string_ops.regex_full_match(
input=input, pattern=pattern, name=name)
if isinstance(pattern, util_compat.bytes_or_text_types):
# When `pattern` is static through the life of the op we can
# use a version which performs the expensive regex compilation once at
# creation time.
return gen_string_ops.static_regex_full_match(
input=input, pattern=pattern, name=name)
return gen_string_ops.regex_full_match(
input=input, pattern=pattern, name=name)
regex_full_match.__doc__ = gen_string_ops.regex_full_match.__doc__
@tf_export(
"strings.regex_replace", v1=["strings.regex_replace", "regex_replace"])
@deprecation.deprecated_endpoints("regex_replace")
@dispatch.add_dispatch_support
def regex_replace(input, pattern, rewrite, replace_global=True, name=None):
r"""Replace elements of `input` matching regex `pattern` with `rewrite`.
Args:
input: string `Tensor`, the source strings to process.
pattern: string or scalar string `Tensor`, regular expression to use,
see more details at https://github.com/google/re2/wiki/Syntax
rewrite: string or scalar string `Tensor`, value to use in match
replacement, supports backslash-escaped digits (\1 to \9) can be to insert
text matching corresponding parenthesized group.
replace_global: `bool`, if `True` replace all non-overlapping matches,
else replace only the first match.
name: A name for the operation (optional).
Returns:
string `Tensor` of the same shape as `input` with specified replacements.
"""
if (isinstance(pattern, util_compat.bytes_or_text_types) and
isinstance(rewrite, util_compat.bytes_or_text_types)):
# When `pattern` and `rewrite` are static through the life of the op we can
# use a version which performs the expensive regex compilation once at
# creation time.
return gen_string_ops.static_regex_replace(
input=input, pattern=pattern,
rewrite=rewrite, replace_global=replace_global,
name=name)
return gen_string_ops.regex_replace(
input=input, pattern=pattern,
rewrite=rewrite, replace_global=replace_global,
name=name)
@tf_export("strings.format")
def string_format(template, inputs, placeholder="{}", summarize=3, name=None):
r"""Formats a string template using a list of tensors.
Formats a string template using a list of tensors, abbreviating tensors by
only printing the first and last `summarize` elements of each dimension
(recursively). If formatting only one tensor into a template, the tensor does
not have to be wrapped in a list.
Example:
Formatting a single-tensor template:
```python
sess = tf.compat.v1.Session()
with sess.as_default():
tensor = tf.range(10)
formatted = tf.strings.format("tensor: {}, suffix", tensor)
out = sess.run(formatted)
expected = "tensor: [0 1 2 ... 7 8 9], suffix"
assert(out.decode() == expected)
```
Formatting a multi-tensor template:
```python
sess = tf.compat.v1.Session()
with sess.as_default():
tensor_one = tf.reshape(tf.range(100), [10, 10])
tensor_two = tf.range(10)
formatted = tf.strings.format("first: {}, second: {}, suffix",
(tensor_one, tensor_two))
out = sess.run(formatted)
expected = ("first: [[0 1 2 ... 7 8 9]\n"
" [10 11 12 ... 17 18 19]\n"
" [20 21 22 ... 27 28 29]\n"
" ...\n"
" [70 71 72 ... 77 78 79]\n"
" [80 81 82 ... 87 88 89]\n"
" [90 91 92 ... 97 98 99]], second: [0 1 2 ... 7 8 9], suffix")
assert(out.decode() == expected)
```
Args:
template: A string template to format tensor values into.
inputs: A list of `Tensor` objects, or a single Tensor.
The list of tensors to format into the template string. If a solitary
tensor is passed in, the input tensor will automatically be wrapped as a
list.
placeholder: An optional `string`. Defaults to `{}`.
At each placeholder occurring in the template, a subsequent tensor
will be inserted.
summarize: An optional `int`. Defaults to `3`.
When formatting the tensors, show the first and last `summarize`
entries of each tensor dimension (recursively). If set to -1, all
elements of the tensor will be shown.
name: A name for the operation (optional).
Returns:
A scalar `Tensor` of type `string`.
Raises:
ValueError: if the number of placeholders does not match the number of
inputs.
"""
# If there is only one tensor to format, we will automatically wrap it in a
# list to simplify the user experience
if tensor_util.is_tensor(inputs):
inputs = [inputs]
if template.count(placeholder) != len(inputs):
raise ValueError("%s placeholder(s) in template does not match %s tensor(s)"
" provided as input" % (template.count(placeholder),
len(inputs)))
return gen_string_ops.string_format(inputs,
template=template,
placeholder=placeholder,
summarize=summarize,
name=name)
# Note: tf.strings.split is exported in ragged/ragged_string_ops.py, which
# defines a wrapper for this function.
def string_split(source, sep=None, skip_empty=True, delimiter=None): # pylint: disable=invalid-name
"""Split elements of `source` based on `delimiter` into a `SparseTensor`.
Let N be the size of source (typically N will be the batch size). Split each
element of `source` based on `delimiter` and return a `SparseTensor`
containing the split tokens. Empty tokens are ignored.
If `sep` is an empty string, each element of the `source` is split
into individual strings, each containing one byte. (This includes splitting
multibyte sequences of UTF-8.) If delimiter contains multiple bytes, it is
treated as a set of delimiters with each considered a potential split point.
For example:
N = 2, source[0] is 'hello world' and source[1] is 'a b c', then the output
will be
st.indices = [0, 0;
0, 1;
1, 0;
1, 1;
1, 2]
st.shape = [2, 3]
st.values = ['hello', 'world', 'a', 'b', 'c']
Args:
source: `1-D` string `Tensor`, the strings to split.
sep: `0-D` string `Tensor`, the delimiter character, the string should
be length 0 or 1. Default is ' '.
skip_empty: A `bool`. If `True`, skip the empty strings from the result.
delimiter: deprecated alias for `sep`.
Raises:
ValueError: If delimiter is not a string.
Returns:
A `SparseTensor` of rank `2`, the strings split according to the delimiter.
The first column of the indices corresponds to the row in `source` and the
second column corresponds to the index of the split component in this row.
"""
delimiter = deprecation.deprecated_argument_lookup(
"sep", sep, "delimiter", delimiter)
if delimiter is None:
delimiter = " "
delimiter = ops.convert_to_tensor(delimiter, dtype=dtypes.string)
source = ops.convert_to_tensor(source, dtype=dtypes.string)
indices, values, shape = gen_string_ops.string_split(
source, delimiter=delimiter, skip_empty=skip_empty)
indices.set_shape([None, 2])
values.set_shape([None])
shape.set_shape([2])
return sparse_tensor.SparseTensor(indices, values, shape)
# Note: tf.strings.split is exported in ragged/ragged_string_ops.py, which
# defines a wrapper for this function.
def string_split_v2(source, sep=None, maxsplit=-1):
"""Split elements of `source` based on `sep` into a `SparseTensor`.
Let N be the size of source (typically N will be the batch size). Split each
element of `source` based on `sep` and return a `SparseTensor`
containing the split tokens. Empty tokens are ignored.
For example, N = 2, source[0] is 'hello world' and source[1] is 'a b c',
then the output will be
st.indices = [0, 0;
0, 1;
1, 0;
1, 1;
1, 2]
st.shape = [2, 3]
st.values = ['hello', 'world', 'a', 'b', 'c']
If `sep` is given, consecutive delimiters are not grouped together and are
deemed to delimit empty strings. For example, source of `"1<>2<><>3"` and
sep of `"<>"` returns `["1", "2", "", "3"]`. If `sep` is None or an empty
string, consecutive whitespace are regarded as a single separator, and the
result will contain no empty strings at the start or end if the string has
leading or trailing whitespace.
Note that the above mentioned behavior matches python's str.split.
Args:
source: `1-D` string `Tensor`, the strings to split.
sep: `0-D` string `Tensor`, the delimiter character.
maxsplit: An `int`. If `maxsplit > 0`, limit of the split of the result.
Raises:
ValueError: If sep is not a string.
Returns:
A `SparseTensor` of rank `2`, the strings split according to the delimiter.
The first column of the indices corresponds to the row in `source` and the
second column corresponds to the index of the split component in this row.
"""
if sep is None:
sep = ""
sep = ops.convert_to_tensor(sep, dtype=dtypes.string)
source = ops.convert_to_tensor(source, dtype=dtypes.string)
indices, values, shape = gen_string_ops.string_split_v2(
source, sep=sep, maxsplit=maxsplit)
indices.set_shape([None, 2])
values.set_shape([None])
shape.set_shape([2])
return sparse_tensor.SparseTensor(indices, values, shape)
def _reduce_join_reduction_dims(x, axis, reduction_indices):
"""Returns range(rank(x) - 1, 0, -1) if reduction_indices is None."""
# TODO(aselle): Remove this after deprecation
if reduction_indices is not None:
if axis is not None:
raise ValueError("Can't specify both 'axis' and 'reduction_indices'.")
axis = reduction_indices
if axis is not None:
return axis
else:
# Fast path: avoid creating Rank and Range ops if ndims is known.
if x.get_shape().ndims is not None:
return constant_op.constant(
np.arange(x.get_shape().ndims - 1, -1, -1), dtype=dtypes.int32)
# Otherwise, we rely on Range and Rank to do the right thing at run-time.
return math_ops.range(array_ops.rank(x) - 1, -1, -1)
@tf_export(v1=["strings.reduce_join", "reduce_join"])
@deprecation.deprecated_endpoints("reduce_join")
def reduce_join(inputs, axis=None, # pylint: disable=missing-docstring
keep_dims=False,
separator="",
name=None,
reduction_indices=None,
keepdims=None):
keep_dims = deprecation.deprecated_argument_lookup(
"keepdims", keepdims, "keep_dims", keep_dims)
inputs_t = ops.convert_to_tensor(inputs)
reduction_indices = _reduce_join_reduction_dims(
inputs_t, axis, reduction_indices)
return gen_string_ops.reduce_join(
inputs=inputs_t,
reduction_indices=reduction_indices,
keep_dims=keep_dims,
separator=separator,
name=name)
@tf_export("strings.reduce_join", v1=[])
def reduce_join_v2( # pylint: disable=missing-docstring
inputs,
axis=None,
keepdims=False,
separator="",
name=None):
return reduce_join(
inputs, axis, keep_dims=keepdims, separator=separator, name=name)
reduce_join.__doc__ = deprecation.rewrite_argument_docstring(
gen_string_ops.reduce_join.__doc__, "reduction_indices", "axis")
reduce_join.__doc__ = reduce_join.__doc__.replace("tf.reduce_join(",
"tf.strings.reduce_join(")
# This wrapper provides backwards compatibility for code that predates the
# unit argument and that passed 'name' as a positional argument.
@tf_export(v1=["strings.length"])
@dispatch.add_dispatch_support
def string_length(input, name=None, unit="BYTE"):
return gen_string_ops.string_length(input, unit=unit, name=name)
@tf_export("strings.length", v1=[])
@dispatch.add_dispatch_support
def string_length_v2(input, unit="BYTE", name=None):
return string_length(input, name, unit)
string_length.__doc__ = gen_string_ops.string_length.__doc__
@tf_export(v1=["substr"])
@deprecation.deprecated(None, "Use `tf.strings.substr` instead of `tf.substr`.")
def substr_deprecated(input, pos, len, name=None, unit="BYTE"):
return substr(input, pos, len, name=name, unit=unit)
substr_deprecated.__doc__ = gen_string_ops.substr.__doc__
@tf_export(v1=["strings.substr"])
@dispatch.add_dispatch_support
def substr(input, pos, len, name=None, unit="BYTE"):
return gen_string_ops.substr(input, pos, len, unit=unit, name=name)
substr.__doc__ = gen_string_ops.substr.__doc__
@tf_export("strings.substr", v1=[])
@dispatch.add_dispatch_support
def substr_v2(input, pos, len, unit="BYTE", name=None):
return gen_string_ops.substr(input, pos, len, unit=unit, name=name)
substr_v2.__doc__ = gen_string_ops.substr.__doc__
ops.NotDifferentiable("RegexReplace")
ops.NotDifferentiable("StringToHashBucket")
ops.NotDifferentiable("StringToHashBucketFast")
ops.NotDifferentiable("StringToHashBucketStrong")
ops.NotDifferentiable("ReduceJoin")
ops.NotDifferentiable("StringJoin")
ops.NotDifferentiable("StringSplit")
ops.NotDifferentiable("AsString")
ops.NotDifferentiable("EncodeBase64")
ops.NotDifferentiable("DecodeBase64")
@tf_export("strings.to_number", v1=[])
@dispatch.add_dispatch_support
def string_to_number(input, out_type=dtypes.float32, name=None):
r"""Converts each string in the input Tensor to the specified numeric type.
(Note that int32 overflow results in an error while float overflow
results in a rounded value.)
Args:
input: A `Tensor` of type `string`.
out_type: An optional `tf.DType` from: `tf.float32, tf.float64, tf.int32,
tf.int64`. Defaults to `tf.float32`.
The numeric type to interpret each string in `string_tensor` as.
name: A name for the operation (optional).
Returns:
A `Tensor` of type `out_type`.
"""
return gen_parsing_ops.string_to_number(input, out_type, name)
@tf_export(v1=["strings.to_number", "string_to_number"])
def string_to_number_v1(
string_tensor=None,
out_type=dtypes.float32,
name=None,
input=None):
string_tensor = deprecation.deprecated_argument_lookup(
"input", input, "string_tensor", string_tensor)
return gen_parsing_ops.string_to_number(string_tensor, out_type, name)
string_to_number_v1.__doc__ = gen_parsing_ops.string_to_number.__doc__
@tf_export("strings.to_hash_bucket", v1=[])
@dispatch.add_dispatch_support
def string_to_hash_bucket(input, num_buckets, name=None):
# pylint: disable=line-too-long
r"""Converts each string in the input Tensor to its hash mod by a number of buckets.
The hash function is deterministic on the content of the string within the
process.
Note that the hash function may change from time to time.
This functionality will be deprecated and it's recommended to use
`tf.strings.to_hash_bucket_fast()` or `tf.strings.to_hash_bucket_strong()`.
Args:
input: A `Tensor` of type `string`.
num_buckets: An `int` that is `>= 1`. The number of buckets.
name: A name for the operation (optional).
Returns:
A `Tensor` of type `int64`.
"""
# pylint: enable=line-too-long
return gen_string_ops.string_to_hash_bucket(input, num_buckets, name)
@tf_export(v1=["strings.to_hash_bucket", "string_to_hash_bucket"])
def string_to_hash_bucket_v1(
string_tensor=None,
num_buckets=None,
name=None,
input=None):
string_tensor = deprecation.deprecated_argument_lookup(
"input", input, "string_tensor", string_tensor)
return gen_string_ops.string_to_hash_bucket(string_tensor, num_buckets, name)
string_to_hash_bucket_v1.__doc__ = gen_string_ops.string_to_hash_bucket.__doc__
|
tensorflow-master
|
tensorflow/python/ops/string_ops.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
# pylint: disable=unused-import
"""Import names of Tensor Flow standard Ops."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import sys as _sys
from tensorflow.python import autograph
# pylint: disable=g-bad-import-order
# Imports the following modules so that @RegisterGradient get executed.
from tensorflow.python.ops import array_grad
from tensorflow.python.ops import cudnn_rnn_grad
from tensorflow.python.ops import data_flow_grad
from tensorflow.python.ops import manip_grad
from tensorflow.python.ops import math_grad
from tensorflow.python.ops import random_grad
from tensorflow.python.ops import sparse_grad
from tensorflow.python.ops import state_grad
from tensorflow.python.ops import tensor_array_grad
# go/tf-wildcard-import
# pylint: disable=wildcard-import
from tensorflow.python.ops.array_ops import *
from tensorflow.python.ops.check_ops import *
from tensorflow.python.ops.clip_ops import *
from tensorflow.python.ops.special_math_ops import *
# TODO(vrv): Switch to import * once we're okay with exposing the module.
from tensorflow.python.ops.confusion_matrix import confusion_matrix
from tensorflow.python.ops.control_flow_ops import Assert
from tensorflow.python.ops.control_flow_ops import case
from tensorflow.python.ops.control_flow_ops import cond
from tensorflow.python.ops.control_flow_ops import group
from tensorflow.python.ops.control_flow_ops import no_op
from tensorflow.python.ops.control_flow_ops import tuple # pylint: disable=redefined-builtin
# pylint: enable=redefined-builtin
from tensorflow.python.eager import wrap_function
from tensorflow.python.ops.control_flow_ops import while_loop
from tensorflow.python.ops.batch_ops import *
from tensorflow.python.ops.critical_section_ops import *
from tensorflow.python.ops.data_flow_ops import *
from tensorflow.python.ops.functional_ops import *
from tensorflow.python.ops.gradients import *
from tensorflow.python.ops.histogram_ops import *
from tensorflow.python.ops.init_ops import *
from tensorflow.python.ops.io_ops import *
from tensorflow.python.ops.linalg_ops import *
from tensorflow.python.ops.logging_ops import Print
from tensorflow.python.ops.logging_ops import get_summary_op
from tensorflow.python.ops.logging_ops import timestamp
from tensorflow.python.ops.lookup_ops import initialize_all_tables
from tensorflow.python.ops.lookup_ops import tables_initializer
from tensorflow.python.ops.manip_ops import *
from tensorflow.python.ops.math_ops import *
from tensorflow.python.ops.numerics import *
from tensorflow.python.ops.parsing_ops import *
from tensorflow.python.ops.partitioned_variables import *
from tensorflow.python.ops.proto_ops import *
from tensorflow.python.ops.ragged import ragged_dispatch as _ragged_dispatch
from tensorflow.python.ops.ragged import ragged_operators as _ragged_operators
from tensorflow.python.ops.random_ops import *
from tensorflow.python.ops.script_ops import py_func
from tensorflow.python.ops.session_ops import *
from tensorflow.python.ops.sort_ops import *
from tensorflow.python.ops.sparse_ops import *
from tensorflow.python.ops.state_ops import assign
from tensorflow.python.ops.state_ops import assign_add
from tensorflow.python.ops.state_ops import assign_sub
from tensorflow.python.ops.state_ops import count_up_to
from tensorflow.python.ops.state_ops import scatter_add
from tensorflow.python.ops.state_ops import scatter_div
from tensorflow.python.ops.state_ops import scatter_mul
from tensorflow.python.ops.state_ops import scatter_sub
from tensorflow.python.ops.state_ops import scatter_min
from tensorflow.python.ops.state_ops import scatter_max
from tensorflow.python.ops.state_ops import scatter_update
from tensorflow.python.ops.state_ops import scatter_nd_add
from tensorflow.python.ops.state_ops import scatter_nd_sub
# TODO(simister): Re-enable once binary size increase due to scatter_nd
# ops is under control.
# from tensorflow.python.ops.state_ops import scatter_nd_mul
# from tensorflow.python.ops.state_ops import scatter_nd_div
from tensorflow.python.ops.state_ops import scatter_nd_update
from tensorflow.python.ops.stateless_random_ops import *
from tensorflow.python.ops.string_ops import *
from tensorflow.python.ops.template import *
from tensorflow.python.ops.tensor_array_ops import *
from tensorflow.python.ops.variable_scope import *
from tensorflow.python.ops.variables import *
from tensorflow.python.ops.parallel_for.control_flow_ops import vectorized_map
# pylint: enable=wildcard-import
# pylint: enable=g-bad-import-order
# These modules were imported to set up RaggedTensor operators and dispatchers:
del _ragged_dispatch, _ragged_operators
|
tensorflow-master
|
tensorflow/python/ops/standard_ops.py
|
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Utilities for managing state of v1 control flow for computing gradients."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import tensor_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import control_flow_util as util
from tensorflow.python.ops import default_gradient
from tensorflow.python.ops import gen_data_flow_ops
from tensorflow.python.ops import gen_resource_variable_ops
# pylint: disable=protected-access
def _GetMaxSizeFromNestedMaximumIterations(value, while_ctxt):
"""Calculate a max_size for use by stack ops inside an XLA while_loop.
Args:
value: The value inside the while_loop forward context. Used for printing
error messages.
while_ctxt: The forward context inside which value resides. This does not
always match the value's immediate context, as `value` may be inside e.g.
a cond context inside the while_loop.
Returns:
A tensor containing the `max_size` to feed to a Stack initializer.
Raises:
ValueError: If `value` is nested inside a `while_loop` that either
lacks a `maximum_iterations` parameter, or the `maximum_iterations`
parameter:
- is inside a `while_loop` that is a parent of the calling context, and
- cannot be evaluated at graph build time to a constant.
"""
value_name = value.name
# curr_ctxt is the context that tf.gradients was called in.
curr_ctxt = ops.get_default_graph()._get_control_flow_context() # pylint: disable=protected-access
curr_ctxt_name = curr_ctxt.name if curr_ctxt is not None else ""
max_size = constant_op.constant(1)
# Loop through all containing while contexts between value and the
# current context, multiplying together each context's
# max_iterations to get the maximum stack size.
while while_ctxt not in (None, curr_ctxt):
max_iter = while_ctxt.maximum_iterations
if max_iter is None:
raise ValueError(
"Cannot create a gradient accumulator for tensor '%s' inside "
"XLA while_loop because maximum_iterations was not passed to "
"the tf.while_loop call ('%s')." % (value_name, while_ctxt.name))
# pylint: disable=protected-access
max_iter_ctxt = max_iter.op._get_control_flow_context()
# pylint: enable=protected-access
# If max_iter_ctxt (non-strictly) contains curr_ctxt, then it's OK to use.
if util.IsContainingContext(curr_ctxt, max_iter_ctxt):
max_size *= max_iter
else:
# We cannot use max_iter because it's defined in a nested while
# or cond context, so will fail if we try to use it as input to
# any ops in curr_ctxt (e.g. max_size or the final accumulator
# stack). Attempt to get a constant value out to use instead.
const_max_iter = tensor_util.constant_value(max_iter)
if const_max_iter is None:
raise ValueError(
"Cannot create a gradient accumulator for tensor '%s' inside XLA "
"while_loop. maximum_iterations tensor '%s' for while_loop context "
"'%s' must be statically known (e.g. a constant value or known "
"shape dimension), or be defined at or outside the while loop "
"context '%s' (currently defined in '%s')." %
(value_name, max_iter.name, while_ctxt.name, curr_ctxt_name,
max_iter_ctxt.name))
max_size *= const_max_iter
# Find the next outer WhileContext (or stop if we reach the
# tf.gradient's context).
while_ctxt = util.GetContainingWhileContext(
while_ctxt.outer_context, stop_ctxt=curr_ctxt)
return max_size
class _GradLoopState(object):
"""The state used for constructing the gradient graph for a while loop.
We create a _GradLoopState for each while loop in forward and its
corresponding while loop in backprop. This gives us access to both
the forward and the backprop WhileContexts.
During the construction of gradient graph, any time when we detect
a forward value that is needed for backprop, we create a history
accumulator and add it to `history_map`. Any time when we backprop
a loop switch op (in _SwitchGrad), we add the grad merge op in
`switch_map`.
"""
def __init__(self, forward_ctxt, outer_grad_state):
# The grad loop state for the outer while loop.
self._outer_grad_state = None
# The while loop context for forward.
self._forward_context = None
# The loop counter added by AddForwardLoopCounter. It is the value
# of the loop counter for the next iteration.
self._forward_index = None
# A sync op for forward.
self._forward_sync = None
# The while loop context for backprop.
self._grad_context = None
# The loop counter added by AddBackpropLoopCounter. It is the value
# of the loop counter for the current iteration.
self._grad_index = None
# A sync op for backprop.
self._grad_sync = None
# Information needed by backprop.
self._history_map = {}
self._switch_map = {}
self._unused_exits = []
self._deferred_exits = []
self._forward_loop_exits = list(forward_ctxt.loop_exits)
self._pending_exits_count = len(forward_ctxt.loop_exits)
self._outer_grad_state = outer_grad_state
if outer_grad_state:
outer_forward_ctxt = outer_grad_state.forward_context
else:
if not hasattr(forward_ctxt, "outer_context"):
raise ValueError("Failed to call gradients on a while loop without"
"properly serializing graph via MetaGraphDef")
outer_forward_ctxt = forward_ctxt.outer_context
# Add the forward loop counter.
with forward_ctxt._graph.as_default(): # pylint: disable=protected-access
if outer_forward_ctxt:
outer_forward_ctxt.Enter()
cnt, forward_index = forward_ctxt.AddForwardLoopCounter(outer_grad_state)
if outer_forward_ctxt:
outer_forward_ctxt.Exit()
self._forward_context = forward_ctxt
self._forward_index = forward_index
# Add the backprop WhileContext, and the backprop loop counter.
if outer_grad_state:
# This is a nested loop. Remember the iteration counts for each
# execution of this inner loop.
outer_forward_ctxt.AddName(cnt.name)
history_cnt = outer_grad_state.AddForwardAccumulator(cnt)
outer_grad_ctxt = outer_grad_state.grad_context
outer_grad_ctxt.Enter()
self._grad_context = control_flow_ops.WhileContext(
maximum_iterations=forward_ctxt.maximum_iterations,
parallel_iterations=forward_ctxt.parallel_iterations,
back_prop=forward_ctxt.back_prop,
swap_memory=forward_ctxt.swap_memory,
name=forward_ctxt.name,
grad_state=self)
real_cnt = outer_grad_state.AddBackpropAccumulatedValue(history_cnt, cnt)
self._grad_index = self._grad_context.AddBackpropLoopCounter(
real_cnt, outer_grad_state)
outer_grad_ctxt.Exit()
else:
if outer_forward_ctxt:
outer_forward_ctxt.Enter()
self._grad_context = control_flow_ops.WhileContext(
maximum_iterations=forward_ctxt.maximum_iterations,
parallel_iterations=forward_ctxt.parallel_iterations,
back_prop=forward_ctxt.back_prop,
swap_memory=forward_ctxt.swap_memory,
name=forward_ctxt.name,
grad_state=self)
self._grad_index = self._grad_context.AddBackpropLoopCounter(
cnt, outer_grad_state)
if outer_forward_ctxt:
outer_forward_ctxt.Exit()
@property
def outer_grad_state(self):
"""The grad loop state for outer loop."""
return self._outer_grad_state
@property
def forward_context(self):
"""The while loop context for forward."""
return self._forward_context
@property
def forward_index(self):
"""The loop index of forward loop."""
return self._forward_index
@property
def forward_sync(self):
"""A control trigger node for synchronization in the forward loop.
One main use is to keep the push ops of a stack executed in the
iteration order.
"""
if self._forward_sync is None:
with ops.control_dependencies(None):
self._forward_sync = control_flow_ops.control_trigger(name="f_sync")
self._forward_sync._set_control_flow_context(self._forward_context)
self._forward_index.op._add_control_input(self._forward_sync)
return self._forward_sync
@property
def grad_context(self):
"""The corresponding WhileContext for gradient."""
return self._grad_context
@property
def grad_index(self):
"""The loop index of backprop loop."""
return self._grad_index
@property
def grad_sync(self):
"""A control trigger node for synchronization in the grad loop.
One main use is to keep the pop ops of a stack executed in the
iteration order.
"""
if self._grad_sync is None:
with ops.control_dependencies(None):
self._grad_sync = control_flow_ops.control_trigger(name="b_sync")
self._grad_sync._set_control_flow_context(self._grad_context)
self._grad_index.op._add_control_input(self._grad_sync)
if self._grad_context.outer_context:
self._grad_context.outer_context.AddInnerOp(self._grad_sync)
return self._grad_sync
@property
def history_map(self):
"""The map that records all the tensors needed for backprop."""
return self._history_map
@property
def switch_map(self):
"""The map that records all the Switch ops for the while loop."""
return self._switch_map
@property
def unused_exits(self):
"""The list of "unused" exits."""
return self._unused_exits
@property
def deferred_exits(self):
"""The list of "deferred" exits."""
return self._deferred_exits
@property
def forward_loop_exits(self):
"""The list of exits of the forward loop."""
return self._forward_loop_exits
@property
def pending_exits_count(self):
"""The number of exits we expect to see but haven't."""
return self._pending_exits_count
@pending_exits_count.setter
def pending_exits_count(self, cnt):
"""Set the pending count to cnt."""
self._pending_exits_count = cnt
def AddForwardAccumulator(self, value, dead_branch=False):
"""Add an accumulator for each forward tensor that is needed in backprop.
This is added to the forward loop at the first time when a tensor
in the forward loop is used by backprop gradient computation loop.
We create an accumulator that accumulates the value of tensor at each
iteration. Called in the control flow context where gradients() is called.
The pseudocode is:
```
acc = stack();
while (_pivot) {
acc = stack_push(acc, value);
}
```
We make sure that the stack push op in one iteration is executed before
next iteration. This is achieved by adding a control edge from
`forward_index.op.inputs[0].op` to the push op, and another control
edge from the push op to either `forward_index.op` or `forward_sync`.
Args:
value: The source tensor in forward that is to be accumulated.
dead_branch: True iff the tensor is on a dead branch of a cond.
Returns:
The stack that contains the accumulated history of the tensor.
Raises:
TypeError: For internal errors involving the value condition context.
ValueError: If `value` is inside a XLA scope and a valid max size
for the stack can't be found.
"""
# curr_ctxt is the context that tf.gradients was called in.
with self._forward_index.graph.as_default():
curr_ctxt = ops.get_default_graph()._get_control_flow_context() # pylint: disable=protected-access
with ops.control_dependencies(None):
if curr_ctxt:
curr_ctxt.Enter()
with ops.colocate_with(value):
# We only need to pass maximum_iterations to the stack if
# we're inside an XLA context.
if not util.IsInXLAContext(value.op):
max_size = constant_op.constant(-1, dtypes.int32)
else:
max_size = _GetMaxSizeFromNestedMaximumIterations(
value, self.forward_context)
acc = gen_data_flow_ops.stack_v2(
max_size=max_size, elem_type=value.dtype.base_dtype, name="f_acc")
if curr_ctxt:
curr_ctxt.Exit()
# Make acc available in the forward context.
enter_acc = self.forward_context.AddValue(acc)
# Add the stack_push op in the context of value.op.
swap_enabled = self.forward_context.swap_memory
value_ctxt = util.GetOutputContext(value.op)
if value_ctxt == self.forward_context:
# value is not nested in the forward context.
self.forward_context.Enter()
push = gen_data_flow_ops.stack_push_v2(
enter_acc, value, swap_memory=swap_enabled)
self.forward_context.Exit()
# Protect stack push and order it before forward_index.
self.forward_index.op._add_control_input(push.op)
else:
# value is in a cond context within the forward context.
if not isinstance(value_ctxt, control_flow_ops.CondContext):
raise TypeError("value_ctxt is not a CondContext: %s" % value_ctxt)
if dead_branch:
# The special case for creating a zero tensor for a dead
# branch of a switch. See _ControlFlowState.ZerosLike().
value_ctxt.outer_context.Enter()
push = gen_data_flow_ops.stack_push_v2(
enter_acc, value, swap_memory=swap_enabled)
value_ctxt.outer_context.Exit()
push.op._set_control_flow_context(value_ctxt)
else:
value_ctxt.Enter()
push = gen_data_flow_ops.stack_push_v2(
enter_acc, value, swap_memory=swap_enabled)
value_ctxt.Exit()
# Protect stack push and order it before forward_sync.
self.forward_sync._add_control_input(push.op)
# Order stack push after the successor of forward_index
add_op = self.forward_index.op.inputs[0].op
push.op._add_control_input(add_op)
return acc
def AddBackpropAccumulatedValue(self, history_value, value,
dead_branch=False):
"""Add the getter for an accumulated value in the grad context.
This is added to the backprop loop. Called in the grad context to
get the value of an accumulated value. The stack pop op must be guarded
by the pred of the controlling cond.
Args:
history_value: The history (a stack) of a value.
value: The value that is pushed onto the stack.
dead_branch: True iff the tensor is on a dead branch of a cond.
Returns:
The current value (the top of the stack).
"""
history_ctxt = history_value.op._get_control_flow_context()
# Find the cond context that controls history_value if any.
cond_ctxt = None
value_ctxt = value.op._get_control_flow_context()
while value_ctxt and value_ctxt != history_ctxt:
if isinstance(value_ctxt, control_flow_ops.CondContext):
cond_ctxt = value_ctxt
break
value_ctxt = value_ctxt.outer_context
with ops.control_dependencies(None):
self.grad_context.Enter()
if cond_ctxt:
# Guard stack pop with a switch if it is controlled by a cond.
grad_state = self
pred = None
while pred is None and grad_state:
pred = grad_state.history_map.get(cond_ctxt.pred.name)
grad_state = grad_state.outer_grad_state
if pred is None:
pred = cond_ctxt.pred
branch = (1 - cond_ctxt.branch) if dead_branch else cond_ctxt.branch
history_value = control_flow_ops._SwitchRefOrTensor(
history_value, pred)[branch]
pop = gen_data_flow_ops.stack_pop_v2(history_value,
value.dtype.base_dtype)
pop.set_shape(value.get_shape())
self.grad_context.Exit()
parallel_iterations = self.grad_context.parallel_iterations
if parallel_iterations > 1:
# All pops are ordered after pivot_for_body and before grad_sync.
self.grad_sync._add_control_input(pop.op)
return pop
def GetRealValue(self, value):
"""Get the real value of `value`.
If backprop "uses" a value produced by forward inference, an accumulator
is added in the forward loop to accumulate its values. We use the
accumulated value. This method must be called in the grad loop context.
`value` must be in forward and needed for backprop.
Args:
value: A tensor to be captured.
Returns:
The same tensor obtained from the saved history.
"""
assert value.op.type not in ["Variable", "VariableV2"]
real_value = self._history_map.get(value.name)
if real_value is None:
cur_value = value
cur_grad_state = self
while True:
enter_op = util.GetLoopConstantEnter(cur_value)
if enter_op:
# Special case: cur_value comes from a constant Enter node.
cur_value = enter_op.inputs[0]
cur_grad_state = cur_grad_state.outer_grad_state
if cur_grad_state is None:
# We are now outside all nested loops for this gradient(),
# so `value` is a loop invariant and there is no need to
# save the history of value. Just make cur_value to enter
# the right control flow context.
real_value = self._grad_context.AddValue(cur_value)
break
elif constant_op.is_constant(cur_value):
# If the value to be forwarded is a constant, clone the constant in
# the gradient loop rather than using a stack.
# TODO(phawkins): consider hoisting the constant out of the loop
# instead.
real_value = constant_op.constant(
tensor_util.constant_value(cur_value), dtype=cur_value.dtype)
break
else:
# Record the history of this value in forward_ctxt.
self._grad_context.Exit()
history_value = cur_grad_state.AddForwardAccumulator(cur_value)
self._grad_context.Enter()
break
if real_value is None:
# Add the stack pop op in the grad context.
real_value = cur_grad_state.AddBackpropAccumulatedValue(
history_value, cur_value)
if cur_grad_state != self:
real_value = self._grad_context.AddValue(real_value)
self._history_map[value.name] = real_value
return real_value
class _ControlFlowState(object):
"""Maintain the mapping from the loops to their grad states."""
def __init__(self):
self._map = {} # maps forward loop context to _GradLoopState
def GetGradState(self, op, before):
"""Return the grad state for this op if it's in a forward loop context."""
if before and util.IsLoopExit(op):
forward_ctxt = op._get_control_flow_context() # pylint: disable=protected-access
forward_ctxt = forward_ctxt.outer_context
if forward_ctxt:
forward_ctxt = forward_ctxt.GetWhileContext()
else:
forward_ctxt = util.GetWhileContext(op)
if forward_ctxt:
return self._map.get(forward_ctxt)
return None
def ProcessUnusedLoopExits(self, pending_count, to_ops_set):
"""Process all the "unused" loop exits.
The "unused" exits of the loops are added to `unused_exits`. An exit is
unused if its pending_count is 0. If there is an exit with real gradient,
all these deferred exits will enter the backprop loop with zero gradient.
Otherwise, they will enter the backprop loop with None. As an example,
people often write:
```python
v1, _ = tf.while_loop(p, b, [x1, x2])
result = gradients(v1, x1)
```
The exit node for x2 is not included by the betweenness analysis. But we
need to backprop x2 if x2 is involved in computing v1.
Args:
pending_count: The number of backprop inputs for every op.
to_ops_set: The set of ops for ys in gradients(ys, xs)
Returns:
The set of unused loop exits that we know at this point we need
to backprop.
"""
loop_exits = []
for grad_state in self._map.values():
for y in grad_state.forward_loop_exits:
if pending_count[y.op] == 0:
grad_state.pending_exits_count -= 1
if y.op not in to_ops_set:
grad_state.unused_exits.append(y)
if grad_state.pending_exits_count == 0:
loop_exits.extend(grad_state.unused_exits)
# Need to include Enters in backprop for higher-order gradients.
for y in grad_state.forward_context.loop_enters:
if pending_count[y.op] == 0:
pending_count[y.op] = 1
return loop_exits
def EnterGradWhileContext(self, op, before):
"""Enter the WhileContext for gradient computation."""
grad_state = self.GetGradState(op, before)
if grad_state:
grad_state.grad_context.Enter()
def ExitGradWhileContext(self, op, before):
"""Exit the WhileContext for gradient computation."""
grad_state = self.GetGradState(op, before)
if grad_state:
grad_state.grad_context.Exit()
def AddWhileContext(self, op, between_op_list, between_ops):
"""Add the grad state for the while loop that op belongs to.
Note that op is an Exit, and this method must be called in
the control flow context where gradients() is called.
Note that this method modifies `between_op_list` and `between_ops`.
"""
forward_ctxt = util.GetWhileContext(op)
grad_state = self._map.get(forward_ctxt)
if grad_state is None:
# This is a new while loop so create a grad state for it.
outer_forward_ctxt = forward_ctxt.outer_context
if outer_forward_ctxt:
outer_forward_ctxt = outer_forward_ctxt.GetWhileContext()
outer_grad_state = None
if outer_forward_ctxt:
outer_grad_state = self._map.get(outer_forward_ctxt)
grad_state = _GradLoopState(forward_ctxt, outer_grad_state)
self._map[forward_ctxt] = grad_state
# We need to include all exits of a loop for backprop.
for loop_exit in grad_state.forward_loop_exits:
if loop_exit.op not in between_ops:
between_ops.add(loop_exit.op)
between_op_list.append(loop_exit.op)
def ZerosLikeForExit(self, val):
"""Create zeros_like gradient for a loop exit.
If the result of a loop variable is not used but is involved in
computing the result of some needed loop variable, we create a
zero-valued tensor that is fed as gradient for the Exit node of that
loop variable. Note that val.op is an Exit, and this method must be
called in the control flow context where gradients() is called.
Args:
val: The output tensor of an Exit op.
Returns:
A zero tensor of the same shape of val.
"""
val_shape = val.get_shape()
forward_ctxt = val.op._get_control_flow_context()
outer_forward_ctxt = forward_ctxt.outer_context
if outer_forward_ctxt:
outer_forward_ctxt = outer_forward_ctxt.GetWhileContext()
outer_grad_state = None
if outer_forward_ctxt:
outer_grad_state = self._map.get(outer_forward_ctxt)
if outer_grad_state:
# This is a nested loop.
if val_shape.is_fully_defined():
# If the shape is known statically, just create a zero tensor
# with the right shape in the right context.
outer_grad_state.grad_context.Enter()
result = array_ops.zeros(val_shape.dims, val.dtype)
outer_grad_state.grad_context.Exit()
else:
# Only the shape of value is needed for backprop.
forward_ctxt.outer_context.Enter()
shape = array_ops.shape_internal(val, optimize=False)
forward_ctxt.outer_context.Exit()
# Save the shape to a stack.
history_shape = outer_grad_state.AddForwardAccumulator(shape)
# Get the shape back from the stack.
outer_grad_ctxt = outer_grad_state.grad_context
outer_grad_ctxt.Enter()
real_shape = outer_grad_state.AddBackpropAccumulatedValue(
history_shape, shape)
result = array_ops.zeros(real_shape, val.dtype)
outer_grad_ctxt.Exit()
else:
# This is not a nested loop.
if val_shape.is_fully_defined():
# If the shape is known statically, just create a zero tensor
# with the right shape.
result = array_ops.zeros(val_shape.dims, val.dtype)
else:
result = array_ops.zeros_like(val, optimize=False)
return result
def ZerosLike(self, op, index):
"""Create zeros_like for the specified output of an op.
If op is in a while loop that is part of gradients(), this method
must be called in its grad loop context.
Args:
op: A tensorflow operation.
index: the index for a specific output of the op.
Returns:
A zero tensor of the same shape of op.outputs[index].
"""
if util.IsLoopSwitch(op):
return None
if op.graph._building_function: # pylint: disable=protected-access
# The optimization here is tricky to apply to functions
return array_ops.zeros_like(op.outputs[index])
dead_branch = util.IsSwitch(op)
forward_ctxt = util.GetWhileContext(op)
grad_state = self._map.get(forward_ctxt)
if grad_state is None:
# op is not in a while loop that is part of gradients().
return ZerosLikeOutsideLoop(op, index)
op_ctxt = op._get_control_flow_context()
val = ops.convert_to_tensor(op.outputs[index], name="tensor")
shape = val.get_shape()
if shape.is_fully_defined():
# If the shape is known statically, just create a zero tensor with
# the right shape in the grad loop context.
result = constant_op.constant(0, shape=shape.dims, dtype=val.dtype)
if dead_branch:
# op is a cond switch. Guard the zero tensor with a switch.
pred = grad_state.history_map.get(op_ctxt.pred.name)
branch = op_ctxt.branch
result = control_flow_ops._SwitchRefOrTensor(result, pred)[1 - branch]
else:
# Unknown shape so keep a history of the shape at runtime.
if dead_branch:
# Need to add a special switch to guard the value.
pred = op_ctxt.pred
branch = op_ctxt.branch
op_ctxt.outer_context.Enter()
val = control_flow_ops._SwitchRefOrTensor(op.inputs[0],
pred)[1 - branch]
zeros_shape = array_ops.shape_internal(val, optimize=False)
op_ctxt.outer_context.Exit()
val.op._set_control_flow_context(op_ctxt)
zeros_shape.op._set_control_flow_context(op_ctxt)
else:
op_ctxt.Enter()
zeros_shape = array_ops.shape_internal(val, optimize=False)
op_ctxt.Exit()
# Add forward accumulator for shape.
grad_state.grad_context.Exit()
history_zeros_shape = grad_state.AddForwardAccumulator(
zeros_shape, dead_branch=dead_branch)
grad_state.grad_context.Enter()
# Create a zero tensor with the right shape.
shape = grad_state.AddBackpropAccumulatedValue(history_zeros_shape,
zeros_shape, dead_branch)
result = array_ops.zeros(shape, val.dtype)
return result
def PostProcessing(self):
"""Perform postprocessing at the end of gradients().
We have created the gradient graph at this point. So this function
can be used to perform any postprocessing on the gradient graph.
We currently perform the following postprocessing:
1. Patch the gradient graph if the output of a loop variable
doesn't depend on its input.
"""
for _, grad_state in self._map.items():
for _, b_merge in grad_state.switch_map.items():
if b_merge.op.inputs[0] == b_merge.op.inputs[1]:
# The value of this loop variable at iteration i+1 doesn't
# depend on its value at iteration i. So use zeros as the
# gradients for all iterations > 0.
dtype = b_merge.op.inputs[0].dtype
shape = b_merge.op.inputs[0].get_shape()
# pylint: disable=protected-access
if shape.is_fully_defined():
grad_state.grad_context.Enter()
# Create a zeros and use it for iterations > 0.
grad_val = constant_op.constant(0, dtype=dtype, shape=shape)
next_grad_val = control_flow_ops._NextIteration(grad_val)
grad_state.grad_context.Exit()
else:
# Create a zeros in the outer grad context.
outer_grad_ctxt = grad_state.grad_context.outer_context
if outer_grad_ctxt:
outer_grad_ctxt.Enter()
enter_grad_op = b_merge.op.inputs[0].op
enter_grad = enter_grad_op.inputs[0]
grad_shape = array_ops.shape_internal(enter_grad, optimize=False)
grad_val = array_ops.zeros(grad_shape)
if outer_grad_ctxt:
outer_grad_ctxt.Exit()
# Use the zeros for iterations > 0.
grad_state.grad_context.Enter()
next_grad_val = control_flow_ops._NextIteration(grad_val)
grad_state.grad_context.Exit()
b_merge.op._update_input(1, next_grad_val)
# pylint: enable=protected-access
def MaybeCreateControlFlowState(between_op_list, between_ops,
colocate_gradients_with_ops):
"""Create the state for all the while loops involved in one gradients().
We create a _ControlFlowState when there are while loops involved in
gradients(). In gradients(), control flow logic is only invoked when
the _ControlFlowState is not None.
Note that this method modifies `between_op_list` and `between_ops`.
"""
loop_state = None
for op in between_op_list:
if util.IsLoopExit(op):
if loop_state is None:
loop_state = _ControlFlowState()
if colocate_gradients_with_ops:
with ops.colocate_with(op):
loop_state.AddWhileContext(op, between_op_list, between_ops)
else:
loop_state.AddWhileContext(op, between_op_list, between_ops)
return loop_state
def ZerosLikeOutsideLoop(op, index):
"""Create zeros_like for the specified output of an op."""
val = op.outputs[index]
if not util.IsSwitch(op):
if val.dtype == dtypes.resource:
return array_ops.zeros(gen_resource_variable_ops.variable_shape(val))
return array_ops.zeros_like(val, optimize=False)
else:
op_ctxt = op._get_control_flow_context()
if op_ctxt:
# We are in a cond context. Use a switch to create zeros only when needed.
pred = op_ctxt.pred
branch = op_ctxt.branch
switch_val = control_flow_ops.switch(op.inputs[0], pred)[1 - branch]
# A op is created along the branch taken as control dependencies are on
# the whole op and not on the tensor output.
pivot = array_ops.identity(switch_val)
if val.dtype == dtypes.resource:
with ops.control_dependencies([pivot]):
return array_ops.zeros(
gen_resource_variable_ops.variable_shape(switch_val),
dtype=default_gradient.get_zeros_dtype(val))
zeros_shape = array_ops.shape_internal(switch_val, optimize=False)
# Ensure ops created within array_ops.zeros are dominated by switch in
# cond context.
with ops.control_dependencies([pivot]):
return array_ops.zeros(zeros_shape, dtype=val.dtype)
else:
return array_ops.zeros_like(val, optimize=False)
|
tensorflow-master
|
tensorflow/python/ops/control_flow_state.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Gradients for operators defined in control_flow_ops.py."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from six.moves import xrange # pylint: disable=redefined-builtin
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import sparse_tensor
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import control_flow_util
from tensorflow.python.ops import math_ops
# go/tf-wildcard-import
# pylint: disable=wildcard-import,undefined-variable
from tensorflow.python.ops.control_flow_ops import *
# pylint: enable=wildcard-import
def _SwitchGrad(op, *grad):
"""Gradients for a Switch op is calculated using a Merge op.
If the switch is a loop switch, it will be visited twice. We create
the merge on the first visit, and update the other input of the merge
on the second visit. A next_iteration is also added on second visit.
"""
graph = ops.get_default_graph()
# pylint: disable=protected-access
op_ctxt = op._get_control_flow_context()
grad_ctxt = graph._get_control_flow_context()
# pylint: enable=protected-access
if isinstance(op_ctxt, WhileContext):
merge_grad = grad_ctxt.grad_state.switch_map.get(op)
if merge_grad is not None:
# This is the second time this Switch is visited. It comes from
# the non-exit branch of the Switch, so update the second input
# to the Merge.
# TODO(yuanbyu): Perform shape inference with this new input.
if grad[1] is not None:
# pylint: disable=protected-access
control_flow_ops._AddNextAndBackEdge(merge_grad, grad[1],
enforce_shape_invariant=False)
# pylint: enable=protected-access
return None, None
elif grad[0] is not None:
# This is the first time this Switch is visited. It comes from
# the Exit branch, which is grad[0]. grad[1] is empty at this point.
# Use grad[0] for both inputs to merge for now, but update the second
# input of merge when we see this Switch the second time.
merge_grad = merge([grad[0], grad[0]], name="b_switch")[0]
grad_ctxt.grad_state.switch_map[op] = merge_grad
return merge_grad, None
else:
# This is the first time this Switch is visited. It comes from the
# Identity branch. Such a Switch has `None` gradient for the Exit branch,
# meaning the output is not differentiable.
return None, None
elif isinstance(op_ctxt, CondContext):
zero_grad = grad[1 - op_ctxt.branch]
# At this point, we have created zero_grad guarded by the right switch.
# Unfortunately, we may still get None here for not trainable data types.
if zero_grad is None:
# For resource variables we get None always on the other branch, so bypass
# this.
if op.inputs[0].dtype == dtypes.resource:
return merge(
[grad[op_ctxt.branch]] * 2, name="cond_resource_grad")[0], None
return None, None
return merge(grad, name="cond_grad")[0], None
else:
false_grad = switch(grad[0], op.inputs[1])[0]
true_grad = switch(grad[1], op.inputs[1])[1]
return merge([false_grad, true_grad])[0], None
ops.RegisterGradient("Switch")(_SwitchGrad)
ops.RegisterGradient("RefSwitch")(_SwitchGrad)
@ops.RegisterGradient("Merge")
def _MergeGrad(op, grad, _):
"""Gradients for a Merge op are calculated using a Switch op."""
input_op = op.inputs[0].op
graph = ops.get_default_graph()
# pylint: disable=protected-access
op_ctxt = control_flow_util.GetOutputContext(input_op)
grad_ctxt = graph._get_control_flow_context()
# pylint: enable=protected-access
if isinstance(op_ctxt, WhileContext):
# pylint: disable=protected-access
return control_flow_ops._SwitchRefOrTensor(grad, grad_ctxt.pivot)
# pylint: enable=protected-access
elif isinstance(op_ctxt, CondContext):
pred = op_ctxt.pred
if grad_ctxt and grad_ctxt.grad_state:
# This Merge node is part of a cond within a loop.
# The backprop needs to have the value of this predicate for every
# iteration. So we must have its values accumulated in the forward, and
# use the accumulated values as the predicate for this backprop switch.
grad_state = grad_ctxt.grad_state
real_pred = grad_state.history_map.get(pred.name)
if real_pred is None:
# Remember the value of pred for every iteration.
grad_ctxt = grad_state.grad_context
grad_ctxt.Exit()
history_pred = grad_state.AddForwardAccumulator(pred)
grad_ctxt.Enter()
# Add the stack pop op. If pred.op is in a (outer) CondContext,
# the stack pop will be guarded with a switch.
real_pred = grad_state.AddBackpropAccumulatedValue(history_pred, pred)
grad_state.history_map[pred.name] = real_pred
pred = real_pred
# pylint: disable=protected-access
return control_flow_ops._SwitchRefOrTensor(grad, pred, name="cond_grad")
# pylint: enable=protected-access
else:
num_inputs = len(op.inputs)
cond = [math_ops.equal(op.outputs[1], i) for i in xrange(num_inputs)]
# pylint: disable=protected-access
return [control_flow_ops._SwitchRefOrTensor(grad, cond[i])[1]
for i in xrange(num_inputs)]
# pylint: enable=protected-access
@ops.RegisterGradient("RefMerge")
def _RefMergeGrad(op, grad, _):
return _MergeGrad(op, grad, _)
@ops.RegisterGradient("Exit")
def _ExitGrad(op, grad):
"""Gradients for an exit op are calculated using an Enter op."""
graph = ops.get_default_graph()
# pylint: disable=protected-access
op_ctxt = op._get_control_flow_context()
grad_ctxt = graph._get_control_flow_context()
# pylint: enable=protected-access
if not grad_ctxt.back_prop:
# The flag `back_prop` is set by users to suppress gradient
# computation for this loop. If the attribute `back_prop` is false,
# no gradient computation.
return None
if op_ctxt.grad_state:
raise TypeError("Second-order gradient for while loops not supported.")
if isinstance(grad, ops.Tensor):
grad_ctxt.AddName(grad.name)
else:
if not isinstance(grad, (ops.IndexedSlices, sparse_tensor.SparseTensor)):
raise TypeError("Type %s not supported" % type(grad))
grad_ctxt.AddName(grad.values.name)
grad_ctxt.AddName(grad.indices.name)
dense_shape = grad.dense_shape
if dense_shape is not None:
grad_ctxt.AddName(dense_shape.name)
grad_ctxt.Enter()
# pylint: disable=protected-access
result = control_flow_ops._Enter(
grad, grad_ctxt.name, is_constant=False,
parallel_iterations=grad_ctxt.parallel_iterations,
name="b_exit")
# pylint: enable=protected-access
grad_ctxt.loop_enters.append(result)
grad_ctxt.Exit()
return result
ops.RegisterGradient("RefExit")(_ExitGrad)
@ops.RegisterGradient("NextIteration")
def _NextIterationGrad(_, grad):
"""A forward next_iteration is translated into a backprop identity.
Note that the backprop next_iteration is added in switch grad.
"""
return grad
@ops.RegisterGradient("RefNextIteration")
def _RefNextIterationGrad(_, grad):
return _NextIterationGrad(_, grad)
@ops.RegisterGradient("Enter")
def _EnterGrad(op, grad):
"""Gradients for an Enter are calculated using an Exit op.
For loop variables, grad is the gradient so just add an exit.
For loop invariants, we need to add an accumulator loop.
"""
graph = ops.get_default_graph()
# pylint: disable=protected-access
grad_ctxt = graph._get_control_flow_context()
# pylint: enable=protected-access
if not grad_ctxt.back_prop:
# Skip gradient computation, if the attribute `back_prop` is false.
return grad
if grad_ctxt.grad_state is None:
# Pass the gradient through if we are not in a gradient while context.
return grad
if op.get_attr("is_constant"):
# Add a gradient accumulator for each loop invariant.
if isinstance(grad, ops.Tensor):
result = grad_ctxt.AddBackpropAccumulator(op, grad)
elif isinstance(grad, ops.IndexedSlices):
result = grad_ctxt.AddBackpropIndexedSlicesAccumulator(op, grad)
else:
# TODO(yuanbyu, lukasr): Add support for SparseTensor.
raise TypeError("Type %s not supported" % type(grad))
else:
result = exit(grad)
grad_ctxt.loop_exits.append(result)
grad_ctxt.ExitResult([result])
return result
@ops.RegisterGradient("RefEnter")
def _RefEnterGrad(op, grad):
return _EnterGrad(op, grad)
@ops.RegisterGradient("LoopCond")
def _LoopCondGrad(_):
"""Stop backprop for the predicate of a while loop."""
return None
|
tensorflow-master
|
tensorflow/python/ops/control_flow_grad.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for Python ops defined in image_grad.py."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.eager import backprop
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import test_util
from tensorflow.python.ops import gradient_checker
from tensorflow.python.ops import gradients_impl
from tensorflow.python.ops import image_ops
from tensorflow.python.platform import test
@test_util.for_all_test_methods(test_util.disable_xla,
'align_corners=False not supported by XLA')
class ResizeNearestNeighborOpTest(test.TestCase):
TYPES = [np.float32, np.float64]
def testShapeIsCorrectAfterOp(self):
in_shape = [1, 2, 2, 1]
out_shape = [1, 4, 6, 1]
for nptype in self.TYPES:
x = np.arange(0, 4).reshape(in_shape).astype(nptype)
with self.cached_session(use_gpu=True):
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_nearest_neighbor(input_tensor,
out_shape[1:3])
self.assertEqual(out_shape, list(resize_out.get_shape()))
resize_out = self.evaluate(resize_out)
self.assertEqual(out_shape, list(resize_out.shape))
@test_util.run_deprecated_v1
def testGradFromResizeToLargerInBothDims(self):
in_shape = [1, 2, 3, 1]
out_shape = [1, 4, 6, 1]
for nptype in self.TYPES:
x = np.arange(0, 6).reshape(in_shape).astype(nptype)
with self.cached_session(use_gpu=True):
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_nearest_neighbor(input_tensor,
out_shape[1:3])
err = gradient_checker.compute_gradient_error(
input_tensor, in_shape, resize_out, out_shape, x_init_value=x)
self.assertLess(err, 1e-3)
@test_util.run_deprecated_v1
def testGradFromResizeToSmallerInBothDims(self):
in_shape = [1, 4, 6, 1]
out_shape = [1, 2, 3, 1]
for nptype in self.TYPES:
x = np.arange(0, 24).reshape(in_shape).astype(nptype)
with self.cached_session(use_gpu=True):
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_nearest_neighbor(input_tensor,
out_shape[1:3])
err = gradient_checker.compute_gradient_error(
input_tensor, in_shape, resize_out, out_shape, x_init_value=x)
self.assertLess(err, 1e-3)
@test_util.run_deprecated_v1
def testCompareGpuVsCpu(self):
in_shape = [1, 4, 6, 3]
out_shape = [1, 8, 16, 3]
for nptype in self.TYPES:
x = np.arange(0, np.prod(in_shape)).reshape(in_shape).astype(nptype)
for align_corners in [True, False]:
with self.cached_session(use_gpu=False):
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_nearest_neighbor(
input_tensor, out_shape[1:3], align_corners=align_corners)
grad_cpu = gradient_checker.compute_gradient(
input_tensor, in_shape, resize_out, out_shape, x_init_value=x)
with self.cached_session(use_gpu=True):
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_nearest_neighbor(
input_tensor, out_shape[1:3], align_corners=align_corners)
grad_gpu = gradient_checker.compute_gradient(
input_tensor, in_shape, resize_out, out_shape, x_init_value=x)
self.assertAllClose(grad_cpu, grad_gpu, rtol=1e-5, atol=1e-5)
class ResizeBilinearOpTest(test.TestCase):
def testShapeIsCorrectAfterOp(self):
in_shape = [1, 2, 2, 1]
out_shape = [1, 4, 6, 1]
x = np.arange(0, 4).reshape(in_shape).astype(np.float32)
with self.cached_session():
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_bilinear(input_tensor, out_shape[1:3])
self.assertEqual(out_shape, list(resize_out.get_shape()))
resize_out = self.evaluate(resize_out)
self.assertEqual(out_shape, list(resize_out.shape))
@test_util.run_deprecated_v1
def testGradFromResizeToLargerInBothDims(self):
in_shape = [1, 2, 3, 1]
out_shape = [1, 4, 6, 1]
x = np.arange(0, 6).reshape(in_shape).astype(np.float32)
with self.cached_session():
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_bilinear(input_tensor, out_shape[1:3])
err = gradient_checker.compute_gradient_error(
input_tensor, in_shape, resize_out, out_shape, x_init_value=x)
self.assertLess(err, 1e-3)
@test_util.run_deprecated_v1
def testGradFromResizeToSmallerInBothDims(self):
in_shape = [1, 4, 6, 1]
out_shape = [1, 2, 3, 1]
x = np.arange(0, 24).reshape(in_shape).astype(np.float32)
with self.cached_session():
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_bilinear(input_tensor, out_shape[1:3])
err = gradient_checker.compute_gradient_error(
input_tensor, in_shape, resize_out, out_shape, x_init_value=x)
self.assertLess(err, 1e-3)
@test_util.run_deprecated_v1
def testCompareGpuVsCpu(self):
in_shape = [2, 4, 6, 3]
out_shape = [2, 8, 16, 3]
size = np.prod(in_shape)
x = 1.0 / size * np.arange(0, size).reshape(in_shape).astype(np.float32)
# Align corners will be deprecated for tf2.0 and the false version is not
# supported by XLA.
align_corner_options = [True
] if test_util.is_xla_enabled() else [True, False]
for align_corners in align_corner_options:
grad = {}
for use_gpu in [False, True]:
with self.cached_session(use_gpu=use_gpu):
input_tensor = constant_op.constant(x, shape=in_shape)
resized_tensor = image_ops.resize_bilinear(
input_tensor, out_shape[1:3], align_corners=align_corners)
grad[use_gpu] = gradient_checker.compute_gradient(
input_tensor, in_shape, resized_tensor, out_shape, x_init_value=x)
self.assertAllClose(grad[False], grad[True], rtol=1e-4, atol=1e-4)
@test_util.run_deprecated_v1
def testTypes(self):
in_shape = [1, 4, 6, 1]
out_shape = [1, 2, 3, 1]
x = np.arange(0, 24).reshape(in_shape)
with self.cached_session() as sess:
for dtype in [np.float16, np.float32, np.float64]:
input_tensor = constant_op.constant(x.astype(dtype), shape=in_shape)
resize_out = image_ops.resize_bilinear(input_tensor, out_shape[1:3])
grad = sess.run(gradients_impl.gradients(resize_out, input_tensor))[0]
self.assertAllEqual(in_shape, grad.shape)
# Not using gradient_checker.compute_gradient as I didn't work out
# the changes required to compensate for the lower precision of
# float16 when computing the numeric jacobian.
# Instead, we just test the theoretical jacobian.
self.assertAllEqual([[[[1.], [0.], [1.], [0.], [1.], [0.]], [[0.], [
0.
], [0.], [0.], [0.], [0.]], [[1.], [0.], [1.], [0.], [1.], [0.]],
[[0.], [0.], [0.], [0.], [0.], [0.]]]], grad)
class ResizeBicubicOpTest(test.TestCase):
def testShapeIsCorrectAfterOp(self):
in_shape = [1, 2, 2, 1]
out_shape = [1, 4, 6, 1]
x = np.arange(0, 4).reshape(in_shape).astype(np.float32)
for align_corners in [True, False]:
with self.cached_session():
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_bicubic(input_tensor, out_shape[1:3],
align_corners=align_corners)
self.assertEqual(out_shape, list(resize_out.get_shape()))
resize_out = self.evaluate(resize_out)
self.assertEqual(out_shape, list(resize_out.shape))
@test_util.run_deprecated_v1
def testGradFromResizeToLargerInBothDims(self):
in_shape = [1, 2, 3, 1]
out_shape = [1, 4, 6, 1]
x = np.arange(0, 6).reshape(in_shape).astype(np.float32)
for align_corners in [True, False]:
with self.cached_session():
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_bicubic(input_tensor, out_shape[1:3],
align_corners=align_corners)
err = gradient_checker.compute_gradient_error(
input_tensor, in_shape, resize_out, out_shape, x_init_value=x)
self.assertLess(err, 1e-3)
@test_util.run_deprecated_v1
def testGradFromResizeToSmallerInBothDims(self):
in_shape = [1, 4, 6, 1]
out_shape = [1, 2, 3, 1]
x = np.arange(0, 24).reshape(in_shape).astype(np.float32)
for align_corners in [True, False]:
with self.cached_session():
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_bicubic(input_tensor, out_shape[1:3],
align_corners=align_corners)
err = gradient_checker.compute_gradient_error(
input_tensor, in_shape, resize_out, out_shape, x_init_value=x)
self.assertLess(err, 1e-3)
@test_util.run_deprecated_v1
def testGradOnUnsupportedType(self):
in_shape = [1, 4, 6, 1]
out_shape = [1, 2, 3, 1]
x = np.arange(0, 24).reshape(in_shape).astype(np.uint8)
with self.cached_session():
input_tensor = constant_op.constant(x, shape=in_shape)
resize_out = image_ops.resize_bicubic(input_tensor, out_shape[1:3])
grad = gradients_impl.gradients(input_tensor, [resize_out])
self.assertEqual([None], grad)
class ScaleAndTranslateOpTest(test.TestCase):
@test_util.run_deprecated_v1
def testGrads(self):
in_shape = [1, 2, 3, 1]
out_shape = [1, 4, 6, 1]
x = np.arange(0, 6).reshape(in_shape).astype(np.float32)
kernel_types = [
'lanczos1', 'lanczos3', 'lanczos5', 'gaussian', 'box', 'triangle',
'keyscubic', 'mitchellcubic'
]
scales = [(1.0, 1.0), (0.37, 0.47), (2.1, 2.1)]
translations = [(0.0, 0.0), (3.14, 1.19), (2.1, 3.1), (100.0, 200.0)]
for scale in scales:
for translation in translations:
for kernel_type in kernel_types:
for antialias in [True, False]:
with self.cached_session():
input_tensor = constant_op.constant(x, shape=in_shape)
scale_and_translate_out = image_ops.scale_and_translate(
input_tensor,
out_shape[1:3],
scale=constant_op.constant(scale),
translation=constant_op.constant(translation),
kernel_type=kernel_type,
antialias=antialias)
err = gradient_checker.compute_gradient_error(
input_tensor,
in_shape,
scale_and_translate_out,
out_shape,
x_init_value=x)
self.assertLess(err, 1e-3)
def testIdentityGrads(self):
"""Tests that Gradients for 1.0 scale should be ones for some kernels."""
in_shape = [1, 2, 3, 1]
out_shape = [1, 4, 6, 1]
x = np.arange(0, 6).reshape(in_shape).astype(np.float32)
kernel_types = ['lanczos1', 'lanczos3', 'lanczos5', 'triangle', 'keyscubic']
scale = (1.0, 1.0)
translation = (0.0, 0.0)
antialias = True
for kernel_type in kernel_types:
with self.cached_session():
input_tensor = constant_op.constant(x, shape=in_shape)
with backprop.GradientTape() as tape:
tape.watch(input_tensor)
scale_and_translate_out = image_ops.scale_and_translate(
input_tensor,
out_shape[1:3],
scale=constant_op.constant(scale),
translation=constant_op.constant(translation),
kernel_type=kernel_type,
antialias=antialias)
grad = tape.gradient(scale_and_translate_out, input_tensor)[0]
grad_v = self.evaluate(grad)
self.assertAllClose(np.ones_like(grad_v), grad_v)
class CropAndResizeOpTest(test.TestCase):
def testShapeIsCorrectAfterOp(self):
batch = 2
image_height = 3
image_width = 4
crop_height = 4
crop_width = 5
depth = 2
num_boxes = 2
image_shape = [batch, image_height, image_width, depth]
crop_size = [crop_height, crop_width]
crops_shape = [num_boxes, crop_height, crop_width, depth]
image = np.arange(0, batch * image_height * image_width *
depth).reshape(image_shape).astype(np.float32)
boxes = np.array([[0, 0, 1, 1], [.1, .2, .7, .8]], dtype=np.float32)
box_ind = np.array([0, 1], dtype=np.int32)
with self.session(use_gpu=True) as sess:
crops = image_ops.crop_and_resize(
constant_op.constant(
image, shape=image_shape),
constant_op.constant(
boxes, shape=[num_boxes, 4]),
constant_op.constant(
box_ind, shape=[num_boxes]),
constant_op.constant(
crop_size, shape=[2]))
self.assertEqual(crops_shape, list(crops.get_shape()))
crops = self.evaluate(crops)
self.assertEqual(crops_shape, list(crops.shape))
def _randomUniformAvoidAnchors(self, low, high, anchors, radius, num_samples):
"""Generate samples that are far enough from a set of anchor points.
We generate uniform samples in [low, high], then reject those that are less
than radius away from any point in anchors. We stop after we have accepted
num_samples samples.
Args:
low: The lower end of the interval.
high: The upper end of the interval.
anchors: A list of length num_crops with anchor points to avoid.
radius: Distance threshold for the samples from the anchors.
num_samples: How many samples to produce.
Returns:
samples: A list of length num_samples with the accepted samples.
"""
self.assertTrue(low < high)
self.assertTrue(radius >= 0)
num_anchors = len(anchors)
# Make sure that at least half of the interval is not forbidden.
self.assertTrue(2 * radius * num_anchors < 0.5 * (high - low))
anchors = np.reshape(anchors, num_anchors)
samples = []
while len(samples) < num_samples:
sample = np.random.uniform(low, high)
if np.all(np.fabs(sample - anchors) > radius):
samples.append(sample)
return samples
@test_util.run_deprecated_v1
def testGradRandomBoxes(self):
"""Test that the gradient is correct for randomly generated boxes.
The mapping is piecewise differentiable with respect to the box coordinates.
The points where the function is not differentiable are those which are
mapped to image pixels, i.e., the normalized y coordinates in
np.linspace(0, 1, image_height) and normalized x coordinates in
np.linspace(0, 1, image_width). Make sure that the box coordinates are
sufficiently far away from those rectangular grid centers that are points of
discontinuity, so that the finite difference Jacobian is close to the
computed one.
"""
np.random.seed(1) # Make it reproducible.
delta = 1e-3
radius = 2 * delta
low, high = -0.5, 1.5 # Also covers the case of extrapolation.
image_height = 4
for image_width in range(1, 3):
for crop_height in range(1, 3):
for crop_width in range(2, 4):
for depth in range(1, 3):
for num_boxes in range(1, 3):
batch = num_boxes
image_shape = [batch, image_height, image_width, depth]
crop_size = [crop_height, crop_width]
crops_shape = [num_boxes, crop_height, crop_width, depth]
boxes_shape = [num_boxes, 4]
image = np.arange(0, batch * image_height * image_width *
depth).reshape(image_shape).astype(np.float32)
boxes = []
for _ in range(num_boxes):
# pylint: disable=unbalanced-tuple-unpacking
y1, y2 = self._randomUniformAvoidAnchors(
low, high, np.linspace(0, 1, image_height), radius, 2)
x1, x2 = self._randomUniformAvoidAnchors(
low, high, np.linspace(0, 1, image_width), radius, 2)
# pylint: enable=unbalanced-tuple-unpacking
boxes.append([y1, x1, y2, x2])
boxes = np.array(boxes, dtype=np.float32)
box_ind = np.arange(batch, dtype=np.int32)
with self.cached_session(use_gpu=True):
image_tensor = constant_op.constant(image, shape=image_shape)
boxes_tensor = constant_op.constant(boxes, shape=[num_boxes, 4])
box_ind_tensor = constant_op.constant(
box_ind, shape=[num_boxes])
crops = image_ops.crop_and_resize(
image_tensor,
boxes_tensor,
box_ind_tensor,
constant_op.constant(
crop_size, shape=[2]))
err = gradient_checker.compute_gradient_error(
[image_tensor, boxes_tensor], [image_shape, boxes_shape],
crops,
crops_shape,
delta=delta,
x_init_value=[image, boxes])
self.assertLess(err, 2e-3)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/ops/image_grad_test.py
|
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Evaluation-related metrics."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
# go/tf-wildcard-import
# pylint: disable=wildcard-import
from tensorflow.python.ops.metrics_impl import *
# pylint: enable=wildcard-import
|
tensorflow-master
|
tensorflow/python/ops/metrics.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
# Tests for this file live in python/kernel_tests/array_ops_test.py
"""Support for manipulating tensors."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import sys
import numpy as np
import six
from tensorflow.python.compat import compat
from tensorflow.python.eager import context
from tensorflow.python.framework import common_shapes
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import sparse_tensor
from tensorflow.python.framework import tensor_shape
from tensorflow.python.framework import tensor_util
# 'Constant' gets imported in the module 'array_ops'.
from tensorflow.python.framework.constant_op import constant
from tensorflow.python.ops import gen_array_ops
from tensorflow.python.ops import gen_math_ops
# go/tf-wildcard-import
# pylint: disable=wildcard-import
from tensorflow.python.ops.gen_array_ops import *
from tensorflow.python.ops.gen_array_ops import reverse_v2 as reverse # pylint: disable=unused-import
from tensorflow.python.util import deprecation
from tensorflow.python.util import dispatch
from tensorflow.python.util import nest
from tensorflow.python.util.tf_export import tf_export
# pylint: enable=wildcard-import
# Used for slicing to specify a new 1 size dimension
newaxis = None
tf_export("newaxis").export_constant(__name__, "newaxis")
# We override the 'slice' for the "slice" op, so we keep python's
# existing 'slice' for later use in this module.
_BaseSlice = slice
@tf_export("identity")
@dispatch.add_dispatch_support
def identity(input, name=None): # pylint: disable=redefined-builtin
r"""Return a tensor with the same shape and contents as input.
Args:
input: A `Tensor`.
name: A name for the operation (optional).
Returns:
A `Tensor`. Has the same type as `input`.
"""
if context.executing_eagerly() and not hasattr(input, "graph"):
input = ops.convert_to_tensor(input)
in_device = input.backing_device
# TODO(ashankar): Does 'identity' need to invoke execution callbacks?
context_device = context.context().device_name
if not context_device:
context_device = "/job:localhost/replica:0/task:0/device:CPU:0"
if context_device == in_device:
return input
else:
copied = input._copy() # pylint: disable=protected-access
if hasattr(copied, "_handle_data"):
copied._handle_data = input._handle_data # pylint: disable=protected-access
return copied
else:
ret = gen_array_ops.identity(input, name=name)
# Propagate handle data for happier shape inference for resource variables.
if hasattr(input, "_handle_data"):
ret._handle_data = input._handle_data # pylint: disable=protected-access
return ret
# pylint: disable=redefined-builtin,protected-access
@tf_export(v1=["expand_dims"])
@dispatch.add_dispatch_support
@deprecation.deprecated_args(None, "Use the `axis` argument instead", "dim")
def expand_dims(input, axis=None, name=None, dim=None):
"""Inserts a dimension of 1 into a tensor's shape.
Given a tensor `input`, this operation inserts a dimension of 1 at the
dimension index `axis` of `input`'s shape. The dimension index `axis` starts
at zero; if you specify a negative number for `axis` it is counted backward
from the end.
This operation is useful if you want to add a batch dimension to a single
element. For example, if you have a single image of shape `[height, width,
channels]`, you can make it a batch of 1 image with `expand_dims(image, 0)`,
which will make the shape `[1, height, width, channels]`.
Other examples:
```python
# 't' is a tensor of shape [2]
tf.shape(tf.expand_dims(t, 0)) # [1, 2]
tf.shape(tf.expand_dims(t, 1)) # [2, 1]
tf.shape(tf.expand_dims(t, -1)) # [2, 1]
# 't2' is a tensor of shape [2, 3, 5]
tf.shape(tf.expand_dims(t2, 0)) # [1, 2, 3, 5]
tf.shape(tf.expand_dims(t2, 2)) # [2, 3, 1, 5]
tf.shape(tf.expand_dims(t2, 3)) # [2, 3, 5, 1]
```
This operation requires that:
`-1-input.dims() <= dim <= input.dims()`
This operation is related to `squeeze()`, which removes dimensions of
size 1.
Args:
input: A `Tensor`.
axis: 0-D (scalar). Specifies the dimension index at which to expand the
shape of `input`. Must be in the range `[-rank(input) - 1, rank(input)]`.
name: The name of the output `Tensor` (optional).
dim: 0-D (scalar). Equivalent to `axis`, to be deprecated.
Returns:
A `Tensor` with the same data as `input`, but its shape has an additional
dimension of size 1 added.
Raises:
ValueError: if either both or neither of `dim` and `axis` are specified.
"""
axis = deprecation.deprecated_argument_lookup("axis", axis, "dim", dim)
if axis is None:
raise ValueError("Must specify an axis argument to tf.expand_dims()")
return expand_dims_v2(input, axis, name)
@tf_export("expand_dims", v1=[])
@dispatch.add_dispatch_support
def expand_dims_v2(input, axis, name=None):
"""Inserts a dimension of 1 into a tensor's shape.
Given a tensor `input`, this operation inserts a dimension of 1 at the
dimension index `axis` of `input`'s shape. The dimension index `axis` starts
at zero; if you specify a negative number for `axis` it is counted backward
from the end.
This operation is useful if you want to add a batch dimension to a single
element. For example, if you have a single image of shape `[height, width,
channels]`, you can make it a batch of 1 image with `expand_dims(image, 0)`,
which will make the shape `[1, height, width, channels]`.
Other examples:
```python
# 't' is a tensor of shape [2]
tf.shape(tf.expand_dims(t, 0)) # [1, 2]
tf.shape(tf.expand_dims(t, 1)) # [2, 1]
tf.shape(tf.expand_dims(t, -1)) # [2, 1]
# 't2' is a tensor of shape [2, 3, 5]
tf.shape(tf.expand_dims(t2, 0)) # [1, 2, 3, 5]
tf.shape(tf.expand_dims(t2, 2)) # [2, 3, 1, 5]
tf.shape(tf.expand_dims(t2, 3)) # [2, 3, 5, 1]
```
This operation requires that:
`-1-input.dims() <= dim <= input.dims()`
This operation is related to `squeeze()`, which removes dimensions of
size 1.
Args:
input: A `Tensor`.
axis: 0-D (scalar). Specifies the dimension index at which to expand the
shape of `input`. Must be in the range `[-rank(input) - 1, rank(input)]`.
name: The name of the output `Tensor` (optional).
Returns:
A `Tensor` with the same data as `input`, but its shape has an additional
dimension of size 1 added.
"""
return gen_array_ops.expand_dims(input, axis, name)
# pylint: enable=redefined-builtin,protected-access
# Aliases for some automatically-generated names.
# pylint: disable=protected-access
@deprecation.deprecated("2016-11-30",
"This op will be removed after the deprecation date. "
"Please switch to tf.setdiff1d().")
def listdiff(x, y, out_idx=None, name=None):
return gen_array_ops.list_diff(x, y, out_idx, name)
listdiff.__doc__ = gen_array_ops.list_diff.__doc__ + "\n" + listdiff.__doc__
# pylint: enable=protected-access
# pylint: disable=undefined-variable
@deprecation.deprecated("2018-11-30",
"This op will be removed after the deprecation date. "
"Please switch to tf.sets.difference().")
@tf_export(v1=["setdiff1d"])
def setdiff1d(x, y, index_dtype=dtypes.int32, name=None):
return gen_array_ops.list_diff(x, y, index_dtype, name)
setdiff1d.__doc__ = gen_array_ops.list_diff.__doc__
@tf_export("broadcast_dynamic_shape")
def broadcast_dynamic_shape(shape_x, shape_y):
"""Computes the shape of a broadcast given symbolic shapes.
When shape_x and shape_y are Tensors representing shapes (i.e. the result of
calling tf.shape on another Tensor) this computes a Tensor which is the shape
of the result of a broadcasting op applied in tensors of shapes shape_x and
shape_y.
For example, if shape_x is [1, 2, 3] and shape_y is [5, 1, 3], the result is a
Tensor whose value is [5, 2, 3].
This is useful when validating the result of a broadcasting operation when the
tensors do not have statically known shapes.
Args:
shape_x: A rank 1 integer `Tensor`, representing the shape of x.
shape_y: A rank 1 integer `Tensor`, representing the shape of y.
Returns:
A rank 1 integer `Tensor` representing the broadcasted shape.
"""
return gen_array_ops.broadcast_args(shape_x, shape_y)
@tf_export("broadcast_static_shape")
def broadcast_static_shape(shape_x, shape_y):
"""Computes the shape of a broadcast given known shapes.
When shape_x and shape_y are fully known TensorShapes this computes a
TensorShape which is the shape of the result of a broadcasting op applied in
tensors of shapes shape_x and shape_y.
For example, if shape_x is [1, 2, 3] and shape_y is [5, 1, 3], the result is a
TensorShape whose value is [5, 2, 3].
This is useful when validating the result of a broadcasting operation when the
tensors have statically known shapes.
Args:
shape_x: A `TensorShape`
shape_y: A `TensorShape`
Returns:
A `TensorShape` representing the broadcasted shape.
Raises:
ValueError: If the two shapes can not be broadcasted.
"""
return common_shapes.broadcast_shape(shape_x, shape_y)
@tf_export("shape", v1=[])
def shape_v2(input, out_type=dtypes.int32, name=None):
# pylint: disable=redefined-builtin
"""Returns the shape of a tensor.
This operation returns a 1-D integer tensor representing the shape of `input`.
For example:
```python
t = tf.constant([[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]])
tf.shape(t) # [2, 2, 3]
```
Args:
input: A `Tensor` or `SparseTensor`.
out_type: (Optional) The specified output type of the operation (`int32` or
`int64`). Defaults to `tf.int32`.
name: A name for the operation (optional).
Returns:
A `Tensor` of type `out_type`.
"""
return shape(input, name, out_type)
@tf_export(v1=["shape"])
def shape(input, name=None, out_type=dtypes.int32):
# pylint: disable=redefined-builtin
"""Returns the shape of a tensor.
This operation returns a 1-D integer tensor representing the shape of `input`.
For example:
```python
t = tf.constant([[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]])
tf.shape(t) # [2, 2, 3]
```
Args:
input: A `Tensor` or `SparseTensor`.
name: A name for the operation (optional).
out_type: (Optional) The specified output type of the operation (`int32` or
`int64`). Defaults to `tf.int32`.
Returns:
A `Tensor` of type `out_type`.
"""
return shape_internal(input, name, optimize=True, out_type=out_type)
def shape_internal(input, name=None, optimize=True, out_type=dtypes.int32):
# pylint: disable=redefined-builtin
"""Returns the shape of a tensor.
Args:
input: A `Tensor` or `SparseTensor`.
name: A name for the operation (optional).
optimize: if true, encode the shape as a constant when possible.
out_type: (Optional) The specified output type of the operation (`int32` or
`int64`). Defaults to tf.int32.
Returns:
A `Tensor` of type `out_type`.
"""
with ops.name_scope(name, "Shape", [input]) as name:
if isinstance(
input, (sparse_tensor.SparseTensor, sparse_tensor.SparseTensorValue)):
return gen_math_ops.cast(input.dense_shape, out_type)
else:
if not context.executing_eagerly():
input_tensor = ops.convert_to_tensor(input)
input_shape = input_tensor.get_shape()
if optimize and input_shape.is_fully_defined():
return constant(input_shape.as_list(), out_type, name=name)
return gen_array_ops.shape(input, name=name, out_type=out_type)
@tf_export("shape_n")
def shape_n(input, out_type=dtypes.int32, name=None):
# pylint: disable=redefined-builtin
"""Returns shape of tensors.
Args:
input: A list of at least 1 `Tensor` object with the same type.
out_type: The specified output type of the operation (`int32` or `int64`).
Defaults to `tf.int32`(optional).
name: A name for the operation (optional).
Returns:
A list with the same length as `input` of `Tensor` objects with
type `out_type`.
"""
return gen_array_ops.shape_n(input, out_type=out_type, name=name)
@tf_export("size", v1=[])
@dispatch.add_dispatch_support
def size_v2(input, out_type=dtypes.int32, name=None):
# pylint: disable=redefined-builtin
return size(input, name, out_type)
@tf_export(v1=["size"])
@dispatch.add_dispatch_support
def size(input, name=None, out_type=dtypes.int32):
# pylint: disable=redefined-builtin
"""Returns the size of a tensor.
Returns a 0-D `Tensor` representing the number of elements in `input`
of type `out_type`. Defaults to tf.int32.
For example:
```python
t = tf.constant([[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]])
tf.size(t) # 12
```
Args:
input: A `Tensor` or `SparseTensor`.
name: A name for the operation (optional).
out_type: (Optional) The specified non-quantized numeric output type of the
operation. Defaults to `tf.int32`.
Returns:
A `Tensor` of type `out_type`. Defaults to `tf.int32`.
@compatibility(numpy)
Equivalent to np.size()
@end_compatibility
"""
return size_internal(input, name, optimize=True, out_type=out_type)
def size_internal(input, name=None, optimize=True, out_type=dtypes.int32):
# pylint: disable=redefined-builtin,protected-access
"""Returns the size of a tensor.
Args:
input: A `Tensor` or `SparseTensor`.
name: A name for the operation (optional).
optimize: if true, encode the size as a constant when possible.
out_type: (Optional) The specified non-quantized numeric output type of the
operation. Defaults to `tf.int32`.
Returns:
A `Tensor` of type `out_type`. Defaults to `tf.int32`.
"""
if (context.executing_eagerly()
and not hasattr(input, "graph")
and not isinstance(
input, (sparse_tensor.SparseTensor, sparse_tensor.SparseTensorValue))
):
input = ops.convert_to_tensor(input)
np_out_type = out_type.as_numpy_dtype
num_elements = np.prod(input._shape_tuple(), dtype=np_out_type) # pylint: disable=protected-access
return ops.convert_to_tensor(num_elements, dtype=out_type)
with ops.name_scope(name, "Size", [input]) as name:
if isinstance(
input, (sparse_tensor.SparseTensor, sparse_tensor.SparseTensorValue)):
return gen_math_ops.prod(
gen_math_ops.cast(input.dense_shape, out_type), 0, name=name)
else:
input_tensor = ops.convert_to_tensor(input)
input_shape = input_tensor.get_shape()
if optimize:
if input_shape.is_fully_defined():
return constant(input_shape.num_elements(), out_type, name=name)
if input_shape.dims and any(dim == 0 for dim in input_shape.dims):
return constant(0, out_type, name=name)
return gen_array_ops.size(input, name=name, out_type=out_type)
@tf_export("rank")
@dispatch.add_dispatch_support
def rank(input, name=None):
# pylint: disable=redefined-builtin
"""Returns the rank of a tensor.
Returns a 0-D `int32` `Tensor` representing the rank of `input`.
For example:
```python
# shape of tensor 't' is [2, 2, 3]
t = tf.constant([[[1, 1, 1], [2, 2, 2]], [[3, 3, 3], [4, 4, 4]]])
tf.rank(t) # 3
```
**Note**: The rank of a tensor is not the same as the rank of a matrix. The
rank of a tensor is the number of indices required to uniquely select each
element of the tensor. Rank is also known as "order", "degree", or "ndims."
Args:
input: A `Tensor` or `SparseTensor`.
name: A name for the operation (optional).
Returns:
A `Tensor` of type `int32`.
@compatibility(numpy)
Equivalent to np.ndim
@end_compatibility
"""
return rank_internal(input, name, optimize=True)
def rank_internal(input, name=None, optimize=True):
# pylint: disable=redefined-builtin
"""Returns the rank of a tensor.
Args:
input: A `Tensor` or `SparseTensor`.
name: A name for the operation (optional).
optimize: if true, encode the rank as a constant when possible.
Returns:
A `Tensor` of type `int32`.
"""
with ops.name_scope(name, "Rank", [input]) as name:
if isinstance(
input, (sparse_tensor.SparseTensor, sparse_tensor.SparseTensorValue)):
return gen_array_ops.size(input.dense_shape, name=name)
else:
input_tensor = ops.convert_to_tensor(input)
input_shape = input_tensor.get_shape()
if optimize and input_shape.ndims is not None:
return constant(input_shape.ndims, dtypes.int32, name=name)
return gen_array_ops.rank(input, name=name)
_SLICE_TYPE_ERROR = (
"Only integers, slices (`:`), ellipsis (`...`), "
"tf.newaxis (`None`) and scalar tf.int32/tf.int64 tensors are valid "
"indices")
_SUPPORTED_SLICE_DTYPES = (dtypes.int32, dtypes.int32_ref, dtypes.int64,
dtypes.int64_ref)
def _check_index(idx):
"""Check if a given value is a valid index into a tensor."""
if isinstance(idx, (six.integer_types, tensor_shape.Dimension)):
return
# Optimistic check. Assumptions:
# * any object with a dtype is supported
# * any object with a dtype has a sizeable shape attribute.
dtype = getattr(idx, "dtype", None)
if (dtype is None or dtypes.as_dtype(dtype) not in _SUPPORTED_SLICE_DTYPES or
idx.shape and len(idx.shape) == 1):
# TODO(slebedev): IndexError seems more appropriate here, but it
# will break `_slice_helper` contract.
raise TypeError(_SLICE_TYPE_ERROR + ", got {!r}".format(idx))
def _slice_helper(tensor, slice_spec, var=None):
"""Overload for Tensor.__getitem__.
This operation extracts the specified region from the tensor.
The notation is similar to NumPy with the restriction that
currently only support basic indexing. That means that
using a non-scalar tensor as input is not currently allowed.
Some useful examples:
```python
# Strip leading and trailing 2 elements
foo = tf.constant([1,2,3,4,5,6])
print(foo[2:-2].eval()) # => [3,4]
# Skip every other row and reverse the order of the columns
foo = tf.constant([[1,2,3], [4,5,6], [7,8,9]])
print(foo[::2,::-1].eval()) # => [[3,2,1], [9,8,7]]
# Use scalar tensors as indices on both dimensions
print(foo[tf.constant(0), tf.constant(2)].eval()) # => 3
# Insert another dimension
foo = tf.constant([[1,2,3], [4,5,6], [7,8,9]])
print(foo[tf.newaxis, :, :].eval()) # => [[[1,2,3], [4,5,6], [7,8,9]]]
print(foo[:, tf.newaxis, :].eval()) # => [[[1,2,3]], [[4,5,6]], [[7,8,9]]]
print(foo[:, :, tf.newaxis].eval()) # => [[[1],[2],[3]], [[4],[5],[6]],
[[7],[8],[9]]]
# Ellipses (3 equivalent operations)
foo = tf.constant([[1,2,3], [4,5,6], [7,8,9]])
print(foo[tf.newaxis, :, :].eval()) # => [[[1,2,3], [4,5,6], [7,8,9]]]
print(foo[tf.newaxis, ...].eval()) # => [[[1,2,3], [4,5,6], [7,8,9]]]
print(foo[tf.newaxis].eval()) # => [[[1,2,3], [4,5,6], [7,8,9]]]
# Masks
foo = tf.constant([[1,2,3], [4,5,6], [7,8,9]])
print(foo[foo > 2].eval()) # => [3, 4, 5, 6, 7, 8, 9]
```
Notes:
- `tf.newaxis` is `None` as in NumPy.
- An implicit ellipsis is placed at the end of the `slice_spec`
- NumPy advanced indexing is currently not supported.
Args:
tensor: An ops.Tensor object.
slice_spec: The arguments to Tensor.__getitem__.
var: In the case of variable slice assignment, the Variable object to slice
(i.e. tensor is the read-only view of this variable).
Returns:
The appropriate slice of "tensor", based on "slice_spec".
Raises:
ValueError: If a slice range is negative size.
TypeError: If the slice indices aren't int, slice, ellipsis,
tf.newaxis or scalar int32/int64 tensors.
"""
if isinstance(slice_spec, bool) or \
(isinstance(slice_spec, ops.Tensor) and slice_spec.dtype == dtypes.bool) or \
(isinstance(slice_spec, np.ndarray) and slice_spec.dtype == bool):
return boolean_mask(tensor=tensor, mask=slice_spec)
if not isinstance(slice_spec, (list, tuple)):
slice_spec = [slice_spec]
begin, end, strides = [], [], []
index = 0
new_axis_mask, shrink_axis_mask = 0, 0
begin_mask, end_mask = 0, 0
ellipsis_mask = 0
for s in slice_spec:
if isinstance(s, _BaseSlice):
# python doesn't always use None when constructing ranges
# for example a[:] gives slice(None,sys.maxsize,None)
# whereas a[::1] gives slice(None,None,None)
if s.start is not None and s.start is not sys.maxsize:
_check_index(s.start)
begin.append(s.start)
else:
begin.append(0)
begin_mask |= (1 << index)
if s.stop is not None and s.stop != sys.maxsize:
_check_index(s.stop)
end.append(s.stop)
else:
end.append(0)
end_mask |= (1 << index)
if s.step is not None:
_check_index(s.step)
strides.append(s.step)
else:
strides.append(1)
elif s is Ellipsis:
begin.append(0)
end.append(0)
strides.append(1)
ellipsis_mask |= (1 << index)
elif s is newaxis:
begin.append(0)
end.append(0)
strides.append(1)
new_axis_mask |= (1 << index)
else:
_check_index(s)
begin.append(s)
end.append(s + 1)
strides.append(1)
shrink_axis_mask |= (1 << index)
index += 1
# stack possibly involves no tensors, so we must use op_scope correct graph.
with ops.name_scope(None, "strided_slice",
[tensor] + begin + end + strides) as name:
if begin:
packed_begin, packed_end, packed_strides = (stack(begin), stack(end),
stack(strides))
if (packed_begin.dtype == dtypes.int64 or
packed_end.dtype == dtypes.int64 or
packed_strides.dtype == dtypes.int64):
if packed_begin.dtype != dtypes.int64:
packed_begin = gen_math_ops.cast(packed_begin, dtypes.int64)
if packed_end.dtype != dtypes.int64:
packed_end = gen_math_ops.cast(packed_end, dtypes.int64)
if packed_strides.dtype != dtypes.int64:
packed_strides = gen_math_ops.cast(packed_strides, dtypes.int64)
else:
var_empty = constant([], dtype=dtypes.int32)
packed_begin = packed_end = packed_strides = var_empty
return strided_slice(
tensor,
packed_begin,
packed_end,
packed_strides,
begin_mask=begin_mask,
end_mask=end_mask,
shrink_axis_mask=shrink_axis_mask,
new_axis_mask=new_axis_mask,
ellipsis_mask=ellipsis_mask,
var=var,
name=name)
# pylint: disable=undefined-variable,protected-access,redefined-outer-name
@tf_export("slice")
def slice(input_, begin, size, name=None):
# pylint: disable=redefined-builtin
"""Extracts a slice from a tensor.
This operation extracts a slice of size `size` from a tensor `input` starting
at the location specified by `begin`. The slice `size` is represented as a
tensor shape, where `size[i]` is the number of elements of the 'i'th dimension
of `input` that you want to slice. The starting location (`begin`) for the
slice is represented as an offset in each dimension of `input`. In other
words, `begin[i]` is the offset into the 'i'th dimension of `input` that you
want to slice from.
Note that `tf.Tensor.__getitem__` is typically a more pythonic way to
perform slices, as it allows you to write `foo[3:7, :-2]` instead of
`tf.slice(foo, [3, 0], [4, foo.get_shape()[1]-2])`.
`begin` is zero-based; `size` is one-based. If `size[i]` is -1,
all remaining elements in dimension i are included in the
slice. In other words, this is equivalent to setting:
`size[i] = input.dim_size(i) - begin[i]`
This operation requires that:
`0 <= begin[i] <= begin[i] + size[i] <= Di for i in [0, n]`
For example:
```python
t = tf.constant([[[1, 1, 1], [2, 2, 2]],
[[3, 3, 3], [4, 4, 4]],
[[5, 5, 5], [6, 6, 6]]])
tf.slice(t, [1, 0, 0], [1, 1, 3]) # [[[3, 3, 3]]]
tf.slice(t, [1, 0, 0], [1, 2, 3]) # [[[3, 3, 3],
# [4, 4, 4]]]
tf.slice(t, [1, 0, 0], [2, 1, 3]) # [[[3, 3, 3]],
# [[5, 5, 5]]]
```
Args:
input_: A `Tensor`.
begin: An `int32` or `int64` `Tensor`.
size: An `int32` or `int64` `Tensor`.
name: A name for the operation (optional).
Returns:
A `Tensor` the same type as `input`.
"""
return gen_array_ops._slice(input_, begin, size, name=name)
# pylint: disable=invalid-name
@tf_export("strided_slice")
def strided_slice(input_,
begin,
end,
strides=None,
begin_mask=0,
end_mask=0,
ellipsis_mask=0,
new_axis_mask=0,
shrink_axis_mask=0,
var=None,
name=None):
"""Extracts a strided slice of a tensor (generalized python array indexing).
**Instead of calling this op directly most users will want to use the
NumPy-style slicing syntax (e.g. `tensor[..., 3:4:-1, tf.newaxis, 3]`), which
is supported via `tf.Tensor.__getitem__` and `tf.Variable.__getitem__`.**
The interface of this op is a low-level encoding of the slicing syntax.
Roughly speaking, this op extracts a slice of size `(end-begin)/stride`
from the given `input_` tensor. Starting at the location specified by `begin`
the slice continues by adding `stride` to the index until all dimensions are
not less than `end`.
Note that a stride can be negative, which causes a reverse slice.
Given a Python slice `input[spec0, spec1, ..., specn]`,
this function will be called as follows.
`begin`, `end`, and `strides` will be vectors of length n.
n in general is not equal to the rank of the `input_` tensor.
In each mask field (`begin_mask`, `end_mask`, `ellipsis_mask`,
`new_axis_mask`, `shrink_axis_mask`) the ith bit will correspond to
the ith spec.
If the ith bit of `begin_mask` is set, `begin[i]` is ignored and
the fullest possible range in that dimension is used instead.
`end_mask` works analogously, except with the end range.
`foo[5:,:,:3]` on a 7x8x9 tensor is equivalent to `foo[5:7,0:8,0:3]`.
`foo[::-1]` reverses a tensor with shape 8.
If the ith bit of `ellipsis_mask` is set, as many unspecified dimensions
as needed will be inserted between other dimensions. Only one
non-zero bit is allowed in `ellipsis_mask`.
For example `foo[3:5,...,4:5]` on a shape 10x3x3x10 tensor is
equivalent to `foo[3:5,:,:,4:5]` and
`foo[3:5,...]` is equivalent to `foo[3:5,:,:,:]`.
If the ith bit of `new_axis_mask` is set, then `begin`,
`end`, and `stride` are ignored and a new length 1 dimension is
added at this point in the output tensor.
For example,
`foo[:4, tf.newaxis, :2]` would produce a shape `(4, 1, 2)` tensor.
If the ith bit of `shrink_axis_mask` is set, it implies that the ith
specification shrinks the dimensionality by 1, taking on the value at index
`begin[i]`. `end[i]` and `strides[i]` are ignored in this case. For example in
Python one might do `foo[:, 3, :]` which would result in `shrink_axis_mask`
equal to 2.
NOTE: `begin` and `end` are zero-indexed.
`strides` entries must be non-zero.
```python
t = tf.constant([[[1, 1, 1], [2, 2, 2]],
[[3, 3, 3], [4, 4, 4]],
[[5, 5, 5], [6, 6, 6]]])
tf.strided_slice(t, [1, 0, 0], [2, 1, 3], [1, 1, 1]) # [[[3, 3, 3]]]
tf.strided_slice(t, [1, 0, 0], [2, 2, 3], [1, 1, 1]) # [[[3, 3, 3],
# [4, 4, 4]]]
tf.strided_slice(t, [1, -1, 0], [2, -3, 3], [1, -1, 1]) # [[[4, 4, 4],
# [3, 3, 3]]]
```
Args:
input_: A `Tensor`.
begin: An `int32` or `int64` `Tensor`.
end: An `int32` or `int64` `Tensor`.
strides: An `int32` or `int64` `Tensor`.
begin_mask: An `int32` mask.
end_mask: An `int32` mask.
ellipsis_mask: An `int32` mask.
new_axis_mask: An `int32` mask.
shrink_axis_mask: An `int32` mask.
var: The variable corresponding to `input_` or None
name: A name for the operation (optional).
Returns:
A `Tensor` the same type as `input`.
"""
if strides is None:
strides = ones_like(begin)
op = gen_array_ops.strided_slice(
input=input_,
begin=begin,
end=end,
strides=strides,
name=name,
begin_mask=begin_mask,
end_mask=end_mask,
ellipsis_mask=ellipsis_mask,
new_axis_mask=new_axis_mask,
shrink_axis_mask=shrink_axis_mask)
parent_name = name
if not (var is None and isinstance(op, ops.EagerTensor)):
def assign(val, name=None):
"""Closure that holds all the arguments to create an assignment."""
if var is None:
raise ValueError("Sliced assignment is only supported for variables")
else:
if name is None:
name = parent_name + "_assign"
return var._strided_slice_assign(
begin=begin,
end=end,
strides=strides,
value=val,
name=name,
begin_mask=begin_mask,
end_mask=end_mask,
ellipsis_mask=ellipsis_mask,
new_axis_mask=new_axis_mask,
shrink_axis_mask=shrink_axis_mask)
op.assign = assign
return op
def _SliceHelperVar(var, slice_spec):
"""Creates a slice helper object given a variable.
This allows creating a sub-tensor from part of the current contents
of a variable. See `tf.Tensor.__getitem__` for detailed examples
of slicing.
This function in addition also allows assignment to a sliced range.
This is similar to `__setitem__` functionality in Python. However,
the syntax is different so that the user can capture the assignment
operation for grouping or passing to `sess.run()`.
For example,
```python
import tensorflow as tf
A = tf.Variable([[1,2,3], [4,5,6], [7,8,9]], dtype=tf.float32)
with tf.compat.v1.Session() as sess:
sess.run(tf.compat.v1.global_variables_initializer())
print(sess.run(A[:2, :2])) # => [[1,2], [4,5]]
op = A[:2,:2].assign(22. * tf.ones((2, 2)))
print(sess.run(op)) # => [[22, 22, 3], [22, 22, 6], [7,8,9]]
```
Note that assignments currently do not support NumPy broadcasting
semantics.
Args:
var: An `ops.Variable` object.
slice_spec: The arguments to `Tensor.__getitem__`.
Returns:
The appropriate slice of "tensor", based on "slice_spec".
As an operator. The operator also has a `assign()` method
that can be used to generate an assignment operator.
Raises:
ValueError: If a slice range is negative size.
TypeError: TypeError: If the slice indices aren't int, slice,
ellipsis, tf.newaxis or int32/int64 tensors.
"""
return _slice_helper(var.value(), slice_spec, var)
ops.Tensor._override_operator("__getitem__", _slice_helper)
@tf_export("parallel_stack")
def parallel_stack(values, name="parallel_stack"):
"""Stacks a list of rank-`R` tensors into one rank-`(R+1)` tensor in parallel.
Requires that the shape of inputs be known at graph construction time.
Packs the list of tensors in `values` into a tensor with rank one higher than
each tensor in `values`, by packing them along the first dimension.
Given a list of length `N` of tensors of shape `(A, B, C)`; the `output`
tensor will have the shape `(N, A, B, C)`.
For example:
```python
x = tf.constant([1, 4])
y = tf.constant([2, 5])
z = tf.constant([3, 6])
tf.parallel_stack([x, y, z]) # [[1, 4], [2, 5], [3, 6]]
```
The difference between `stack` and `parallel_stack` is that `stack` requires
all the inputs be computed before the operation will begin but doesn't require
that the input shapes be known during graph construction.
`parallel_stack` will copy pieces of the input into the output as they become
available, in some situations this can provide a performance benefit.
Unlike `stack`, `parallel_stack` does NOT support backpropagation.
This is the opposite of unstack. The numpy equivalent is
tf.parallel_stack([x, y, z]) = np.asarray([x, y, z])
Args:
values: A list of `Tensor` objects with the same shape and type.
name: A name for this operation (optional).
Returns:
output: A stacked `Tensor` with the same type as `values`.
"""
with ops.name_scope(name):
value_t = ops.convert_to_tensor(values[0])
value_shape = ops.convert_to_tensor(value_t).get_shape()
output_shape = tensor_shape.TensorShape([len(values)])
output_shape = output_shape.concatenate(value_shape)
# expand_dims converts concat to stack.
return gen_array_ops.parallel_concat(
[expand_dims(value, 0) for value in values], shape=output_shape)
@tf_export("stack")
@dispatch.add_dispatch_support
def stack(values, axis=0, name="stack"):
"""Stacks a list of rank-`R` tensors into one rank-`(R+1)` tensor.
Packs the list of tensors in `values` into a tensor with rank one higher than
each tensor in `values`, by packing them along the `axis` dimension.
Given a list of length `N` of tensors of shape `(A, B, C)`;
if `axis == 0` then the `output` tensor will have the shape `(N, A, B, C)`.
if `axis == 1` then the `output` tensor will have the shape `(A, N, B, C)`.
Etc.
For example:
```python
x = tf.constant([1, 4])
y = tf.constant([2, 5])
z = tf.constant([3, 6])
tf.stack([x, y, z]) # [[1, 4], [2, 5], [3, 6]] (Pack along first dim.)
tf.stack([x, y, z], axis=1) # [[1, 2, 3], [4, 5, 6]]
```
This is the opposite of unstack. The numpy equivalent is
```python
tf.stack([x, y, z]) = np.stack([x, y, z])
```
Args:
values: A list of `Tensor` objects with the same shape and type.
axis: An `int`. The axis to stack along. Defaults to the first dimension.
Negative values wrap around, so the valid range is `[-(R+1), R+1)`.
name: A name for this operation (optional).
Returns:
output: A stacked `Tensor` with the same type as `values`.
Raises:
ValueError: If `axis` is out of the range [-(R+1), R+1).
"""
if axis == 0:
try:
# If the input is a constant list, it can be converted to a constant op
return ops.convert_to_tensor(values, name=name)
except (TypeError, ValueError):
pass # Input list contains non-constant tensors
value_shape = ops.convert_to_tensor(values[0], name=name)._shape_tuple() # pylint: disable=protected-access
if value_shape is not None:
expanded_num_dims = len(value_shape) + 1
if axis < -expanded_num_dims or axis >= expanded_num_dims:
raise ValueError("axis = %d not in [%d, %d)" %
(axis, -expanded_num_dims, expanded_num_dims))
return gen_array_ops.pack(values, axis=axis, name=name)
# pylint: disable=invalid-name
def _autopacking_helper(list_or_tuple, dtype, name):
"""Converts the given list or tuple to a tensor by packing.
Args:
list_or_tuple: A (possibly nested) list or tuple containing a tensor.
dtype: The element type of the returned tensor.
name: A name for the returned tensor.
Returns:
A `tf.Tensor` with value equivalent to `list_or_tuple`.
"""
if context.executing_eagerly():
# NOTE: Fast path when all the items are tensors, this doesn't do any type
# checking.
if all(ops.is_dense_tensor_like(elem) for elem in list_or_tuple):
return gen_array_ops.pack(list_or_tuple, name=name)
must_pack = False
converted_elems = []
with ops.name_scope(name) as scope:
for i, elem in enumerate(list_or_tuple):
if ops.is_dense_tensor_like(elem):
if dtype is not None and elem.dtype.base_dtype != dtype:
raise TypeError("Cannot convert a list containing a tensor of dtype "
"%s to %s (Tensor is: %r)" %
(elem.dtype, dtype, elem))
converted_elems.append(elem)
must_pack = True
elif isinstance(elem, (list, tuple)):
converted_elem = _autopacking_helper(elem, dtype, str(i))
if ops.is_dense_tensor_like(converted_elem):
must_pack = True
converted_elems.append(converted_elem)
else:
converted_elems.append(elem)
if must_pack:
elems_as_tensors = []
for i, elem in enumerate(converted_elems):
if ops.is_dense_tensor_like(elem):
elems_as_tensors.append(elem)
else:
# NOTE(mrry): This is inefficient, but it enables us to
# handle the case where the list arguments are other
# convertible-to-tensor types, such as numpy arrays.
elems_as_tensors.append(
constant_op.constant(elem, dtype=dtype, name=str(i)))
return gen_array_ops.pack(elems_as_tensors, name=scope)
else:
return converted_elems
def _get_dtype_from_nested_lists(list_or_tuple):
"""Returns the dtype of any tensor-like object in `list_or_tuple`, if found.
Args:
list_or_tuple: A list or tuple representing an object that can be converted
to a `tf.Tensor`.
Returns:
The dtype of any tensor-like object in `list_or_tuple`, or `None` if no
such object exists.
"""
for elem in list_or_tuple:
if ops.is_dense_tensor_like(elem):
return elem.dtype.base_dtype
elif isinstance(elem, (list, tuple)):
maybe_dtype = _get_dtype_from_nested_lists(elem)
if maybe_dtype is not None:
return maybe_dtype
return None
def _cast_nested_seqs_to_dtype(dtype):
def _maybe_cast(elem):
if ops.is_dense_tensor_like(elem):
if dtype != elem.dtype.base_dtype:
elem = gen_math_ops.cast(elem, dtype)
return elem
return _maybe_cast
def _autopacking_conversion_function(v, dtype=None, name=None, as_ref=False):
"""Tensor conversion function that automatically packs arguments."""
if as_ref:
return NotImplemented
inferred_dtype = _get_dtype_from_nested_lists(v)
if inferred_dtype is None:
# We did not find any tensor-like objects in the nested lists, so defer to
# other conversion functions.
return NotImplemented
if dtype is None:
dtype = inferred_dtype
elif dtype != inferred_dtype:
v = nest.map_structure(_cast_nested_seqs_to_dtype(dtype), v)
return _autopacking_helper(v, dtype, name or "packed")
# pylint: enable=invalid-name
# NOTE: Register this conversion function to run *before* one that
# assumes every element is a value.
ops.register_tensor_conversion_function((list, tuple),
_autopacking_conversion_function, 99)
@tf_export("unstack")
def unstack(value, num=None, axis=0, name="unstack"):
"""Unpacks the given dimension of a rank-`R` tensor into rank-`(R-1)` tensors.
Unpacks `num` tensors from `value` by chipping it along the `axis` dimension.
If `num` is not specified (the default), it is inferred from `value`'s shape.
If `value.shape[axis]` is not known, `ValueError` is raised.
For example, given a tensor of shape `(A, B, C, D)`;
If `axis == 0` then the i'th tensor in `output` is the slice
`value[i, :, :, :]` and each tensor in `output` will have shape `(B, C, D)`.
(Note that the dimension unpacked along is gone, unlike `split`).
If `axis == 1` then the i'th tensor in `output` is the slice
`value[:, i, :, :]` and each tensor in `output` will have shape `(A, C, D)`.
Etc.
This is the opposite of stack.
Args:
value: A rank `R > 0` `Tensor` to be unstacked.
num: An `int`. The length of the dimension `axis`. Automatically inferred if
`None` (the default).
axis: An `int`. The axis to unstack along. Defaults to the first dimension.
Negative values wrap around, so the valid range is `[-R, R)`.
name: A name for the operation (optional).
Returns:
The list of `Tensor` objects unstacked from `value`.
Raises:
ValueError: If `num` is unspecified and cannot be inferred.
ValueError: If `axis` is out of the range [-R, R).
"""
if num is None:
value = ops.convert_to_tensor(value)
value_shape = value.get_shape()
if value_shape.ndims is not None:
if axis < -value_shape.ndims or axis >= value_shape.ndims:
raise ValueError("axis = %d not in [%d, %d)" %
(axis, -value_shape.ndims, value_shape.ndims))
num = value_shape.dims[axis].value
if num is None:
raise ValueError("Cannot infer num from shape %s" % value_shape)
return gen_array_ops.unpack(value, num=num, axis=axis, name=name)
@tf_export("concat")
@dispatch.add_dispatch_support
def concat(values, axis, name="concat"):
"""Concatenates tensors along one dimension.
Concatenates the list of tensors `values` along dimension `axis`. If
`values[i].shape = [D0, D1, ... Daxis(i), ...Dn]`, the concatenated
result has shape
[D0, D1, ... Raxis, ...Dn]
where
Raxis = sum(Daxis(i))
That is, the data from the input tensors is joined along the `axis`
dimension.
The number of dimensions of the input tensors must match, and all dimensions
except `axis` must be equal.
For example:
```python
t1 = [[1, 2, 3], [4, 5, 6]]
t2 = [[7, 8, 9], [10, 11, 12]]
tf.concat([t1, t2], 0) # [[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12]]
tf.concat([t1, t2], 1) # [[1, 2, 3, 7, 8, 9], [4, 5, 6, 10, 11, 12]]
# tensor t3 with shape [2, 3]
# tensor t4 with shape [2, 3]
tf.shape(tf.concat([t3, t4], 0)) # [4, 3]
tf.shape(tf.concat([t3, t4], 1)) # [2, 6]
```
As in Python, the `axis` could also be negative numbers. Negative `axis`
are interpreted as counting from the end of the rank, i.e.,
`axis + rank(values)`-th dimension.
For example:
```python
t1 = [[[1, 2], [2, 3]], [[4, 4], [5, 3]]]
t2 = [[[7, 4], [8, 4]], [[2, 10], [15, 11]]]
tf.concat([t1, t2], -1)
```
would produce:
```python
[[[ 1, 2, 7, 4],
[ 2, 3, 8, 4]],
[[ 4, 4, 2, 10],
[ 5, 3, 15, 11]]]
```
Note: If you are concatenating along a new axis consider using stack.
E.g.
```python
tf.concat([tf.expand_dims(t, axis) for t in tensors], axis)
```
can be rewritten as
```python
tf.stack(tensors, axis=axis)
```
Args:
values: A list of `Tensor` objects or a single `Tensor`.
axis: 0-D `int32` `Tensor`. Dimension along which to concatenate. Must be
in the range `[-rank(values), rank(values))`. As in Python, indexing for
axis is 0-based. Positive axis in the rage of `[0, rank(values))` refers
to `axis`-th dimension. And negative axis refers to `axis +
rank(values)`-th dimension.
name: A name for the operation (optional).
Returns:
A `Tensor` resulting from concatenation of the input tensors.
"""
if not isinstance(values, (list, tuple)):
values = [values]
# TODO(mrry): Change to return values?
if len(values) == 1: # Degenerate case of one tensor.
# Make a throwaway call to convert_to_tensor to make sure
# that axis is of the correct type, and make sure that
# the returned tensor is a scalar.
# TODO(keveman): Implement a standalone type and shape checker.
with ops.name_scope(name) as scope:
ops.convert_to_tensor(
axis, name="concat_dim",
dtype=dtypes.int32).get_shape().assert_is_compatible_with(
tensor_shape.scalar())
return identity(values[0], name=scope)
return gen_array_ops.concat_v2(values=values, axis=axis, name=name)
@tf_export(v1=["boolean_mask"])
def boolean_mask(tensor, mask, name="boolean_mask", axis=None):
"""Apply boolean mask to tensor.
Numpy equivalent is `tensor[mask]`.
```python
# 1-D example
tensor = [0, 1, 2, 3]
mask = np.array([True, False, True, False])
boolean_mask(tensor, mask) # [0, 2]
```
In general, `0 < dim(mask) = K <= dim(tensor)`, and `mask`'s shape must match
the first K dimensions of `tensor`'s shape. We then have:
`boolean_mask(tensor, mask)[i, j1,...,jd] = tensor[i1,...,iK,j1,...,jd]`
where `(i1,...,iK)` is the ith `True` entry of `mask` (row-major order).
The `axis` could be used with `mask` to indicate the axis to mask from.
In that case, `axis + dim(mask) <= dim(tensor)` and `mask`'s shape must match
the first `axis + dim(mask)` dimensions of `tensor`'s shape.
See also: `tf.ragged.boolean_mask`, which can be applied to both dense and
ragged tensors, and can be used if you need to preserve the masked dimensions
of `tensor` (rather than flattening them, as `tf.boolean_mask` does).
Args:
tensor: N-D tensor.
mask: K-D boolean tensor, K <= N and K must be known statically.
name: A name for this operation (optional).
axis: A 0-D int Tensor representing the axis in `tensor` to mask from. By
default, axis is 0 which will mask from the first dimension. Otherwise K +
axis <= N.
Returns:
(N-K+1)-dimensional tensor populated by entries in `tensor` corresponding
to `True` values in `mask`.
Raises:
ValueError: If shapes do not conform.
Examples:
```python
# 2-D example
tensor = [[1, 2], [3, 4], [5, 6]]
mask = np.array([True, False, True])
boolean_mask(tensor, mask) # [[1, 2], [5, 6]]
```
"""
def _apply_mask_1d(reshaped_tensor, mask, axis=None):
"""Mask tensor along dimension 0 with a 1-D mask."""
indices = squeeze(where(mask), axis=[1])
return gather(reshaped_tensor, indices, axis=axis)
with ops.name_scope(name, values=[tensor, mask]):
tensor = ops.convert_to_tensor(tensor, name="tensor")
mask = ops.convert_to_tensor(mask, name="mask")
shape_mask = mask.get_shape()
ndims_mask = shape_mask.ndims
shape_tensor = tensor.get_shape()
if ndims_mask == 0:
raise ValueError("mask cannot be scalar.")
if ndims_mask is None:
raise ValueError(
"Number of mask dimensions must be specified, even if some dimensions"
" are None. E.g. shape=[None] is ok, but shape=None is not.")
axis = 0 if axis is None else axis
shape_tensor[axis:axis + ndims_mask].assert_is_compatible_with(shape_mask)
leading_size = gen_math_ops.prod(shape(tensor)[axis:axis + ndims_mask], [0])
tensor = reshape(
tensor,
concat([
shape(tensor)[:axis], [leading_size],
shape(tensor)[axis + ndims_mask:]
], 0))
first_dim = shape_tensor[axis:axis + ndims_mask].num_elements()
tensor.set_shape(
tensor_shape.as_shape(shape_tensor[:axis]).concatenate(
[first_dim]).concatenate(shape_tensor[axis + ndims_mask:]))
mask = reshape(mask, [-1])
return _apply_mask_1d(tensor, mask, axis)
@tf_export("boolean_mask", v1=[])
@dispatch.add_dispatch_support
def boolean_mask_v2(tensor, mask, axis=None, name="boolean_mask"):
"""Apply boolean mask to tensor.
Numpy equivalent is `tensor[mask]`.
```python
# 1-D example
tensor = [0, 1, 2, 3]
mask = np.array([True, False, True, False])
boolean_mask(tensor, mask) # [0, 2]
```
In general, `0 < dim(mask) = K <= dim(tensor)`, and `mask`'s shape must match
the first K dimensions of `tensor`'s shape. We then have:
`boolean_mask(tensor, mask)[i, j1,...,jd] = tensor[i1,...,iK,j1,...,jd]`
where `(i1,...,iK)` is the ith `True` entry of `mask` (row-major order).
The `axis` could be used with `mask` to indicate the axis to mask from.
In that case, `axis + dim(mask) <= dim(tensor)` and `mask`'s shape must match
the first `axis + dim(mask)` dimensions of `tensor`'s shape.
See also: `tf.ragged.boolean_mask`, which can be applied to both dense and
ragged tensors, and can be used if you need to preserve the masked dimensions
of `tensor` (rather than flattening them, as `tf.boolean_mask` does).
Args:
tensor: N-D tensor.
mask: K-D boolean tensor, K <= N and K must be known statically.
axis: A 0-D int Tensor representing the axis in `tensor` to mask from. By
default, axis is 0 which will mask from the first dimension. Otherwise K +
axis <= N.
name: A name for this operation (optional).
Returns:
(N-K+1)-dimensional tensor populated by entries in `tensor` corresponding
to `True` values in `mask`.
Raises:
ValueError: If shapes do not conform.
Examples:
```python
# 2-D example
tensor = [[1, 2], [3, 4], [5, 6]]
mask = np.array([True, False, True])
boolean_mask(tensor, mask) # [[1, 2], [5, 6]]
```
"""
return boolean_mask(tensor, mask, name, axis)
@tf_export("sparse.mask", v1=["sparse.mask", "sparse_mask"])
@deprecation.deprecated_endpoints("sparse_mask")
def sparse_mask(a, mask_indices, name=None):
"""Masks elements of `IndexedSlices`.
Given an `IndexedSlices` instance `a`, returns another `IndexedSlices` that
contains a subset of the slices of `a`. Only the slices at indices not
specified in `mask_indices` are returned.
This is useful when you need to extract a subset of slices in an
`IndexedSlices` object.
For example:
```python
# `a` contains slices at indices [12, 26, 37, 45] from a large tensor
# with shape [1000, 10]
a.indices # [12, 26, 37, 45]
tf.shape(a.values) # [4, 10]
# `b` will be the subset of `a` slices at its second and third indices, so
# we want to mask its first and last indices (which are at absolute
# indices 12, 45)
b = tf.sparse.mask(a, [12, 45])
b.indices # [26, 37]
tf.shape(b.values) # [2, 10]
```
Args:
a: An `IndexedSlices` instance.
mask_indices: Indices of elements to mask.
name: A name for the operation (optional).
Returns:
The masked `IndexedSlices` instance.
"""
with ops.name_scope(name, "sparse_mask", [a, mask_indices]) as name:
indices = a.indices
out_indices, to_gather = gen_array_ops.list_diff(indices, mask_indices)
out_values = gather(a.values, to_gather, name=name)
return ops.IndexedSlices(out_values, out_indices, a.dense_shape)
@tf_export("unique")
def unique(x, out_idx=dtypes.int32, name=None):
# TODO(yongtang): switch to v2 once API deprecation
# period (3 weeks) pass.
# TODO(yongtang): The documentation should also
# be updated when switch to v2.
return gen_array_ops.unique(x, out_idx, name)
unique.__doc__ = gen_array_ops.unique.__doc__
@tf_export("unique_with_counts")
def unique_with_counts(x, out_idx=dtypes.int32, name=None):
# TODO(yongtang): switch to v2 once API deprecation
# period (3 weeks) pass.
# TODO(yongtang): The documentation should also
# be updated when switch to v2.
return gen_array_ops.unique_with_counts(x, out_idx, name)
unique_with_counts.__doc__ = gen_array_ops.unique_with_counts.__doc__
@tf_export("split")
def split(value, num_or_size_splits, axis=0, num=None, name="split"):
"""Splits a tensor into sub tensors.
If `num_or_size_splits` is an integer, then `value` is split along dimension
`axis` into `num_split` smaller tensors. This requires that `num_split` evenly
divides `value.shape[axis]`.
If `num_or_size_splits` is a 1-D Tensor (or list), we call it `size_splits`
and `value` is split into `len(size_splits)` elements. The shape of the `i`-th
element has the same size as the `value` except along dimension `axis` where
the size is `size_splits[i]`.
For example:
```python
# 'value' is a tensor with shape [5, 30]
# Split 'value' into 3 tensors with sizes [4, 15, 11] along dimension 1
split0, split1, split2 = tf.split(value, [4, 15, 11], 1)
tf.shape(split0) # [5, 4]
tf.shape(split1) # [5, 15]
tf.shape(split2) # [5, 11]
# Split 'value' into 3 tensors along dimension 1
split0, split1, split2 = tf.split(value, num_or_size_splits=3, axis=1)
tf.shape(split0) # [5, 10]
```
Args:
value: The `Tensor` to split.
num_or_size_splits: Either an integer indicating the number of splits along
split_dim or a 1-D integer `Tensor` or Python list containing the sizes of
each output tensor along split_dim. If a scalar then it must evenly divide
`value.shape[axis]`; otherwise the sum of sizes along the split dimension
must match that of the `value`.
axis: An integer or scalar `int32` `Tensor`. The dimension along which to
split. Must be in the range `[-rank(value), rank(value))`. Defaults to 0.
num: Optional, used to specify the number of outputs when it cannot be
inferred from the shape of `size_splits`.
name: A name for the operation (optional).
Returns:
if `num_or_size_splits` is a scalar returns `num_or_size_splits` `Tensor`
objects; if `num_or_size_splits` is a 1-D Tensor returns
`num_or_size_splits.get_shape[0]` `Tensor` objects resulting from splitting
`value`.
Raises:
ValueError: If `num` is unspecified and cannot be inferred.
"""
size_splits = ops.convert_to_tensor(num_or_size_splits)
if isinstance(num_or_size_splits,
six.integer_types + (tensor_shape.Dimension,)):
return gen_array_ops.split(
axis=axis, num_split=num_or_size_splits, value=value, name=name)
if size_splits._rank() == 0:
raise ValueError(
"Rank-0 tensors are not supported as the num_or_size_splits argument "
"to split. Argument provided: %s" % (num_or_size_splits,))
if num is None:
size_splits_shape = size_splits._shape_tuple()
if size_splits_shape:
num = size_splits_shape[0]
if num is None:
raise ValueError("Cannot infer num from shape %s" % num_or_size_splits)
return gen_array_ops.split_v(
value=value, size_splits=size_splits, axis=axis, num_split=num, name=name)
@tf_export("transpose", v1=[])
def transpose_v2(a, perm=None, conjugate=False, name="transpose"):
"""Transposes `a`.
Permutes the dimensions according to `perm`.
The returned tensor's dimension i will correspond to the input dimension
`perm[i]`. If `perm` is not given, it is set to (n-1...0), where n is
the rank of the input tensor. Hence by default, this operation performs a
regular matrix transpose on 2-D input Tensors. If conjugate is True and
`a.dtype` is either `complex64` or `complex128` then the values of `a`
are conjugated and transposed.
@compatibility(numpy)
In `numpy` transposes are memory-efficient constant time operations as they
simply return a new view of the same data with adjusted `strides`.
TensorFlow does not support strides, so `transpose` returns a new tensor with
the items permuted.
@end_compatibility
For example:
```python
x = tf.constant([[1, 2, 3], [4, 5, 6]])
tf.transpose(x) # [[1, 4]
# [2, 5]
# [3, 6]]
# Equivalently
tf.transpose(x, perm=[1, 0]) # [[1, 4]
# [2, 5]
# [3, 6]]
# If x is complex, setting conjugate=True gives the conjugate transpose
x = tf.constant([[1 + 1j, 2 + 2j, 3 + 3j],
[4 + 4j, 5 + 5j, 6 + 6j]])
tf.transpose(x, conjugate=True) # [[1 - 1j, 4 - 4j],
# [2 - 2j, 5 - 5j],
# [3 - 3j, 6 - 6j]]
# 'perm' is more useful for n-dimensional tensors, for n > 2
x = tf.constant([[[ 1, 2, 3],
[ 4, 5, 6]],
[[ 7, 8, 9],
[10, 11, 12]]])
# Take the transpose of the matrices in dimension-0
# (this common operation has a shorthand `linalg.matrix_transpose`)
tf.transpose(x, perm=[0, 2, 1]) # [[[1, 4],
# [2, 5],
# [3, 6]],
# [[7, 10],
# [8, 11],
# [9, 12]]]
```
Args:
a: A `Tensor`.
perm: A permutation of the dimensions of `a`.
conjugate: Optional bool. Setting it to `True` is mathematically equivalent
to tf.math.conj(tf.transpose(input)).
name: A name for the operation (optional).
Returns:
A transposed `Tensor`.
"""
return transpose(a=a, perm=perm, name=name, conjugate=conjugate)
@tf_export(v1=["transpose"])
def transpose(a, perm=None, name="transpose", conjugate=False):
"""Transposes `a`.
Permutes the dimensions according to `perm`.
The returned tensor's dimension i will correspond to the input dimension
`perm[i]`. If `perm` is not given, it is set to (n-1...0), where n is
the rank of the input tensor. Hence by default, this operation performs a
regular matrix transpose on 2-D input Tensors. If conjugate is True and
`a.dtype` is either `complex64` or `complex128` then the values of `a`
are conjugated and transposed.
@compatibility(numpy)
In `numpy` transposes are memory-efficient constant time operations as they
simply return a new view of the same data with adjusted `strides`.
TensorFlow does not support strides, so `transpose` returns a new tensor with
the items permuted.
@end_compatibility
For example:
```python
x = tf.constant([[1, 2, 3], [4, 5, 6]])
tf.transpose(x) # [[1, 4]
# [2, 5]
# [3, 6]]
# Equivalently
tf.transpose(x, perm=[1, 0]) # [[1, 4]
# [2, 5]
# [3, 6]]
# If x is complex, setting conjugate=True gives the conjugate transpose
x = tf.constant([[1 + 1j, 2 + 2j, 3 + 3j],
[4 + 4j, 5 + 5j, 6 + 6j]])
tf.transpose(x, conjugate=True) # [[1 - 1j, 4 - 4j],
# [2 - 2j, 5 - 5j],
# [3 - 3j, 6 - 6j]]
# 'perm' is more useful for n-dimensional tensors, for n > 2
x = tf.constant([[[ 1, 2, 3],
[ 4, 5, 6]],
[[ 7, 8, 9],
[10, 11, 12]]])
# Take the transpose of the matrices in dimension-0
# (this common operation has a shorthand `linalg.matrix_transpose`)
tf.transpose(x, perm=[0, 2, 1]) # [[[1, 4],
# [2, 5],
# [3, 6]],
# [[7, 10],
# [8, 11],
# [9, 12]]]
```
Args:
a: A `Tensor`.
perm: A permutation of the dimensions of `a`.
name: A name for the operation (optional).
conjugate: Optional bool. Setting it to `True` is mathematically equivalent
to tf.math.conj(tf.transpose(input)).
Returns:
A transposed `Tensor`.
"""
with ops.name_scope(name, "transpose", [a]) as name:
transpose_fn = (
gen_array_ops.conjugate_transpose if
(conjugate and a.dtype.is_complex) else gen_array_ops.transpose)
if perm is None:
a = ops.convert_to_tensor(a, name="a")
if not a.get_shape().ndims:
rank = gen_array_ops.rank(a)
perm = (rank - 1) - gen_math_ops._range(0, rank, 1)
else:
rank = a.get_shape().ndims
perm = (rank - 1) - np.arange(rank)
ret = transpose_fn(a, perm, name=name)
# NOTE(mrry): Setting the shape explicitly because
# reverse is not handled by the shape function.
if not context.executing_eagerly():
input_shape = ret.op.inputs[0].get_shape().dims
if input_shape is not None:
ret.set_shape(input_shape[::-1])
else:
ret = transpose_fn(a, perm, name=name)
return ret
# pylint: disable=invalid-name
@tf_export(
"linalg.matrix_transpose",
v1=["linalg.transpose", "linalg.matrix_transpose", "matrix_transpose"])
@deprecation.deprecated_endpoints("matrix_transpose", "linalg.transpose")
def matrix_transpose(a, name="matrix_transpose", conjugate=False):
"""Transposes last two dimensions of tensor `a`.
For example:
```python
x = tf.constant([[1, 2, 3], [4, 5, 6]])
tf.linalg.matrix_transpose(x) # [[1, 4],
# [2, 5],
# [3, 6]]
x = tf.constant([[1 + 1j, 2 + 2j, 3 + 3j],
[4 + 4j, 5 + 5j, 6 + 6j]])
tf.linalg.matrix_transpose(x, conjugate=True) # [[1 - 1j, 4 - 4j],
# [2 - 2j, 5 - 5j],
# [3 - 3j, 6 - 6j]]
# Matrix with two batch dimensions.
# x.shape is [1, 2, 3, 4]
# tf.linalg.matrix_transpose(x) is shape [1, 2, 4, 3]
```
Note that `tf.matmul` provides kwargs allowing for transpose of arguments.
This is done with minimal cost, and is preferable to using this function. E.g.
```python
# Good! Transpose is taken at minimal additional cost.
tf.matmul(matrix, b, transpose_b=True)
# Inefficient!
tf.matmul(matrix, tf.linalg.matrix_transpose(b))
```
@compatibility(numpy)
In `numpy` transposes are memory-efficient constant time operations as they
simply return a new view of the same data with adjusted `strides`.
TensorFlow does not support strides, `linalg.matrix_transpose` returns a new
tensor with the items permuted.
@end_compatibility
Args:
a: A `Tensor` with `rank >= 2`.
name: A name for the operation (optional).
conjugate: Optional bool. Setting it to `True` is mathematically equivalent
to tf.math.conj(tf.linalg.matrix_transpose(input)).
Returns:
A transposed batch matrix `Tensor`.
Raises:
ValueError: If `a` is determined statically to have `rank < 2`.
"""
with ops.name_scope(name, values=[a]):
a = ops.convert_to_tensor(a, name="a")
# If we know the number of dimensions (statically), we can do two things:
# 1. Check that `a` is a (batch) matrix.
# 2. Use a python list for perm. This preserves static shape information
# and avoids extra computations.
a_shape = a.get_shape()
ndims = a_shape.ndims
if ndims is not None:
if ndims < 2:
raise ValueError(
"Argument 'a' should be a (batch) matrix, with rank >= 2. Found: "
"%s" % a_shape)
perm = list(range(ndims - 2)) + [ndims - 1] + [ndims - 2]
else:
a_rank = rank(a)
perm = concat(
(gen_math_ops._range(0, a_rank - 2, 1), [a_rank - 1, a_rank - 2]), 0)
return transpose(a, perm=perm, conjugate=conjugate)
@tf_export("linalg.diag", v1=["linalg.diag", "matrix_diag"])
@deprecation.deprecated_endpoints("matrix_diag")
def matrix_diag(diagonal,
name="diag",
k=0,
num_rows=-1,
num_cols=-1,
padding_value=0):
"""Returns a batched diagonal tensor with given batched diagonal values.
Returns a tensor with the contents in `diagonal` as `k[0]`-th to `k[1]`-th
diagonals of a matrix, with everything else padded with `padding`. `num_rows`
and `num_cols` specify the dimension of the innermost matrix of the output. If
both are not specified, the op assumes the innermost matrix is square and
infers its size from `k` and the innermost dimension of `diagonal`. If only
one of them is specified, the op assumes the unspecified value is the smallest
possible based on other criteria.
Let `diagonal` have `r` dimensions `[I, J, ..., L, M, N]`. The output tensor
has rank `r+1` with shape `[I, J, ..., L, M, num_rows, num_cols]` when only
one diagonal is given (`k` is an integer or `k[0] == k[1]`). Otherwise, it has
rank `r` with shape `[I, J, ..., L, num_rows, num_cols]`.
The second innermost dimension of `diagonal` has double meaning. When `k` is
scalar or `k[0] == k[1]`, `M` is part of the batch size [I, J, ..., M], and
the output tensor is:
```
output[i, j, ..., l, m, n]
= diagonal[i, j, ..., l, n-max(d_upper, 0)] ; if n - m == d_upper
output[i, j, ..., l, m, n] ; otherwise
```
Otherwise, `M` is treated as the number of diagonals for the matrix in the
same batch (`M = k[1]-k[0]+1`), and the output tensor is:
```
output[i, j, ..., l, m, n]
= diagonal[i, j, ..., l, k[1]-d, n-max(d, 0)] ; if d_lower <= d <= d_upper
input[i, j, ..., l, m, n] ; otherwise
```
where `d = n - m`
For example:
```
# The main diagonal.
diagonal = np.array([[1, 2, 3, 4], # Input shape: (2, 4)
[5, 6, 7, 8]])
tf.matrix_diag(diagonal) ==> [[[1, 0, 0, 0], # Output shape: (2, 4, 4)
[0, 2, 0, 0],
[0, 0, 3, 0],
[0, 0, 0, 4]],
[[5, 0, 0, 0],
[0, 6, 0, 0],
[0, 0, 7, 0],
[0, 0, 0, 8]]]
# A superdiagonal (per batch).
diagonal = np.array([[1, 2, 3], # Input shape: (2, 3)
[4, 5, 6]])
tf.matrix_diag(diagonal, k = 1)
==> [[[0, 1, 0, 0], # Output shape: (2, 4, 4)
[0, 0, 2, 0],
[0, 0, 0, 3],
[0, 0, 0, 0]],
[[0, 4, 0, 0],
[0, 0, 5, 0],
[0, 0, 0, 6],
[0, 0, 0, 0]]]
# A band of diagonals.
diagonals = np.array([[[1, 2, 3], # Input shape: (2, 2, 3)
[4, 5, 0]],
[[6, 7, 9],
[9, 1, 0]]])
tf.matrix_diag(diagonals, k = (-1, 0))
==> [[[1, 0, 0], # Output shape: (2, 3, 3)
[4, 2, 0],
[0, 5, 3]],
[[6, 0, 0],
[9, 7, 0],
[0, 1, 9]]]
# Rectangular matrix.
diagonal = np.array([1, 2]) # Input shape: (2)
tf.matrix_diag(diagonal, k = -1, num_rows = 3, num_cols = 4)
==> [[0, 0, 0, 0], # Output shape: (3, 4)
[1, 0, 0, 0],
[0, 2, 0, 0]]
# Rectangular matrix with inferred num_cols and padding = 9.
tf.matrix_diag(diagonal, k = -1, num_rows = 3, padding = 9)
==> [[9, 9], # Output shape: (3, 2)
[1, 9],
[9, 2]]
```
Args:
diagonal: A `Tensor` with `rank k >= 1`.
name: A name for the operation (optional).
k: Diagonal offset(s). Positive value means superdiagonal, 0 refers to the
main diagonal, and negative value means subdiagonals. `k` can be a single
integer (for a single diagonal) or a pair of integers specifying the low
and high ends of a matrix band. `k[0]` must not be larger than `k[1]`.
num_rows: The number of rows of the output matrix. If it is not provided,
the op assumes the output matrix is a square matrix and infers the matrix
size from `d_lower`, `d_upper`, and the innermost dimension of `diagonal`.
num_cols: The number of columns of the output matrix. If it is not provided,
the op assumes the output matrix is a square matrix and infers the matrix
size from `d_lower`, `d_upper`, and the innermost dimension of `diagonal`.
padding_value: The value to fill the area outside the specified diagonal
band with. Default is 0.
Returns:
A Tensor. Has the same type as `diagonal`.
"""
if compat.forward_compatible(2019, 7, 4):
# Special case to sidestep the tf.constant conversion error:
# TypeError: Expected bool, got 0 of type 'int' instead.
if hasattr(diagonal, "dtype") and diagonal.dtype == "bool":
padding_value = bool(padding_value)
return gen_array_ops.matrix_diag_v2(
diagonal=diagonal,
k=k,
num_rows=num_rows,
num_cols=num_cols,
padding_value=padding_value,
name=name)
# Call v1 to maintain forward compatibility.
return gen_array_ops.matrix_diag(diagonal=diagonal, name=name)
@tf_export("linalg.diag_part", v1=["linalg.diag_part", "matrix_diag_part"])
@deprecation.deprecated_endpoints("matrix_diag_part")
def matrix_diag_part(
input, # pylint:disable=redefined-builtin
name="diag_part",
k=0,
padding_value=0):
"""Returns the batched diagonal part of a batched tensor.
Returns a tensor with the `k[0]`-th to `k[1]`-th diagonals of the batched
`input`.
Assume `input` has `r` dimensions `[I, J, ..., L, M, N]`.
Let `max_diag_len` be the maximum length among all diagonals to be extracted,
`max_diag_len = min(M + min(k[1], 0), N + min(-k[0], 0))`
Let `num_diags` be the number of diagonals to extract,
`num_diags = k[1] - k[0] + 1`.
If `num_diags == 1`, the output tensor is of rank `r - 1` with shape
`[I, J, ..., L, max_diag_len]` and values:
```
diagonal[i, j, ..., l, n]
= input[i, j, ..., l, n+y, n+x] ; when 0 <= n-y < M and 0 <= n-x < N,
0 ; otherwise.
```
where `y = max(-k[1], 0)`, `x = max(k[1], 0)`.
Otherwise, the output tensor has rank `r` with dimensions
`[I, J, ..., L, num_diags, max_diag_len]` with values:
```
diagonal[i, j, ..., l, m, n]
= input[i, j, ..., l, n+y, n+x] ; when 0 <= n-y < M and 0 <= n-x < N,
0 ; otherwise.
```
where `d = k[1] - m`, `y = max(-d, 0)`, and `x = max(d, 0)`.
The input must be at least a matrix.
For example:
```
input = np.array([[[1, 2, 3, 4], # Input shape: (2, 3, 4)
[5, 6, 7, 8],
[9, 8, 7, 6]],
[[5, 4, 3, 2],
[1, 2, 3, 4],
[5, 6, 7, 8]]])
# A main diagonal from each batch.
tf.matrix_diag_part(input) ==> [[1, 6, 7], # Output shape: (2, 3)
[5, 2, 7]]
# A superdiagonal from each batch.
tf.matrix_diag_part(input, k = 1)
==> [[2, 7, 6], # Output shape: (2, 3)
[4, 3, 8]]
# A tridiagonal band from each batch.
tf.matrix_diag_part(input, k = (-1, 1))
==> [[[2, 7, 6], # Output shape: (2, 3, 3)
[1, 6, 7],
[5, 8, 0]],
[[4, 3, 8],
[5, 2, 7],
[1, 6, 0]]]
# Padding = 9
tf.matrix_diag_part(input, k = (1, 3), padding = 9)
==> [[[4, 9, 9], # Output shape: (2, 3, 3)
[3, 8, 9],
[2, 7, 6]],
[[2, 9, 9],
[3, 4, 9],
[4, 3, 8]]]
```
Args:
input: A `Tensor` with `rank k >= 2`.
name: A name for the operation (optional).
k: Diagonal offset(s). Positive value means superdiagonal, 0 refers to the
main diagonal, and negative value means subdiagonals. `k` can be a single
integer (for a single diagonal) or a pair of integers specifying the low
and high ends of a matrix band. `k[0]` must not be larger than `k[1]`.
padding_value: The value to fill the area outside the specified diagonal
band with. Default is 0.
Returns:
A Tensor containing diagonals of `input`. Has the same type as `input`.
"""
if compat.forward_compatible(2019, 7, 4):
# Special case to sidestep the tf.constant conversion error:
# TypeError: Expected bool, got 0 of type 'int' instead.
if hasattr(input, "dtype") and input.dtype == "bool":
padding_value = bool(padding_value)
return gen_array_ops.matrix_diag_part_v2(
input=input, k=k, padding_value=padding_value, name=name)
# Call v1 to maintain forward compatibility.
return gen_array_ops.matrix_diag_part(input=input, name=name)
@tf_export("linalg.set_diag", v1=["linalg.set_diag", "matrix_set_diag"])
@deprecation.deprecated_endpoints("matrix_set_diag")
def matrix_set_diag(
input, # pylint:disable=redefined-builtin
diagonal,
name="set_diag",
k=0):
"""Returns a batched matrix tensor with new batched diagonal values.
Given `input` and `diagonal`, this operation returns a tensor with the
same shape and values as `input`, except for the specified diagonals of the
innermost matrices. These will be overwritten by the values in `diagonal`.
`input` has `r+1` dimensions `[I, J, ..., L, M, N]`. When `k` is scalar or
`k[0] == k[1]`, `diagonal` has `r` dimensions `[I, J, ..., L, max_diag_len]`.
Otherwise, it has `r+1` dimensions `[I, J, ..., L, num_diags, max_diag_len]`.
`num_diags` is the number of diagonals, `num_diags = k[1] - k[0] + 1`.
`max_diag_len` is the longest diagonal in the range `[k[0], k[1]]`,
`max_diag_len = min(M + min(k[1], 0), N + min(-k[0], 0))`
The output is a tensor of rank `k+1` with dimensions `[I, J, ..., L, M, N]`.
If `k` is scalar or `k[0] == k[1]`:
```
output[i, j, ..., l, m, n]
= diagonal[i, j, ..., l, n-max(k[1], 0)] ; if n - m == k[1]
output[i, j, ..., l, m, n] ; otherwise
```
Otherwise,
```
output[i, j, ..., l, m, n]
= diagonal[i, j, ..., l, k[1]-d, n-max(d, 0)] ; if d_lower <= d <= d_upper
input[i, j, ..., l, m, n] ; otherwise
```
where `d = n - m`
For example:
```
# The main diagonal.
input = np.array([[[7, 7, 7, 7], # Input shape: (2, 3, 4)
[7, 7, 7, 7],
[7, 7, 7, 7]],
[[7, 7, 7, 7],
[7, 7, 7, 7],
[7, 7, 7, 7]]])
diagonal = np.array([[1, 2, 3], # Diagonal shape: (2, 3)
[4, 5, 6]])
tf.matrix_diag(diagonal) ==> [[[1, 7, 7, 7], # Output shape: (2, 3, 4)
[7, 2, 7, 7],
[7, 7, 3, 7]],
[[4, 7, 7, 7],
[7, 5, 7, 7],
[7, 7, 6, 7]]]
# A superdiagonal (per batch).
tf.matrix_diag(diagonal, k = 1)
==> [[[7, 1, 7, 7], # Output shape: (2, 3, 4)
[7, 7, 2, 7],
[7, 7, 7, 3]],
[[7, 4, 7, 7],
[7, 7, 5, 7],
[7, 7, 7, 6]]]
# A band of diagonals.
diagonals = np.array([[[1, 2, 3], # Diagonal shape: (2, 2, 3)
[4, 5, 0]],
[[6, 1, 2],
[3, 4, 0]]])
tf.matrix_diag(diagonals, k = (-1, 0))
==> [[[1, 7, 7, 7], # Output shape: (2, 3, 4)
[4, 2, 7, 7],
[0, 5, 3, 7]],
[[6, 7, 7, 7],
[3, 1, 7, 7],
[7, 4, 2, 7]]]
```
Args:
input: A `Tensor` with rank `k + 1`, where `k >= 1`.
diagonal: A `Tensor` with rank `k`, when `d_lower == d_upper`, or `k + 1`,
otherwise. `k >= 1`.
name: A name for the operation (optional).
k: Diagonal offset(s). Positive value means superdiagonal, 0 refers to the
main diagonal, and negative value means subdiagonals. `k` can be a single
integer (for a single diagonal) or a pair of integers specifying the low
and high ends of a matrix band. `k[0]` must not be larger than `k[1]`.
"""
if compat.forward_compatible(2019, 7, 4):
return gen_array_ops.matrix_set_diag_v2(
input=input, diagonal=diagonal, k=k, name=name)
# Call v1 to maintain forward compatibility.
return gen_array_ops.matrix_set_diag(
input=input, diagonal=diagonal, name=name)
# pylint: enable=invalid-name
def _constant_if_small(value, shape, dtype, name):
try:
if np.prod(shape) < 1000:
return constant(value, shape=shape, dtype=dtype, name=name)
except TypeError:
# Happens when shape is a Tensor, list with Tensor elements, etc.
pass
return None
@tf_export("zeros")
def zeros(shape, dtype=dtypes.float32, name=None):
"""Creates a tensor with all elements set to zero.
This operation returns a tensor of type `dtype` with shape `shape` and
all elements set to zero.
For example:
```python
tf.zeros([3, 4], tf.int32) # [[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]]
```
Args:
shape: A list of integers, a tuple of integers, or a 1-D `Tensor` of type
`int32`.
dtype: The type of an element in the resulting `Tensor`.
name: A name for the operation (optional).
Returns:
A `Tensor` with all elements set to zero.
"""
dtype = dtypes.as_dtype(dtype).base_dtype
with ops.name_scope(name, "zeros", [shape]) as name:
if dtype == dtypes.bool:
zero = False
elif dtype == dtypes.string:
zero = ""
else:
zero = 0
if not isinstance(shape, ops.Tensor):
try:
# Create a constant if it won't be very big. Otherwise create a fill op
# to prevent serialized GraphDefs from becoming too large.
output = _constant_if_small(zero, shape, dtype, name)
if output is not None:
return output
# Go through tensor shapes to get int64-if-needed semantics
shape = constant_op._tensor_shape_tensor_conversion_function(
tensor_shape.TensorShape(shape))
except (TypeError, ValueError):
# Happens when shape is a list with tensor elements
shape = ops.convert_to_tensor(shape, dtype=dtypes.int32)
if not shape._shape_tuple():
shape = reshape(shape, [-1]) # Ensure it's a vector
output = fill(shape, constant(zero, dtype=dtype), name=name)
assert output.dtype.base_dtype == dtype
return output
@tf_export(v1=["zeros_like"])
@dispatch.add_dispatch_support
def zeros_like(tensor, dtype=None, name=None, optimize=True):
"""Creates a tensor with all elements set to zero.
Given a single tensor (`tensor`), this operation returns a tensor of the
same type and shape as `tensor` with all elements set to zero. Optionally,
you can use `dtype` to specify a new type for the returned tensor.
For example:
```python
tensor = tf.constant([[1, 2, 3], [4, 5, 6]])
tf.zeros_like(tensor) # [[0, 0, 0], [0, 0, 0]]
```
Args:
tensor: A `Tensor`.
dtype: A type for the returned `Tensor`. Must be `float16`, `float32`,
`float64`, `int8`, `uint8`, `int16`, `uint16`, `int32`, `int64`,
`complex64`, `complex128`, `bool` or `string`.
name: A name for the operation (optional).
optimize: if true, attempt to statically determine the shape of 'tensor' and
encode it as a constant.
Returns:
A `Tensor` with all elements set to zero.
"""
return zeros_like_impl(tensor, dtype, name, optimize)
@tf_export("zeros_like", v1=[])
@dispatch.add_dispatch_support
def zeros_like_v2(
input, # pylint: disable=redefined-builtin
dtype=None,
name=None):
"""Creates a tensor with all elements set to zero.
Given a single tensor (`tensor`), this operation returns a tensor of the
same type and shape as `tensor` with all elements set to zero. Optionally,
you can use `dtype` to specify a new type for the returned tensor.
For example:
```python
tensor = tf.constant([[1, 2, 3], [4, 5, 6]])
tf.zeros_like(tensor) # [[0, 0, 0], [0, 0, 0]]
```
Args:
input: A `Tensor`.
dtype: A type for the returned `Tensor`. Must be `float16`, `float32`,
`float64`, `int8`, `uint8`, `int16`, `uint16`, `int32`, `int64`,
`complex64`, `complex128`, `bool` or `string`.
name: A name for the operation (optional).
Returns:
A `Tensor` with all elements set to zero.
"""
return zeros_like_impl(input, dtype, name, optimize=True)
def zeros_like_impl(tensor, dtype, name, optimize=True):
"""Internal implementation for the v1/v2 zeros_like API calls."""
with ops.name_scope(name, "zeros_like", [tensor]) as name:
tensor = ops.convert_to_tensor(tensor, name="tensor")
if context.executing_eagerly():
if dtype is not None and dtype != tensor.dtype:
return zeros(
shape_internal(tensor, optimize=optimize), dtype=dtype, name=name)
with ops.device(tensor.device):
return gen_array_ops.zeros_like(tensor, name=name)
# For now, variant types must be created via zeros_like; as we need to
# pass the input variant object to the proper zeros callback.
if (optimize and tensor.shape.is_fully_defined() and
tensor.dtype != dtypes.variant):
# We can produce a zeros tensor independent of the value of 'tensor',
# since the shape is known statically.
return zeros(tensor.shape, dtype=dtype or tensor.dtype, name=name)
if dtype is not None and dtype != tensor.dtype and dtype != dtypes.variant:
return zeros(
shape_internal(tensor, optimize=optimize), dtype=dtype, name=name)
else:
return gen_array_ops.zeros_like(tensor, name=name)
@tf_export(v1=["ones_like"])
@dispatch.add_dispatch_support
def ones_like(tensor, dtype=None, name=None, optimize=True):
"""Creates a tensor with all elements set to 1.
Given a single tensor (`tensor`), this operation returns a tensor of the same
type and shape as `tensor` with all elements set to 1. Optionally, you can
specify a new type (`dtype`) for the returned tensor.
For example:
```python
tensor = tf.constant([[1, 2, 3], [4, 5, 6]])
tf.ones_like(tensor) # [[1, 1, 1], [1, 1, 1]]
```
Args:
tensor: A `Tensor`.
dtype: A type for the returned `Tensor`. Must be `float32`, `float64`,
`int8`, `uint8`, `int16`, `uint16`, `int32`, `int64`, `complex64`,
`complex128` or `bool`.
name: A name for the operation (optional).
optimize: if true, attempt to statically determine the shape of 'tensor' and
encode it as a constant.
Returns:
A `Tensor` with all elements set to 1.
"""
return ones_like_impl(tensor, dtype, name, optimize)
@tf_export("ones_like", v1=[])
@dispatch.add_dispatch_support
def ones_like_v2(
input, # pylint: disable=redefined-builtin
dtype=None,
name=None):
"""Creates a tensor with all elements set to zero.
Given a single tensor (`tensor`), this operation returns a tensor of the
same type and shape as `tensor` with all elements set to 1. Optionally,
you can use `dtype` to specify a new type for the returned tensor.
For example:
```python
tensor = tf.constant([[1, 2, 3], [4, 5, 6]])
tf.ones_like(tensor) # [[1, 1, 1], [1, 1, 1]]
```
Args:
input: A `Tensor`.
dtype: A type for the returned `Tensor`. Must be `float16`, `float32`,
`float64`, `int8`, `uint8`, `int16`, `uint16`, `int32`, `int64`,
`complex64`, `complex128`, `bool` or `string`.
name: A name for the operation (optional).
Returns:
A `Tensor` with all elements set to zero.
"""
return ones_like_impl(input, dtype, name, optimize=True)
def ones_like_impl(tensor, dtype, name, optimize=True):
"""Internal implementation for the v1/v2 ones_like API calls."""
with ops.name_scope(name, "ones_like", [tensor]) as name:
tensor = ops.convert_to_tensor(tensor, name="tensor")
ones_shape = shape_internal(tensor, optimize=optimize)
if dtype is None:
dtype = tensor.dtype
ret = ones(ones_shape, dtype=dtype, name=name)
if not context.executing_eagerly():
ret.set_shape(tensor.get_shape())
return ret
@tf_export("ones")
def ones(shape, dtype=dtypes.float32, name=None):
"""Creates a tensor with all elements set to 1.
This operation returns a tensor of type `dtype` with shape `shape` and all
elements set to 1.
For example:
```python
tf.ones([2, 3], tf.int32) # [[1, 1, 1], [1, 1, 1]]
```
Args:
shape: A list of integers, a tuple of integers, or a 1-D `Tensor` of type
`int32`.
dtype: The type of an element in the resulting `Tensor`.
name: A name for the operation (optional).
Returns:
A `Tensor` with all elements set to 1.
"""
dtype = dtypes.as_dtype(dtype).base_dtype
with ops.name_scope(name, "ones", [shape]) as name:
one = True if dtype == dtypes.bool else 1
if not isinstance(shape, ops.Tensor):
try:
# Create a constant if it won't be very big. Otherwise create a fill op
# to prevent serialized GraphDefs from becoming too large.
output = _constant_if_small(one, shape, dtype, name)
if output is not None:
return output
# Go through tensor shapes to get int64-if-needed semantics
shape = constant_op._tensor_shape_tensor_conversion_function(
tensor_shape.TensorShape(shape))
except (TypeError, ValueError):
# Happens when shape is a list with tensor elements
shape = ops.convert_to_tensor(shape, dtype=dtypes.int32)
if not shape._shape_tuple():
shape = reshape(shape, [-1]) # Ensure it's a vector
output = fill(shape, constant(one, dtype=dtype), name=name)
assert output.dtype.base_dtype == dtype
return output
@tf_export(v1=["placeholder"])
def placeholder(dtype, shape=None, name=None):
"""Inserts a placeholder for a tensor that will be always fed.
**Important**: This tensor will produce an error if evaluated. Its value must
be fed using the `feed_dict` optional argument to `Session.run()`,
`Tensor.eval()`, or `Operation.run()`.
For example:
```python
x = tf.compat.v1.placeholder(tf.float32, shape=(1024, 1024))
y = tf.matmul(x, x)
with tf.compat.v1.Session() as sess:
print(sess.run(y)) # ERROR: will fail because x was not fed.
rand_array = np.random.rand(1024, 1024)
print(sess.run(y, feed_dict={x: rand_array})) # Will succeed.
```
@compatibility(eager)
Placeholders are not compatible with eager execution.
@end_compatibility
Args:
dtype: The type of elements in the tensor to be fed.
shape: The shape of the tensor to be fed (optional). If the shape is not
specified, you can feed a tensor of any shape.
name: A name for the operation (optional).
Returns:
A `Tensor` that may be used as a handle for feeding a value, but not
evaluated directly.
Raises:
RuntimeError: if eager execution is enabled
"""
if context.executing_eagerly():
raise RuntimeError("tf.placeholder() is not compatible with "
"eager execution.")
return gen_array_ops.placeholder(dtype=dtype, shape=shape, name=name)
@tf_export(v1=["placeholder_with_default"])
def placeholder_with_default(input, shape, name=None): # pylint: disable=redefined-builtin
"""A placeholder op that passes through `input` when its output is not fed.
Args:
input: A `Tensor`. The default value to produce when output is not fed.
shape: A `tf.TensorShape` or list of `int`s. The (possibly partial) shape of
the tensor.
name: A name for the operation (optional).
Returns:
A `Tensor`. Has the same type as `input`.
"""
return gen_array_ops.placeholder_with_default(input, shape, name)
# pylint: disable=redefined-outer-name
def _normalize_sparse_shape(shape, name):
"""Returns a tuple of (Tensor or None, rank or None)."""
if shape is None:
return (None, None)
rank = shape.get_shape()[0] if isinstance(shape, ops.Tensor) else len(shape)
if not isinstance(shape, ops.Tensor) and None in shape:
return (None, rank)
return (ops.convert_to_tensor(shape, dtype=dtypes.int64, name=name), rank)
@tf_export(v1=["sparse.placeholder", "sparse_placeholder"])
@deprecation.deprecated_endpoints("sparse_placeholder")
def sparse_placeholder(dtype, shape=None, name=None):
"""Inserts a placeholder for a sparse tensor that will be always fed.
**Important**: This sparse tensor will produce an error if evaluated.
Its value must be fed using the `feed_dict` optional argument to
`Session.run()`, `Tensor.eval()`, or `Operation.run()`.
For example:
```python
x = tf.compat.v1.sparse.placeholder(tf.float32)
y = tf.sparse.reduce_sum(x)
with tf.compat.v1.Session() as sess:
print(sess.run(y)) # ERROR: will fail because x was not fed.
indices = np.array([[3, 2, 0], [4, 5, 1]], dtype=np.int64)
values = np.array([1.0, 2.0], dtype=np.float32)
shape = np.array([7, 9, 2], dtype=np.int64)
print(sess.run(y, feed_dict={
x: tf.compat.v1.SparseTensorValue(indices, values, shape)})) # Will
succeed.
print(sess.run(y, feed_dict={
x: (indices, values, shape)})) # Will succeed.
sp = tf.SparseTensor(indices=indices, values=values, dense_shape=shape)
sp_value = sp.eval(session=sess)
print(sess.run(y, feed_dict={x: sp_value})) # Will succeed.
```
@compatibility{eager} Placeholders are not compatible with eager execution.
Args:
dtype: The type of `values` elements in the tensor to be fed.
shape: The shape of the tensor to be fed (optional). If the shape is not
specified, you can feed a sparse tensor of any shape.
name: A name for prefixing the operations (optional).
Returns:
A `SparseTensor` that may be used as a handle for feeding a value, but not
evaluated directly.
Raises:
RuntimeError: if eager execution is enabled
"""
if context.executing_eagerly():
raise RuntimeError("tf.placeholder() is not compatible with "
"eager execution.")
shape_name = (name + "/shape") if name is not None else None
shape, rank = _normalize_sparse_shape(shape, shape_name)
if shape is None:
shape = placeholder(dtypes.int64, shape=[rank], name=shape_name)
return sparse_tensor.SparseTensor(
values=placeholder(
dtype,
shape=[None],
name=(name + "/values") if name is not None else None),
indices=placeholder(
dtypes.int64,
shape=[None, rank],
name=(name + "/indices") if name is not None else None),
dense_shape=shape)
# pylint: enable=redefined-outer-name
@tf_export("pad", v1=[])
def pad_v2(tensor, paddings, mode="CONSTANT", constant_values=0, name=None):
"""Pads a tensor.
This operation pads a `tensor` according to the `paddings` you specify.
`paddings` is an integer tensor with shape `[n, 2]`, where n is the rank of
`tensor`. For each dimension D of `input`, `paddings[D, 0]` indicates how
many values to add before the contents of `tensor` in that dimension, and
`paddings[D, 1]` indicates how many values to add after the contents of
`tensor` in that dimension. If `mode` is "REFLECT" then both `paddings[D, 0]`
and `paddings[D, 1]` must be no greater than `tensor.dim_size(D) - 1`. If
`mode` is "SYMMETRIC" then both `paddings[D, 0]` and `paddings[D, 1]` must be
no greater than `tensor.dim_size(D)`.
The padded size of each dimension D of the output is:
`paddings[D, 0] + tensor.dim_size(D) + paddings[D, 1]`
For example:
```python
t = tf.constant([[1, 2, 3], [4, 5, 6]])
paddings = tf.constant([[1, 1,], [2, 2]])
# 'constant_values' is 0.
# rank of 't' is 2.
tf.pad(t, paddings, "CONSTANT") # [[0, 0, 0, 0, 0, 0, 0],
# [0, 0, 1, 2, 3, 0, 0],
# [0, 0, 4, 5, 6, 0, 0],
# [0, 0, 0, 0, 0, 0, 0]]
tf.pad(t, paddings, "REFLECT") # [[6, 5, 4, 5, 6, 5, 4],
# [3, 2, 1, 2, 3, 2, 1],
# [6, 5, 4, 5, 6, 5, 4],
# [3, 2, 1, 2, 3, 2, 1]]
tf.pad(t, paddings, "SYMMETRIC") # [[2, 1, 1, 2, 3, 3, 2],
# [2, 1, 1, 2, 3, 3, 2],
# [5, 4, 4, 5, 6, 6, 5],
# [5, 4, 4, 5, 6, 6, 5]]
```
Args:
tensor: A `Tensor`.
paddings: A `Tensor` of type `int32`.
mode: One of "CONSTANT", "REFLECT", or "SYMMETRIC" (case-insensitive)
constant_values: In "CONSTANT" mode, the scalar pad value to use. Must be
same type as `tensor`.
name: A name for the operation (optional).
Returns:
A `Tensor`. Has the same type as `tensor`.
Raises:
ValueError: When mode is not one of "CONSTANT", "REFLECT", or "SYMMETRIC".
"""
return pad(tensor, paddings, mode, name, constant_values)
@tf_export(v1=["pad"])
def pad(tensor, paddings, mode="CONSTANT", name=None, constant_values=0): # pylint: disable=invalid-name
"""Pads a tensor.
This operation pads a `tensor` according to the `paddings` you specify.
`paddings` is an integer tensor with shape `[n, 2]`, where n is the rank of
`tensor`. For each dimension D of `input`, `paddings[D, 0]` indicates how
many values to add before the contents of `tensor` in that dimension, and
`paddings[D, 1]` indicates how many values to add after the contents of
`tensor` in that dimension. If `mode` is "REFLECT" then both `paddings[D, 0]`
and `paddings[D, 1]` must be no greater than `tensor.dim_size(D) - 1`. If
`mode` is "SYMMETRIC" then both `paddings[D, 0]` and `paddings[D, 1]` must be
no greater than `tensor.dim_size(D)`.
The padded size of each dimension D of the output is:
`paddings[D, 0] + tensor.dim_size(D) + paddings[D, 1]`
For example:
```python
t = tf.constant([[1, 2, 3], [4, 5, 6]])
paddings = tf.constant([[1, 1,], [2, 2]])
# 'constant_values' is 0.
# rank of 't' is 2.
tf.pad(t, paddings, "CONSTANT") # [[0, 0, 0, 0, 0, 0, 0],
# [0, 0, 1, 2, 3, 0, 0],
# [0, 0, 4, 5, 6, 0, 0],
# [0, 0, 0, 0, 0, 0, 0]]
tf.pad(t, paddings, "REFLECT") # [[6, 5, 4, 5, 6, 5, 4],
# [3, 2, 1, 2, 3, 2, 1],
# [6, 5, 4, 5, 6, 5, 4],
# [3, 2, 1, 2, 3, 2, 1]]
tf.pad(t, paddings, "SYMMETRIC") # [[2, 1, 1, 2, 3, 3, 2],
# [2, 1, 1, 2, 3, 3, 2],
# [5, 4, 4, 5, 6, 6, 5],
# [5, 4, 4, 5, 6, 6, 5]]
```
Args:
tensor: A `Tensor`.
paddings: A `Tensor` of type `int32`.
mode: One of "CONSTANT", "REFLECT", or "SYMMETRIC" (case-insensitive)
name: A name for the operation (optional).
constant_values: In "CONSTANT" mode, the scalar pad value to use. Must be
same type as `tensor`.
Returns:
A `Tensor`. Has the same type as `tensor`.
Raises:
ValueError: When mode is not one of "CONSTANT", "REFLECT", or "SYMMETRIC".
"""
# Convert lower/mixed case to upper for NumPy compatibility
# NumPy uses all lower-case modes.
mode = mode.upper()
if mode == "CONSTANT":
# TODO(rjryan): Once the forward compatibility period (3 weeks) have passed
# remove the "Pad" fallback here.
if constant_values != 0:
result = gen_array_ops.pad_v2(
tensor, paddings, constant_values, name=name)
else:
result = gen_array_ops.pad(tensor, paddings, name=name)
elif mode == "REFLECT":
result = gen_array_ops.mirror_pad(
tensor, paddings, mode="REFLECT", name=name)
elif mode == "SYMMETRIC":
result = gen_array_ops.mirror_pad(
tensor, paddings, mode="SYMMETRIC", name=name)
else:
raise ValueError("Unknown padding mode: %s" % mode)
# Restore shape information where possible.
if not context.executing_eagerly():
paddings_constant = tensor_util.constant_value(
result.op.inputs[1], partial=True)
input_shape = result.op.inputs[0].shape
if (input_shape.ndims is not None and
not result.shape.is_fully_defined() and paddings_constant is not None):
new_shape = []
for padding, dim in zip(paddings_constant, input_shape.as_list()):
if padding is None or dim is None or any((x is None for x in padding)):
new_shape.append(None)
else:
new_shape.append(sum(padding) + dim)
result.set_shape(new_shape)
return result
@tf_export("meshgrid")
def meshgrid(*args, **kwargs):
"""Broadcasts parameters for evaluation on an N-D grid.
Given N one-dimensional coordinate arrays `*args`, returns a list `outputs`
of N-D coordinate arrays for evaluating expressions on an N-D grid.
Notes:
`meshgrid` supports cartesian ('xy') and matrix ('ij') indexing conventions.
When the `indexing` argument is set to 'xy' (the default), the broadcasting
instructions for the first two dimensions are swapped.
Examples:
Calling `X, Y = meshgrid(x, y)` with the tensors
```python
x = [1, 2, 3]
y = [4, 5, 6]
X, Y = tf.meshgrid(x, y)
# X = [[1, 2, 3],
# [1, 2, 3],
# [1, 2, 3]]
# Y = [[4, 4, 4],
# [5, 5, 5],
# [6, 6, 6]]
```
Args:
*args: `Tensor`s with rank 1.
**kwargs:
- indexing: Either 'xy' or 'ij' (optional, default: 'xy').
- name: A name for the operation (optional).
Returns:
outputs: A list of N `Tensor`s with rank N.
Raises:
TypeError: When no keyword arguments (kwargs) are passed.
ValueError: When indexing keyword argument is not one of `xy` or `ij`.
"""
indexing = kwargs.pop("indexing", "xy")
name = kwargs.pop("name", "meshgrid")
if kwargs:
key = list(kwargs.keys())[0]
raise TypeError("'{}' is an invalid keyword argument "
"for this function".format(key))
if indexing not in ("xy", "ij"):
raise ValueError("indexing parameter must be either 'xy' or 'ij'")
with ops.name_scope(name, "meshgrid", args) as name:
ndim = len(args)
s0 = (1,) * ndim
# Prepare reshape by inserting dimensions with size 1 where needed
output = []
for i, x in enumerate(args):
output.append(reshape(stack(x), (s0[:i] + (-1,) + s0[i + 1::])))
# Create parameters for broadcasting each tensor to the full size
shapes = [size(x) for x in args]
output_dtype = ops.convert_to_tensor(args[0]).dtype.base_dtype
if indexing == "xy" and ndim > 1:
output[0] = reshape(output[0], (1, -1) + (1,) * (ndim - 2))
output[1] = reshape(output[1], (-1, 1) + (1,) * (ndim - 2))
shapes[0], shapes[1] = shapes[1], shapes[0]
# TODO(nolivia): improve performance with a broadcast
mult_fact = ones(shapes, output_dtype)
return [x * mult_fact for x in output]
NEW_AXIS = -1
SHRINK_AXIS = -2
# PEP-8 naming
# pylint: disable=invalid-name,redefined-outer-name
def _compute_size_of_strided_dim(shrink, spec, size):
"""Computes the size of a single strided slice dimension."""
unknown = None # Document what None means here.
use_full_range = None # Document other use of None.
# if this is a shrink axis (i.e. a non-range index)
# it either will produce an error or return 1
if shrink:
return 1
if size is unknown or size.value is unknown:
return unknown
size = size.value
stride = spec.step
if stride is not unknown:
if stride == 0:
return unknown
stride = spec.step
valid_range = [0, size] if stride > 0 else [-1, size - 1]
# PEP-8 naming
# pylint: disable=invalid-name
def canonical(x, c):
if x is use_full_range:
return valid_range[c] if stride > 0 else valid_range[(c + 1) & 1]
else:
x_fwd = size + x if x < 0 else x # make negative indices positive
return max(valid_range[0], min(valid_range[1], x_fwd))
begin = canonical(spec.start, 0)
end = canonical(spec.stop, 1)
interval_length = end - begin
if interval_length == 0 or ((interval_length < 0) != (stride < 0)):
return 0
else:
remainder = 1 if interval_length % stride != 0 else 0
return interval_length // stride + remainder
else:
return unknown # unknown because stride is unknown
def _TileGradShape(op):
"""Shape function for the TileGrad op."""
multiples_shape = op.inputs[1].get_shape().with_rank(1)
input_shape = op.inputs[0].get_shape().with_rank(multiples_shape[0])
# NOTE(mrry): Represent `multiples` as a `TensorShape` because (i)
# it is a vector of non-negative integers, and (ii) doing so allows
# us to handle partially-known multiples.
multiples = tensor_util.constant_value_as_shape(op.inputs[1]).with_rank(
input_shape.ndims)
if multiples.ndims is None:
return [tensor_shape.unknown_shape()]
else:
output_dims = []
for dim, multiple in zip(input_shape.dims, multiples.dims):
output_dims.append(dim // multiple)
return [tensor_shape.TensorShape(output_dims)]
@tf_export("edit_distance")
def edit_distance(hypothesis, truth, normalize=True, name="edit_distance"):
"""Computes the Levenshtein distance between sequences.
This operation takes variable-length sequences (`hypothesis` and `truth`),
each provided as a `SparseTensor`, and computes the Levenshtein distance.
You can normalize the edit distance by length of `truth` by setting
`normalize` to true.
For example, given the following input:
```python
# 'hypothesis' is a tensor of shape `[2, 1]` with variable-length values:
# (0,0) = ["a"]
# (1,0) = ["b"]
hypothesis = tf.SparseTensor(
[[0, 0, 0],
[1, 0, 0]],
["a", "b"],
(2, 1, 1))
# 'truth' is a tensor of shape `[2, 2]` with variable-length values:
# (0,0) = []
# (0,1) = ["a"]
# (1,0) = ["b", "c"]
# (1,1) = ["a"]
truth = tf.SparseTensor(
[[0, 1, 0],
[1, 0, 0],
[1, 0, 1],
[1, 1, 0]],
["a", "b", "c", "a"],
(2, 2, 2))
normalize = True
```
This operation would return the following:
```python
# 'output' is a tensor of shape `[2, 2]` with edit distances normalized
# by 'truth' lengths.
output ==> [[inf, 1.0], # (0,0): no truth, (0,1): no hypothesis
[0.5, 1.0]] # (1,0): addition, (1,1): no hypothesis
```
Args:
hypothesis: A `SparseTensor` containing hypothesis sequences.
truth: A `SparseTensor` containing truth sequences.
normalize: A `bool`. If `True`, normalizes the Levenshtein distance by
length of `truth.`
name: A name for the operation (optional).
Returns:
A dense `Tensor` with rank `R - 1`, where R is the rank of the
`SparseTensor` inputs `hypothesis` and `truth`.
Raises:
TypeError: If either `hypothesis` or `truth` are not a `SparseTensor`.
"""
if not isinstance(
hypothesis,
(sparse_tensor.SparseTensor, sparse_tensor.SparseTensorValue)):
raise TypeError("Hypothesis must be a SparseTensor.")
if not isinstance(
truth, (sparse_tensor.SparseTensor, sparse_tensor.SparseTensorValue)):
raise TypeError("Truth must be a SparseTensor.")
return gen_array_ops.edit_distance(
hypothesis.indices,
hypothesis.values,
hypothesis.dense_shape,
truth.indices,
truth.values,
truth.dense_shape,
normalize=normalize,
name=name)
@ops.RegisterGradient("FakeQuantWithMinMaxArgs")
def _FakeQuantWithMinMaxArgsGradient(op, grad):
"""Gradient for FakeQuantWithMinMaxArgs op."""
return fake_quant_with_min_max_args_gradient(
grad,
op.inputs[0],
min=op.get_attr("min"),
max=op.get_attr("max"),
num_bits=op.get_attr("num_bits"),
narrow_range=op.get_attr("narrow_range"))
@ops.RegisterGradient("FakeQuantWithMinMaxVars")
def _FakeQuantWithMinMaxVarsGradient(op, grad):
"""Gradient for FakeQuantWithMinMaxVars op."""
return fake_quant_with_min_max_vars_gradient(
grad,
op.inputs[0],
op.inputs[1],
op.inputs[2],
num_bits=op.get_attr("num_bits"),
narrow_range=op.get_attr("narrow_range"))
@ops.RegisterGradient("FakeQuantWithMinMaxVarsPerChannel")
def _FakeQuantWithMinMaxVarsPerChannelGradient(op, grad):
"""Gradient for FakeQuantWithMinMaxVarsPerChannel op."""
return fake_quant_with_min_max_vars_per_channel_gradient(
grad,
op.inputs[0],
op.inputs[1],
op.inputs[2],
num_bits=op.get_attr("num_bits"),
narrow_range=op.get_attr("narrow_range"))
@tf_export("required_space_to_batch_paddings")
def required_space_to_batch_paddings(input_shape,
block_shape,
base_paddings=None,
name=None):
"""Calculate padding required to make block_shape divide input_shape.
This function can be used to calculate a suitable paddings argument for use
with space_to_batch_nd and batch_to_space_nd.
Args:
input_shape: int32 Tensor of shape [N].
block_shape: int32 Tensor of shape [N].
base_paddings: Optional int32 Tensor of shape [N, 2]. Specifies the minimum
amount of padding to use. All elements must be >= 0. If not specified,
defaults to 0.
name: string. Optional name prefix.
Returns:
(paddings, crops), where:
`paddings` and `crops` are int32 Tensors of rank 2 and shape [N, 2]
satisfying:
paddings[i, 0] = base_paddings[i, 0].
0 <= paddings[i, 1] - base_paddings[i, 1] < block_shape[i]
(input_shape[i] + paddings[i, 0] + paddings[i, 1]) % block_shape[i] == 0
crops[i, 0] = 0
crops[i, 1] = paddings[i, 1] - base_paddings[i, 1]
Raises: ValueError if called with incompatible shapes.
"""
with ops.name_scope(name, "required_space_to_batch_paddings",
[input_shape, block_shape]):
input_shape = ops.convert_to_tensor(
input_shape, dtype=dtypes.int32, name="input_shape")
block_shape = ops.convert_to_tensor(
block_shape, dtype=dtypes.int32, name="block_shape")
block_shape.get_shape().assert_is_fully_defined()
block_shape.get_shape().assert_has_rank(1)
num_block_dims = block_shape.get_shape().dims[0].value
if num_block_dims == 0:
return zeros([0, 2], dtypes.int32), zeros([0, 2], dtypes.int32)
input_shape.get_shape().assert_is_compatible_with([num_block_dims])
if base_paddings is not None:
base_paddings = ops.convert_to_tensor(
base_paddings, dtype=dtypes.int32, name="base_paddings")
base_paddings.get_shape().assert_is_compatible_with([num_block_dims, 2])
else:
base_paddings = zeros([num_block_dims, 2], dtypes.int32)
const_block_shape = tensor_util.constant_value(block_shape)
const_input_shape = tensor_util.constant_value(input_shape)
const_base_paddings = tensor_util.constant_value(base_paddings)
if (const_block_shape is not None and const_input_shape is not None and
const_base_paddings is not None):
block_shape = const_block_shape
input_shape = const_input_shape
base_paddings = const_base_paddings
# Use same expression for both constant and non-constant case.
pad_start = base_paddings[:, 0]
orig_pad_end = base_paddings[:, 1]
full_input_shape = input_shape + pad_start + orig_pad_end
pad_end_extra = (block_shape - full_input_shape % block_shape) % block_shape
pad_end = orig_pad_end + pad_end_extra
result_paddings = stack(
[[pad_start[i], pad_end[i]] for i in range(num_block_dims)],
name="paddings")
result_crops = stack([[0, pad_end_extra[i]] for i in range(num_block_dims)],
name="crops")
return result_paddings, result_crops
@tf_export(v1=["nn.space_to_batch", "space_to_batch"])
@deprecation.deprecated_endpoints("space_to_batch")
def space_to_batch( # pylint: disable=missing-docstring
input, paddings, block_size=None, name=None, block_shape=None): # pylint: disable=redefined-builtin
block_size = deprecation.deprecated_argument_lookup("block_shape",
block_shape, "block_size",
block_size)
result = space_to_batch_nd(
input,
paddings=paddings,
block_shape=np.array([block_size, block_size], dtype=np.int64),
name=name)
result.set_shape(result.get_shape().with_rank(4))
return result
space_to_batch.__doc__ = gen_array_ops.space_to_batch.__doc__
@tf_export("space_to_batch", "nn.space_to_batch", v1=[])
def space_to_batch_v2(input, block_shape, paddings, name=None): # pylint: disable=redefined-builtin
return space_to_batch_nd(input, block_shape, paddings, name)
space_to_batch_v2.__doc__ = gen_array_ops.space_to_batch_nd.__doc__
@tf_export(v1=["nn.space_to_depth", "space_to_depth"])
@deprecation.deprecated_endpoints("space_to_depth")
def space_to_depth(input, block_size, name=None, data_format="NHWC"): # pylint: disable=redefined-builtin
return gen_array_ops.space_to_depth(input, block_size, data_format, name=name)
space_to_depth.__doc__ = gen_array_ops.space_to_depth.__doc__
@tf_export("nn.space_to_depth", v1=[])
def space_to_depth_v2(input, block_size, data_format="NHWC", name=None): # pylint: disable=redefined-builtin
return gen_array_ops.space_to_depth(input, block_size, data_format, name=name)
space_to_depth_v2.__doc__ = gen_array_ops.space_to_depth.__doc__
@tf_export(v1=["nn.depth_to_space", "depth_to_space"])
@deprecation.deprecated_endpoints("depth_to_space")
def depth_to_space(input, block_size, name=None, data_format="NHWC"): # pylint: disable=redefined-builtin
return gen_array_ops.depth_to_space(input, block_size, data_format, name=name)
depth_to_space.__doc__ = gen_array_ops.depth_to_space.__doc__
@tf_export("nn.depth_to_space", v1=[])
def depth_to_space_v2(input, block_size, data_format="NHWC", name=None): # pylint: disable=redefined-builtin
return gen_array_ops.depth_to_space(input, block_size, data_format, name=name)
depth_to_space_v2.__doc__ = gen_array_ops.depth_to_space.__doc__
@tf_export(v1=["batch_to_space"])
def batch_to_space(input, crops, block_size, name=None, block_shape=None): # pylint: disable=redefined-builtin,missing-docstring
block_size = deprecation.deprecated_argument_lookup("block_shape",
block_shape, "block_size",
block_size)
result = batch_to_space_nd(
input,
crops=crops,
block_shape=np.array([block_size, block_size], dtype=np.int64),
name=name)
result.set_shape(result.get_shape().with_rank(4))
return result
batch_to_space.__doc__ = gen_array_ops.batch_to_space.__doc__
@tf_export("batch_to_space", v1=[])
def batch_to_space_v2(input, block_shape, crops, name=None): # pylint: disable=redefined-builtin
"""BatchToSpace for N-D tensors of type T.
This operation reshapes the "batch" dimension 0 into `M + 1` dimensions of
shape `block_shape + [batch]`, interleaves these blocks back into the grid
defined by the spatial dimensions `[1, ..., M]`, to obtain a result with the
same rank as the input. The spatial dimensions of this intermediate result
are then optionally cropped according to `crops` to produce the output. This
is the reverse of SpaceToBatch. See below for a precise description.
Args:
input: A `Tensor`. N-D with shape `input_shape = [batch] + spatial_shape +
remaining_shape`, where spatial_shape has M dimensions.
block_shape: A `Tensor`. Must be one of the following types: `int32`,
`int64`. 1-D with shape `[M]`, all values must be >= 1. For backwards
compatibility with TF 1.0, this parameter may be an int, in which case it
is converted to `numpy.array([block_shape, block_shape],
dtype=numpy.int64)`.
crops: A `Tensor`. Must be one of the following types: `int32`, `int64`. 2-D
with shape `[M, 2]`, all values must be >= 0. `crops[i] = [crop_start,
crop_end]` specifies the amount to crop from input dimension `i + 1`,
which corresponds to spatial dimension `i`. It is required that
`crop_start[i] + crop_end[i] <= block_shape[i] * input_shape[i + 1]`.
This operation is equivalent to the following steps:
1. Reshape `input` to `reshaped` of shape: [block_shape[0], ...,
block_shape[M-1], batch / prod(block_shape), input_shape[1], ...,
input_shape[N-1]] 2. Permute dimensions of `reshaped` to produce
`permuted` of shape [batch / prod(block_shape), input_shape[1],
block_shape[0], ..., input_shape[M], block_shape[M-1],
input_shape[M+1], ..., input_shape[N-1]] 3. Reshape `permuted` to
produce `reshaped_permuted` of shape [batch / prod(block_shape),
input_shape[1] * block_shape[0], ..., input_shape[M] * block_shape[M-1],
input_shape[M+1], ..., input_shape[N-1]] 4. Crop the start and end of
dimensions `[1, ..., M]` of `reshaped_permuted` according to `crops` to
produce the
output of shape: [batch / prod(block_shape), input_shape[1] *
block_shape[0] - crops[0,0] - crops[0,1], ..., input_shape[M] *
block_shape[M-1] - crops[M-1,0] - crops[M-1,1], input_shape[M+1],
..., input_shape[N-1]]
Some examples: (1) For the following input of shape `[4, 1, 1, 1]`,
`block_shape = [2, 2]`, and `crops = [[0, 0], [0, 0]]`: ``` [[[[1]]],
[[[2]]], [[[3]]], [[[4]]]] ```
The output tensor has shape `[1, 2, 2, 1]` and value: ``` x = [[[[1],
[2]], [[3], [4]]]] ``` (2) For the following input of shape `[4, 1, 1,
3]`,
`block_shape = [2, 2]`, and `crops = [[0, 0], [0, 0]]`: ``` [[[1, 2,
3]], [[4, 5, 6]], [[7, 8, 9]], [[10, 11, 12]]] ```
The output tensor has shape `[1, 2, 2, 3]` and value: ``` x = [[[[1, 2,
3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]]] ``` (3) For the following
input of shape `[4, 2, 2, 1]`,
`block_shape = [2, 2]`, and `crops = [[0, 0], [0, 0]]`: ``` x =
[[[[1], [3]], [[9], [11]]], [[[2], [4]], [[10], [12]]], [[[5], [7]],
[[13], [15]]], [[[6], [8]], [[14], [16]]]] ```
The output tensor has shape `[1, 4, 4, 1]` and value: ``` x = [[[1],
[2], [3], [4]], [[5], [6], [7], [8]], [[9], [10], [11], [12]],
[[13], [14], [15], [16]]] ``` (4) For the following input of shape
`[8, 1, 3, 1]`,
`block_shape = [2, 2]`, and `crops = [[0, 0], [2, 0]]`: ``` x =
[[[[0], [1], [3]]], [[[0], [9], [11]]], [[[0], [2], [4]]], [[[0],
[10], [12]]], [[[0], [5], [7]]], [[[0], [13], [15]]], [[[0], [6],
[8]]], [[[0], [14], [16]]]] ```
The output tensor has shape `[2, 2, 4, 1]` and value: ``` x = [[[[1],
[2], [3], [4]], [[5], [6], [7], [8]]], [[[9], [10], [11], [12]],
[[13], [14], [15], [16]]]] ```
name: A name for the operation (optional).
Returns:
A `Tensor`. Has the same type as `input`.
"""
if isinstance(block_shape, int):
block_shape = np.array([block_shape, block_shape], dtype=np.int64)
return batch_to_space_nd(
input=input, block_shape=block_shape, crops=crops, name=name)
@tf_export("one_hot")
def one_hot(indices,
depth,
on_value=None,
off_value=None,
axis=None,
dtype=None,
name=None):
"""Returns a one-hot tensor.
The locations represented by indices in `indices` take value `on_value`,
while all other locations take value `off_value`.
`on_value` and `off_value` must have matching data types. If `dtype` is also
provided, they must be the same data type as specified by `dtype`.
If `on_value` is not provided, it will default to the value `1` with type
`dtype`
If `off_value` is not provided, it will default to the value `0` with type
`dtype`
If the input `indices` is rank `N`, the output will have rank `N+1`. The
new axis is created at dimension `axis` (default: the new axis is appended
at the end).
If `indices` is a scalar the output shape will be a vector of length `depth`
If `indices` is a vector of length `features`, the output shape will be:
```
features x depth if axis == -1
depth x features if axis == 0
```
If `indices` is a matrix (batch) with shape `[batch, features]`, the output
shape will be:
```
batch x features x depth if axis == -1
batch x depth x features if axis == 1
depth x batch x features if axis == 0
```
If `dtype` is not provided, it will attempt to assume the data type of
`on_value` or `off_value`, if one or both are passed in. If none of
`on_value`, `off_value`, or `dtype` are provided, `dtype` will default to the
value `tf.float32`.
Note: If a non-numeric data type output is desired (`tf.string`, `tf.bool`,
etc.), both `on_value` and `off_value` _must_ be provided to `one_hot`.
For example:
```python
indices = [0, 1, 2]
depth = 3
tf.one_hot(indices, depth) # output: [3 x 3]
# [[1., 0., 0.],
# [0., 1., 0.],
# [0., 0., 1.]]
indices = [0, 2, -1, 1]
depth = 3
tf.one_hot(indices, depth,
on_value=5.0, off_value=0.0,
axis=-1) # output: [4 x 3]
# [[5.0, 0.0, 0.0], # one_hot(0)
# [0.0, 0.0, 5.0], # one_hot(2)
# [0.0, 0.0, 0.0], # one_hot(-1)
# [0.0, 5.0, 0.0]] # one_hot(1)
indices = [[0, 2], [1, -1]]
depth = 3
tf.one_hot(indices, depth,
on_value=1.0, off_value=0.0,
axis=-1) # output: [2 x 2 x 3]
# [[[1.0, 0.0, 0.0], # one_hot(0)
# [0.0, 0.0, 1.0]], # one_hot(2)
# [[0.0, 1.0, 0.0], # one_hot(1)
# [0.0, 0.0, 0.0]]] # one_hot(-1)
```
Args:
indices: A `Tensor` of indices.
depth: A scalar defining the depth of the one hot dimension.
on_value: A scalar defining the value to fill in output when `indices[j]
= i`. (default: 1)
off_value: A scalar defining the value to fill in output when `indices[j]
!= i`. (default: 0)
axis: The axis to fill (default: -1, a new inner-most axis).
dtype: The data type of the output tensor.
name: A name for the operation (optional).
Returns:
output: The one-hot tensor.
Raises:
TypeError: If dtype of either `on_value` or `off_value` don't match `dtype`
TypeError: If dtype of `on_value` and `off_value` don't match one another
"""
with ops.name_scope(
name, "one_hot",
[indices, depth, on_value, off_value, axis, dtype]) as name:
on_exists = on_value is not None
off_exists = off_value is not None
on_dtype = (
ops.convert_to_tensor(on_value).dtype.base_dtype if on_exists else None)
off_dtype = (
ops.convert_to_tensor(off_value).dtype.base_dtype
if off_exists else None)
if on_exists or off_exists:
if dtype is not None:
# Ensure provided on_value and/or off_value match dtype
if on_exists and on_dtype != dtype:
raise TypeError("dtype {0} of on_value does not match "
"dtype parameter {1}".format(on_dtype, dtype))
if off_exists and off_dtype != dtype:
raise TypeError("dtype {0} of off_value does not match "
"dtype parameter {1}".format(off_dtype, dtype))
else:
# dtype not provided: automatically assign it
dtype = on_dtype if on_exists else off_dtype
elif dtype is None:
# None of on_value, off_value, or dtype provided. Default dtype to float32
dtype = dtypes.float32
if not on_exists:
# on_value not provided: assign to value 1 of type dtype
on_value = ops.convert_to_tensor(1, dtype, name="on_value")
on_dtype = dtype
if not off_exists:
# off_value not provided: assign to value 0 of type dtype
off_value = ops.convert_to_tensor(0, dtype, name="off_value")
off_dtype = dtype
if on_dtype != off_dtype:
raise TypeError("dtype {0} of on_value does not match "
"dtype {1} of off_value".format(on_dtype, off_dtype))
return gen_array_ops.one_hot(indices, depth, on_value, off_value, axis,
name)
def _all_dimensions(x):
"""Returns a 1D-tensor listing all dimensions in x."""
# Fast path: avoid creating Rank and Range ops if ndims is known.
if isinstance(x, ops.Tensor) and x.get_shape().ndims is not None:
return constant_op.constant(
np.arange(x.get_shape().ndims), dtype=dtypes.int32)
if (isinstance(x, sparse_tensor.SparseTensor) and
x.dense_shape.get_shape().is_fully_defined()):
r = x.dense_shape.get_shape().dims[0].value # sparse.dense_shape is 1-D.
return constant_op.constant(np.arange(r), dtype=dtypes.int32)
# Otherwise, we rely on `range` and `rank` to do the right thing at runtime.
return gen_math_ops._range(0, rank(x), 1)
@tf_export("sequence_mask")
def sequence_mask(lengths, maxlen=None, dtype=dtypes.bool, name=None):
"""Returns a mask tensor representing the first N positions of each cell.
If `lengths` has shape `[d_1, d_2, ..., d_n]` the resulting tensor `mask` has
dtype `dtype` and shape `[d_1, d_2, ..., d_n, maxlen]`, with
```
mask[i_1, i_2, ..., i_n, j] = (j < lengths[i_1, i_2, ..., i_n])
```
Examples:
```python
tf.sequence_mask([1, 3, 2], 5) # [[True, False, False, False, False],
# [True, True, True, False, False],
# [True, True, False, False, False]]
tf.sequence_mask([[1, 3],[2,0]]) # [[[True, False, False],
# [True, True, True]],
# [[True, True, False],
# [False, False, False]]]
```
Args:
lengths: integer tensor, all its values <= maxlen.
maxlen: scalar integer tensor, size of last dimension of returned tensor.
Default is the maximum value in `lengths`.
dtype: output type of the resulting tensor.
name: name of the op.
Returns:
A mask tensor of shape `lengths.shape + (maxlen,)`, cast to specified dtype.
Raises:
ValueError: if `maxlen` is not a scalar.
"""
with ops.name_scope(name, "SequenceMask", [lengths, maxlen]):
lengths = ops.convert_to_tensor(lengths)
if maxlen is None:
maxlen = gen_math_ops._max(lengths, _all_dimensions(lengths))
maxlen = gen_math_ops.maximum(constant(0, maxlen.dtype), maxlen)
else:
maxlen = ops.convert_to_tensor(maxlen)
if maxlen.get_shape().ndims is not None and maxlen.get_shape().ndims != 0:
raise ValueError("maxlen must be scalar for sequence_mask")
# The basic idea is to compare a range row vector of size maxlen:
# [0, 1, 2, 3, 4]
# to length as a matrix with 1 column: [[1], [3], [2]].
# Because of broadcasting on both arguments this comparison results
# in a matrix of size (len(lengths), maxlen)
row_vector = gen_math_ops._range(
constant(0, maxlen.dtype), maxlen, constant(1, maxlen.dtype))
# Since maxlen >= max(lengths), it is safe to use maxlen as a cast
# authoritative type. Whenever maxlen fits into tf.int32, so do the lengths.
matrix = gen_math_ops.cast(expand_dims(lengths, -1), maxlen.dtype)
result = row_vector < matrix
if dtype is None or result.dtype.base_dtype == dtype.base_dtype:
return result
else:
return gen_math_ops.cast(result, dtype)
@tf_export(v1=["squeeze"])
@dispatch.add_dispatch_support
@deprecation.deprecated_args(None, "Use the `axis` argument instead",
"squeeze_dims")
def squeeze(input, axis=None, name=None, squeeze_dims=None):
# pylint: disable=redefined-builtin
"""Removes dimensions of size 1 from the shape of a tensor.
Given a tensor `input`, this operation returns a tensor of the same type with
all dimensions of size 1 removed. If you don't want to remove all size 1
dimensions, you can remove specific size 1 dimensions by specifying
`axis`.
For example:
```python
# 't' is a tensor of shape [1, 2, 1, 3, 1, 1]
tf.shape(tf.squeeze(t)) # [2, 3]
```
Or, to remove specific size 1 dimensions:
```python
# 't' is a tensor of shape [1, 2, 1, 3, 1, 1]
tf.shape(tf.squeeze(t, [2, 4])) # [1, 2, 3, 1]
```
Note: if `input` is a `tf.RaggedTensor`, then this operation takes `O(N)`
time, where `N` is the number of elements in the squeezed dimensions.
Args:
input: A `Tensor`. The `input` to squeeze.
axis: An optional list of `ints`. Defaults to `[]`. If specified, only
squeezes the dimensions listed. The dimension index starts at 0. It is an
error to squeeze a dimension that is not 1. Must be in the range
`[-rank(input), rank(input))`.
Must be specified if `input` is a `RaggedTensor`.
name: A name for the operation (optional).
squeeze_dims: Deprecated keyword argument that is now axis.
Returns:
A `Tensor`. Has the same type as `input`.
Contains the same data as `input`, but has one or more dimensions of
size 1 removed.
Raises:
ValueError: When both `squeeze_dims` and `axis` are specified.
"""
axis = deprecation.deprecated_argument_lookup("axis", axis, "squeeze_dims",
squeeze_dims)
if np.isscalar(axis):
axis = [axis]
return gen_array_ops.squeeze(input, axis, name)
@tf_export("squeeze", v1=[])
@dispatch.add_dispatch_support
def squeeze_v2(input, axis=None, name=None):
"""Removes dimensions of size 1 from the shape of a tensor.
Given a tensor `input`, this operation returns a tensor of the same type with
all dimensions of size 1 removed. If you don't want to remove all size 1
dimensions, you can remove specific size 1 dimensions by specifying
`axis`.
For example:
```python
# 't' is a tensor of shape [1, 2, 1, 3, 1, 1]
tf.shape(tf.squeeze(t)) # [2, 3]
```
Or, to remove specific size 1 dimensions:
```python
# 't' is a tensor of shape [1, 2, 1, 3, 1, 1]
tf.shape(tf.squeeze(t, [2, 4])) # [1, 2, 3, 1]
```
Unlike the older op `tf.compat.v1.squeeze`, this op does not accept a
deprecated `squeeze_dims` argument.
Note: if `input` is a `tf.RaggedTensor`, then this operation takes `O(N)`
time, where `N` is the number of elements in the squeezed dimensions.
Args:
input: A `Tensor`. The `input` to squeeze.
axis: An optional list of `ints`. Defaults to `[]`. If specified, only
squeezes the dimensions listed. The dimension index starts at 0. It is an
error to squeeze a dimension that is not 1. Must be in the range
`[-rank(input), rank(input))`. Must be specified if `input` is a
`RaggedTensor`.
name: A name for the operation (optional).
Returns:
A `Tensor`. Has the same type as `input`.
Contains the same data as `input`, but has one or more dimensions of
size 1 removed.
Raises:
ValueError: The input cannot be converted to a tensor, or the specified
axis cannot be squeezed.
"""
# pylint: disable=redefined-builtin
return squeeze(input, axis, name)
@tf_export(v1=["where"])
@deprecation.deprecated(
date=None,
instructions="Use tf.where in 2.0, "
"which has the same broadcast rule as np.where")
@dispatch.add_dispatch_support
def where(condition, x=None, y=None, name=None):
"""Return the elements, either from `x` or `y`, depending on the `condition`.
If both `x` and `y` are None, then this operation returns the coordinates of
true elements of `condition`. The coordinates are returned in a 2-D tensor
where the first dimension (rows) represents the number of true elements, and
the second dimension (columns) represents the coordinates of the true
elements. Keep in mind, the shape of the output tensor can vary depending on
how many true values there are in input. Indices are output in row-major
order.
If both non-None, `x` and `y` must have the same shape.
The `condition` tensor must be a scalar if `x` and `y` are scalar.
If `x` and `y` are vectors of higher rank, then `condition` must be either a
vector with size matching the first dimension of `x`, or must have the same
shape as `x`.
The `condition` tensor acts as a mask that chooses, based on the value at each
element, whether the corresponding element / row in the output should be taken
from `x` (if true) or `y` (if false).
If `condition` is a vector and `x` and `y` are higher rank matrices, then it
chooses which row (outer dimension) to copy from `x` and `y`. If `condition`
has the same shape as `x` and `y`, then it chooses which element to copy from
`x` and `y`.
Args:
condition: A `Tensor` of type `bool`
x: A Tensor which may have the same shape as `condition`. If `condition` is
rank 1, `x` may have higher rank, but its first dimension must match the
size of `condition`.
y: A `tensor` with the same shape and type as `x`.
name: A name of the operation (optional)
Returns:
A `Tensor` with the same type and shape as `x`, `y` if they are non-None.
Otherwise, a `Tensor` with shape `(num_true, rank(condition))`.
Raises:
ValueError: When exactly one of `x` or `y` is non-None.
"""
if x is None and y is None:
with ops.name_scope(name, "Where", [condition]) as name:
condition = ops.convert_to_tensor(
condition, preferred_dtype=dtypes.bool, name="condition")
return gen_array_ops.where(condition=condition, name=name)
elif x is not None and y is not None:
return gen_math_ops.select(condition=condition, x=x, y=y, name=name)
else:
raise ValueError("x and y must both be non-None or both be None.")
@tf_export("where", v1=["where_v2"])
def where_v2(condition, x=None, y=None, name=None):
"""Return the elements, either from `x` or `y`, depending on the `condition`.
If both `x` and `y` are None, then this operation returns the coordinates of
true elements of `condition`. The coordinates are returned in a 2-D tensor
where the first dimension (rows) represents the number of true elements, and
the second dimension (columns) represents the coordinates of the true
elements. Keep in mind, the shape of the output tensor can vary depending on
how many true values there are in input. Indices are output in row-major
order.
If both non-None, `condition`, `x` and `y` must be broadcastable to the same
shape.
The `condition` tensor acts as a mask that chooses, based on the value at each
element, whether the corresponding element / row in the output should be taken
from `x` (if true) or `y` (if false).
Args:
condition: A `Tensor` of type `bool`
x: A Tensor which is of the same type as `y`, and may be broadcastable with
`condition` and `y`.
y: A Tensor which is of the same type as `x`, and may be broadcastable with
`condition` and `x`.
name: A name of the operation (optional).
Returns:
A `Tensor` with the same type as `x` and `y`, and shape that
is broadcasted from `condition`, `x`, and `y`, if `x`, `y` are non-None.
A `Tensor` with shape `(num_true, dim_size(condition))`.
Raises:
ValueError: When exactly one of `x` or `y` is non-None.
"""
if x is None and y is None:
with ops.name_scope(name, "Where", [condition]) as name:
condition = ops.convert_to_tensor(
condition, preferred_dtype=dtypes.bool, name="condition")
return gen_array_ops.where(condition=condition, name=name)
elif x is not None and y is not None:
return gen_math_ops.select_v2(condition=condition, t=x, e=y, name=name)
else:
raise ValueError("x and y must both be non-None or both be None.")
# pylint: disable=redefined-builtin
@tf_export(v1=["reverse_sequence"])
@deprecation.deprecated_args(None,
"seq_dim is deprecated, use seq_axis instead",
"seq_dim")
@deprecation.deprecated_args(None,
"batch_dim is deprecated, use batch_axis instead",
"batch_dim")
def reverse_sequence(input,
seq_lengths,
seq_axis=None,
batch_axis=None,
name=None,
seq_dim=None,
batch_dim=None):
seq_axis = deprecation.deprecated_argument_lookup("seq_axis", seq_axis,
"seq_dim", seq_dim)
batch_axis = deprecation.deprecated_argument_lookup("batch_axis", batch_axis,
"batch_dim", batch_dim)
return gen_array_ops.reverse_sequence(
input=input,
seq_lengths=seq_lengths,
seq_dim=seq_axis,
batch_dim=batch_axis,
name=name)
reverse_sequence.__doc__ = deprecation.rewrite_argument_docstring(
deprecation.rewrite_argument_docstring(
gen_array_ops.reverse_sequence.__doc__, "batch_dim", "batch_axis"),
"seq_dim", "seq_axis")
@tf_export("reverse_sequence", v1=[])
def reverse_sequence_v2(input,
seq_lengths,
seq_axis=None,
batch_axis=None,
name=None):
return gen_array_ops.reverse_sequence(
input=input,
seq_lengths=seq_lengths,
seq_dim=seq_axis,
batch_dim=batch_axis,
name=name)
reverse_sequence_v2.__doc__ = deprecation.rewrite_argument_docstring(
deprecation.rewrite_argument_docstring(
gen_array_ops.reverse_sequence.__doc__, "batch_dim", "batch_axis"),
"seq_dim", "seq_axis")
# pylint: enable=redefined-builtin
@tf_export(v1=["gather"])
@dispatch.add_dispatch_support
def gather(params,
indices,
validate_indices=None,
name=None,
axis=None,
batch_dims=0): # pylint: disable=g-doc-args
r"""Gather slices from params axis axis according to indices.
Gather slices from params axis `axis` according to `indices`. `indices` must
be an integer tensor of any dimension (usually 0-D or 1-D).
For 0-D (scalar) `indices`:
> `output`$$[p_0, ..., p_{axis-1}, \hspace{5.1em}
> p_{axis + 1}, ..., p_{N-1}]$$ =\
> `params`$$[p_0, ..., p_{axis-1}, \hspace{1em}
> indices, \hspace{1em}
> p_{axis + 1}, ..., p_{N-1}]$$.
For 1-D (vector) `indices` with `batch_dims=0`:
> `output`$$[p_0, ..., p_{axis-1}, \hspace{2.6em}
> i, \hspace{2.6em}
> p_{axis + 1}, ..., p_{N-1}]$$ =\
> `params`$$[p_0, ..., p_{axis-1}, \hspace{1em}
> indices[i], \hspace{1em}
> p_{axis + 1}, ..., p_{N-1}]$$.
In the general case, produces an output tensor where:
> `output`$$[p_0, ..., p_{axis-1}, \hspace{1.2em}
> i_{batch\_dims}, ..., i_{M-1}, \hspace{1.3em}
> p_{axis + 1}, ..., p_{N-1}]$$ =\
> `params`$$[p_0, ..., p_{axis-1}, \hspace{1em}
> indices[i_0, ..., i_{M-1}], \hspace{1em}
> p_{axis + 1}, ..., p_{N-1}]$$.
Where $$N$$=`ndims(params)` and $$M$$=`ndims(indices)`.
The shape of the output tensor is:
> `output.shape = params.shape[:axis] + indices.shape[batch_dims:] +
> params.shape[axis + 1:]`.
Note that on CPU, if an out of bound index is found, an error is returned.
On GPU, if an out of bound index is found, a 0 is stored in the corresponding
output value.
See also `tf.gather_nd`.
<div style="width:70%; margin:auto; margin-bottom:10px; margin-top:20px;">
<img style="width:100%" src="https://www.tensorflow.org/images/Gather.png"
alt>
</div>
Args:
params: The `Tensor` from which to gather values. Must be at least rank
`axis + 1`.
indices: The index `Tensor`. Must be one of the following types: `int32`,
`int64`. Must be in range `[0, params.shape[axis])`.
validate_indices: Deprecated, does nothing.
axis: A `Tensor`. Must be one of the following types: `int32`, `int64`. The
`axis` in `params` to gather `indices` from. Must be greater than or equal
to `batch_dims`. Defaults to the first non-batch dimension. Supports
negative indexes.
batch_dims: An `integer`. The number of batch dimensions. Must be less
than `rank(indices)`.
name: A name for the operation (optional).
Returns:
A `Tensor`. Has the same type as `params`.
"""
del validate_indices
if compat.forward_compatible(2019, 7, 10):
if axis is None:
axis = batch_dims
if axis != 0:
return gen_array_ops.gather_v2(
params, indices, axis, batch_dims=batch_dims, name=name)
try:
# TODO(apassos) find a less bad way of detecting resource variables
# without introducing a circular dependency.
return params.sparse_read(indices, name=name)
except AttributeError:
return gen_array_ops.gather_v2(
params, indices, axis, name=name)
if batch_dims != 0:
with ops.name_scope(name, "Gather", [params, indices, axis]):
return _batch_gather(params, indices, batch_dims, axis)
if axis is None:
axis = batch_dims
if axis != 0:
# Note that we do a sparse_read here to avoid snapshotting the entire
# resource variable and doing a gather, which can be inefficient and lead to
# subtle race conditions. TODO(apassos) implement axis != 0 on sparse_read
return gen_array_ops.gather_v2(params, indices, axis, name=name)
try:
# TODO(apassos) find a less bad way of detecting resource variables without
# introducing a circular dependency.
return params.sparse_read(indices, name=name)
except AttributeError:
return gen_array_ops.gather_v2(params, indices, axis, name=name)
@tf_export("gather", v1=[])
@dispatch.add_dispatch_support
def gather_v2(params,
indices,
validate_indices=None,
axis=None,
batch_dims=0,
name=None):
return gather(
params,
indices,
validate_indices=validate_indices,
name=name,
axis=axis,
batch_dims=batch_dims)
gather_v2.__doc__ = gather.__doc__
@tf_export(v1=["batch_gather"])
@dispatch.add_dispatch_support
@deprecation.deprecated(
"2017-10-25", "`tf.batch_gather` is deprecated, please use `tf.gather` "
"with `batch_dims=-1` instead.") # pylint: disable=missing-docstring
def batch_gather(params, indices, name=None):
"""Gather slices from params according to indices with leading batch dims."""
with ops.name_scope(name, "BatchGather", [params, indices]):
indices = ops.convert_to_tensor(indices, name="indices")
params = ops.convert_to_tensor(params, name="params")
if indices.shape.ndims is None:
raise ValueError(
"batch_gather does not allow indices with unknown shape.")
return _batch_gather(params, indices, batch_dims=indices.shape.ndims - 1)
def _batch_gather(params, indices, batch_dims, axis=None):
r"""Gather slices from params according to indices with leading batch dims.
This operation assumes that the leading `batch_dims` dimensions of `indices`
and `params` are batch dimensions; and performs a `tf.gather` operation within
each batch. (If `batch_dims` is not specified, then it defaults to
`rank(indices)-1`.) In the case in which `batch_dims==0`, this operation
is equivalent to `tf.gather`.
Args:
params: A Tensor. The tensor from which to gather values.
indices: A Tensor. Must be one of the following types: int32, int64. Index
tensor. Must be in range `[0, params.shape[batch_dims]]`.
batch_dims: An integer or none. The number of batch dimensions. Must be
less than `rank(indices)`. Defaults to `rank(indices) - 1` if None.
axis: A `Tensor`. Must be one of the following types: `int32`, `int64`. The
`axis` in `params` to gather `indices` from. Must be greater than or equal
to `batch_dims`. Defaults to the first non-batch dimension. Supports
negative indexes.
Returns:
A Tensor. Has the same type as `params`.
Raises:
ValueError: if `indices` has an unknown shape.
"""
if batch_dims is not None and not isinstance(batch_dims, int):
raise TypeError("batch_dims must be an int; got %r" % (batch_dims,))
indices = ops.convert_to_tensor(indices, name="indices")
params = ops.convert_to_tensor(params, name="params")
indices_ndims = indices.shape.ndims
if indices_ndims is None:
raise ValueError("tf.gather does not allow indices with unknown "
"rank when batch_dims is specified.")
if batch_dims is None:
batch_dims = indices_ndims - 1
if batch_dims < 0:
batch_dims += indices_ndims
if batch_dims < 0 or batch_dims >= indices_ndims:
raise ValueError("batch_dims = %d must be less than rank(indices) = %d" %
(batch_dims, indices_ndims))
if params.shape.ndims is not None and batch_dims >= params.shape.ndims:
raise ValueError("batch_dims = %d must be less than rank(params) = %d" %
(batch_dims, params.shape.ndims))
# Handle axis by transposing the axis dimension to be the first non-batch
# dimension, recursively calling batch_gather with axis=0, and then
# transposing the result to put the pre-axis dimensions before the indices
# dimensions.
if axis is not None and axis != batch_dims:
# Adjust axis to be positive.
if not isinstance(axis, int):
axis = tf.where(axis < 0, axis + array_ops.rank(params), axis)
elif axis < 0 and params.shape.ndims is None:
axis = axis + array_ops.rank(params)
else:
if (axis < -params.shape.ndims) or (axis >= params.shape.ndims):
raise ValueError("axis (%d) out of range [%d, %d)" %
(axis, -params.shape.ndims, params.shape.ndims))
if axis < 0:
axis += params.shape.ndims
if axis < batch_dims:
raise ValueError("batch_dims = %d must be less than or equal to "
"axis = %d" % (batch_dims, axis))
# Move params[axis] up to params[batch_dims].
perm = [
list(range(batch_dims)), [axis],
gen_math_ops._range(batch_dims, axis, 1),
gen_math_ops._range(axis + 1, rank(params), 1)
]
params = transpose(params, concat(perm, axis=0))
result = _batch_gather(params, indices, batch_dims=batch_dims)
# Move the result dimensions corresponding to params[batch_dims:axis]
# to just before the dimensions corresponding to indices[batch_dims:].
params_start = indices_ndims + axis - batch_dims
perm = [
list(range(batch_dims)),
gen_math_ops._range(indices_ndims, params_start, 1),
list(range(batch_dims, indices_ndims)),
gen_math_ops._range(params_start, rank(result), 1)
]
return transpose(result, perm=concat(perm, axis=0))
indices_shape = shape(indices)
params_shape = shape(params)
batch_indices = indices
indices_dtype = indices.dtype.base_dtype
accum_dim_value = ones((), dtype=indices_dtype)
# Use correct type for offset index computation
casted_params_shape = gen_math_ops.cast(params_shape, indices_dtype)
for dim in range(batch_dims, 0, -1):
dim_value = casted_params_shape[dim - 1]
accum_dim_value *= casted_params_shape[dim]
start = zeros((), dtype=indices_dtype)
step = ones((), dtype=indices_dtype)
dim_indices = gen_math_ops._range(start, dim_value, step)
dim_indices *= accum_dim_value
dim_shape = stack(
[1] * (dim - 1) + [dim_value] + [1] * (indices_ndims - dim), axis=0)
batch_indices += reshape(dim_indices, dim_shape)
flat_indices = reshape(batch_indices, [-1])
outer_shape = params_shape[batch_dims + 1:]
flat_inner_shape = gen_math_ops.prod(params_shape[:batch_dims + 1], [0],
False)
flat_params = reshape(params, concat([[flat_inner_shape], outer_shape],
axis=0))
flat_result = gather(flat_params, flat_indices)
result = reshape(flat_result, concat([indices_shape, outer_shape], axis=0))
final_shape = indices.get_shape()[:batch_dims].merge_with(
params.get_shape()[:batch_dims])
final_shape = final_shape.concatenate(indices.get_shape().dims[batch_dims:])
final_shape = final_shape.concatenate(params.get_shape()[batch_dims + 1:])
result.set_shape(final_shape)
return result
@tf_export(v1=["gather_nd", "manip.gather_nd"])
@dispatch.add_dispatch_support
@deprecated_endpoints("manip.gather_nd")
def gather_nd(params, indices, name=None, batch_dims=0):
r"""Gather slices from `params` into a Tensor with shape specified by `indices`.
`indices` is an K-dimensional integer tensor, best thought of as a
(K-1)-dimensional tensor of indices into `params`, where each element defines
a slice of `params`:
output[\\(i_0, ..., i_{K-2}\\)] = params[indices[\\(i_0, ..., i_{K-2}\\)]]
Whereas in `tf.gather` `indices` defines slices into the first
dimension of `params`, in `tf.gather_nd`, `indices` defines slices into the
first `N` dimensions of `params`, where `N = indices.shape[-1]`.
The last dimension of `indices` can be at most the rank of
`params`:
indices.shape[-1] <= params.rank
The last dimension of `indices` corresponds to elements
(if `indices.shape[-1] == params.rank`) or slices
(if `indices.shape[-1] < params.rank`) along dimension `indices.shape[-1]`
of `params`. The output tensor has shape
indices.shape[:-1] + params.shape[indices.shape[-1]:]
Additionally both 'params' and 'indices' can have M leading batch
dimensions that exactly match. In this case 'batch_dims' must be M.
Note that on CPU, if an out of bound index is found, an error is returned.
On GPU, if an out of bound index is found, a 0 is stored in the
corresponding output value.
Some examples below.
Simple indexing into a matrix:
```python
indices = [[0, 0], [1, 1]]
params = [['a', 'b'], ['c', 'd']]
output = ['a', 'd']
```
Slice indexing into a matrix:
```python
indices = [[1], [0]]
params = [['a', 'b'], ['c', 'd']]
output = [['c', 'd'], ['a', 'b']]
```
Indexing into a 3-tensor:
```python
indices = [[1]]
params = [[['a0', 'b0'], ['c0', 'd0']],
[['a1', 'b1'], ['c1', 'd1']]]
output = [[['a1', 'b1'], ['c1', 'd1']]]
indices = [[0, 1], [1, 0]]
params = [[['a0', 'b0'], ['c0', 'd0']],
[['a1', 'b1'], ['c1', 'd1']]]
output = [['c0', 'd0'], ['a1', 'b1']]
indices = [[0, 0, 1], [1, 0, 1]]
params = [[['a0', 'b0'], ['c0', 'd0']],
[['a1', 'b1'], ['c1', 'd1']]]
output = ['b0', 'b1']
```
The examples below are for the case when only indices have leading extra
dimensions. If both 'params' and 'indices' have leading batch dimensions, use
the 'batch_dims' parameter to run gather_nd in batch mode.
Batched indexing into a matrix:
```python
indices = [[[0, 0]], [[0, 1]]]
params = [['a', 'b'], ['c', 'd']]
output = [['a'], ['b']]
```
Batched slice indexing into a matrix:
```python
indices = [[[1]], [[0]]]
params = [['a', 'b'], ['c', 'd']]
output = [[['c', 'd']], [['a', 'b']]]
```
Batched indexing into a 3-tensor:
```python
indices = [[[1]], [[0]]]
params = [[['a0', 'b0'], ['c0', 'd0']],
[['a1', 'b1'], ['c1', 'd1']]]
output = [[[['a1', 'b1'], ['c1', 'd1']]],
[[['a0', 'b0'], ['c0', 'd0']]]]
indices = [[[0, 1], [1, 0]], [[0, 0], [1, 1]]]
params = [[['a0', 'b0'], ['c0', 'd0']],
[['a1', 'b1'], ['c1', 'd1']]]
output = [[['c0', 'd0'], ['a1', 'b1']],
[['a0', 'b0'], ['c1', 'd1']]]
indices = [[[0, 0, 1], [1, 0, 1]], [[0, 1, 1], [1, 1, 0]]]
params = [[['a0', 'b0'], ['c0', 'd0']],
[['a1', 'b1'], ['c1', 'd1']]]
output = [['b0', 'b1'], ['d0', 'c1']]
```
Examples with batched 'params' and 'indices':
```python
batch_dims = 1
indices = [[1], [0]]
params = [[['a0', 'b0'], ['c0', 'd0']],
[['a1', 'b1'], ['c1', 'd1']]]
output = [['c0', 'd0'], ['a1', 'b1']]
batch_dims = 1
indices = [[[1]], [[0]]]
params = [[['a0', 'b0'], ['c0', 'd0']],
[['a1', 'b1'], ['c1', 'd1']]]
output = [[['c0', 'd0']], [['a1', 'b1']]]
batch_dims = 1
indices = [[[1, 0]], [[0, 1]]]
params = [[['a0', 'b0'], ['c0', 'd0']],
[['a1', 'b1'], ['c1', 'd1']]]
output = [['c0'], ['b1']]
```
See also `tf.gather`.
Args:
params: A `Tensor`. The tensor from which to gather values.
indices: A `Tensor`. Must be one of the following types: `int32`, `int64`.
Index tensor.
name: A name for the operation (optional).
batch_dims: An integer or a scalar 'Tensor'. The number of batch dimensions.
Returns:
A `Tensor`. Has the same type as `params`.
"""
batch_dims_ = tensor_util.constant_value(batch_dims)
if batch_dims_ is not None:
batch_dims = int(batch_dims_)
if batch_dims == 0:
if compat.forward_compatible(2019, 4, 29):
try:
# TODO(apassos) find a less bad way of detecting resource variables
# without introducing a circular dependency.
return params.gather_nd(indices, name=name)
except AttributeError:
return gen_array_ops.gather_nd(params, indices, name=name)
else:
return gen_array_ops.gather_nd(params, indices, name=name)
else:
return batch_gather_nd(params, indices, batch_dims=batch_dims, name=name)
@tf_export("gather_nd", v1=[])
@dispatch.add_dispatch_support
def gather_nd_v2(params, indices, batch_dims=0, name=None):
return gather_nd(params, indices, name=name, batch_dims=batch_dims)
gather_nd_v2.__doc__ = gather_nd.__doc__
def batch_gather_nd(params, indices, batch_dims, name=None):
"""gather_nd implementation with batch support."""
with ops.name_scope(name, "BatchGatherND", [params, indices]):
indices = ops.convert_to_tensor(indices, name="indices")
params = ops.convert_to_tensor(params, name="params")
if not isinstance(batch_dims, int):
raise TypeError("batch_dims must be an int; got %r" % (batch_dims,))
if batch_dims < 0:
raise ValueError("tf.gather_nd does not allow negative batch_dims.")
params_ndims = params.shape.ndims
indices_ndims = indices.shape.ndims
if indices_ndims is not None and batch_dims >= indices_ndims:
raise ValueError("batch_dims = %d must be less than rank(indices) = %d" %
(batch_dims, indices_ndims))
if params_ndims is not None and batch_dims >= params_ndims:
raise ValueError("batch_dims = %d must be less than rank(params) = %d" %
(batch_dims, params_ndims))
expand = batch_dims == 0
if expand:
# Normally gather_nd will be called when batch_dims == 0.
# But if this function is called with batch_dims = 0, e.g. for testing
# purposes, this adds a dummy batch dimension to make batch_dims = 1.
params = expand_dims(params, axis=0)
indices = expand_dims(indices, axis=0)
batch_dims = 1
params_shape = shape(params)
indices_shape = shape(indices)
batch_shape = params_shape[:batch_dims]
batch_size = gen_math_ops.prod(batch_shape, [0])
index_internal_ndims = rank(indices) - batch_dims - 1
indices_internal_shape = indices_shape[batch_dims:-1]
# Assuming a 'params' with shape [b1, ..., bM, g1, ..., gN] and an 'indices'
# with shape [b1, ..., bM, i1, ..., iK, C], where C <= N, we need to modify
# 'indices' s.t. it has shape [i1, ..., iK, D], where D <= M + N and slices
# to the entire 'params' tensor.
# Assuming we have a batch of shape [B1, B2], we use meshgrid to create a
# grid of size B1 x B2.
batch_dim_list = unstack(batch_shape, axis=0)
dim_ranges = [
gen_math_ops.cast(gen_math_ops._range(0, x, 1), indices.dtype)
for x in batch_dim_list
]
mesh_list = meshgrid(*dim_ranges, indexing="ij") if dim_ranges else []
# Then we flatten and stack the tensors to form a (B1.B2) by 2 matrix.
flat_list = [reshape(x, shape=(-1,)) for x in mesh_list]
index_grid = transpose(stack(flat_list, axis=0))
# We need to concatenate these batch coordinates with the internal indices.
# concat -> index_grid [B1.B2, 2] with indices [i1, ..., iK, C]
# So we reshape them both to [(B1.B2), i1, ..., iK, *]
index_grid_shape = shape(index_grid)
index_grid = reshape(
index_grid,
concat([
index_grid_shape[:1],
ones(index_internal_ndims, dtype=dtypes.int32), index_grid_shape[1:]
],
axis=0))
tile_shape = concat(((1,), indices_internal_shape, (1,)), axis=0)
index_grid = tile(index_grid, multiples=tile_shape)
# index_grid now has shape [(B1.B2), i1, ..., iK, 2]
flat_shape = concat(([batch_size], indices_shape[batch_dims:]), axis=0)
flat_indices = reshape(indices, shape=flat_shape)
# flat_indices now has shape [(B1.B2), i1, ..., iK, C]
indices = concat((index_grid, flat_indices), axis=-1)
# indices has shape [(B1.B2), i1, ..., iK, 2+C]
out = gen_array_ops.gather_nd(params, indices)
# out has shape [(B1.B2), i1, ..., iK, N-C]. Now we reshape batch to
# its original form.
out_shape = shape(out)
out = reshape(out, shape=concat((batch_shape, out_shape[1:]), axis=0))
if expand:
out = squeeze(out, axis=0)
return out
# Define quantize_v2 here in order to make name the second-to-last attribute,
# because round_mode was added later.
@tf_export(v1=["quantize_v2"])
@deprecation.deprecated(
"2017-10-25",
"`tf.quantize_v2` is deprecated, please use `tf.quantization.quantize` "
"instead.") # pylint: disable=missing-docstring
def quantize_v2(
input, # pylint: disable=redefined-builtin
min_range,
max_range,
T,
mode="MIN_COMBINED",
name=None,
round_mode="HALF_AWAY_FROM_ZERO"):
return gen_array_ops.quantize_v2(
input,
min_range,
max_range,
T=T,
mode=mode,
name=name,
round_mode=round_mode)
quantize_v2.__doc__ = """Please use `tf.quantization.quantize` instead."""
# We want to expose tf.quantization.quantize instead of
# tf.quantization.quantize; we can deprecate tf.quantization.quantize in next
# version of TensorFlow.
@tf_export("quantization.quantize", v1=["quantization.quantize", "quantize"])
@deprecation.deprecated_endpoints("quantize")
def quantize(input, # pylint: disable=redefined-builtin
min_range,
max_range,
T,
mode="MIN_COMBINED",
round_mode="HALF_AWAY_FROM_ZERO",
name=None):
return gen_array_ops.quantize_v2(
input,
min_range,
max_range,
T,
mode=mode,
round_mode=round_mode,
name=name)
@tf_export("quantization.quantize_and_dequantize")
def quantize_and_dequantize(input, # pylint: disable=redefined-builtin
input_min,
input_max,
signed_input=True,
num_bits=8,
range_given=False,
round_mode="HALF_TO_EVEN",
name=None,
narrow_range=False):
"""Quantizes then dequantizes a tensor.
Args:
input: A `Tensor` to quantize and dequantize.
input_min: If range_given=True, the minimum input value that needs to be
represented in the quantized representation.
input_max: If range_given=True, the maximum input value that needs to be
represented in the quantized representation.
signed_input: True if the quantization is signed or unsigned.
num_bits: The bitwidth of the quantization.
range_given: If true use `input_min` and `input_max` for the range of the
input, otherwise determine min and max from the input `Tensor`.
round_mode: Rounding mode when rounding from float values to quantized ones.
name: Optional name for the operation.
narrow_range: If true, then the absolute value of the quantized minimum
value is the same as the quantized maximum value, instead of 1 greater.
i.e. for 8 bit quantization, the minimum value is -127 instead of -128.
Returns:
A `Tensor`. Each element is the result of quantizing and dequantizing the
corresponding element of `input`.
"""
return gen_array_ops.quantize_and_dequantize_v2(
input,
input_min=input_min,
input_max=input_max,
signed_input=signed_input,
num_bits=num_bits,
range_given=range_given,
round_mode=round_mode,
narrow_range=narrow_range,
name=name)
@tf_export("searchsorted")
def searchsorted(sorted_sequence,
values,
side="left",
out_type=dtypes.int32,
name=None):
"""Searches input tensor for values on the innermost dimension.
A 2-D example:
```
sorted_sequence = [[0, 3, 9, 9, 10],
[1, 2, 3, 4, 5]]
values = [[2, 4, 9],
[0, 2, 6]]
result = searchsorted(sorted_sequence, values, side="left")
result == [[1, 2, 2],
[0, 1, 5]]
result = searchsorted(sorted_sequence, values, side="right")
result == [[1, 2, 4],
[0, 2, 5]]
```
Args:
sorted_sequence: N-D `Tensor` containing a sorted sequence.
values: N-D `Tensor` containing the search values.
side: 'left' or 'right'; 'left' corresponds to lower_bound and 'right' to
upper_bound.
out_type: The output type (`int32` or `int64`). Default is `tf.int32`.
name: Optional name for the operation.
Returns:
An N-D `Tensor` the size of values containing the result of applying either
lower_bound or upper_bound (depending on side) to each value. The result
is not a global index to the entire `Tensor`, but the index in the last
dimension.
Raises:
ValueError: If the last dimension of `sorted_sequence >= 2^31-1` elements.
If the total size of values exceeds `2^31 - 1` elements.
If the first `N-1` dimensions of the two tensors don't match.
"""
sequence_size = shape_internal(sorted_sequence)[-1]
values_size = shape_internal(values)[-1]
sorted_sequence_2d = reshape(sorted_sequence, [-1, sequence_size])
values_2d = reshape(values, [-1, values_size])
if side == "right":
output = gen_array_ops.upper_bound(sorted_sequence_2d, values_2d, out_type,
name)
elif side == "left":
output = gen_array_ops.lower_bound(sorted_sequence_2d, values_2d, out_type,
name)
else:
raise ValueError("side must be either 'right' or 'left'. Saw: %s." % side)
return reshape(output, shape_internal(values))
quantize.__doc__ = gen_array_ops.quantize_v2.__doc__
@tf_export("image.extract_patches")
def extract_image_patches_v2(images, sizes, strides, rates, padding, name=None):
# pylint: disable=line-too-long
r"""Extract `patches` from `images` and put them in the \"depth\" output dimension.
Args:
images: A 4-D Tensor with shape `[batch, in_rows, in_cols, depth]
sizes: The size of the sliding window for each dimension of `images`.
strides: A 1-D Tensor of length 4. How far the centers of two consecutive
patches are in the images. Must be: `[1, stride_rows, stride_cols, 1]`.
rates: A 1-D Tensor of length 4. Must be: `[1, rate_rows, rate_cols, 1]`.
This is the input stride, specifying how far two consecutive patch samples
are in the input. Equivalent to extracting patches with `patch_sizes_eff =
patch_sizes + (patch_sizes - 1) * (rates - 1)`, followed by subsampling
them spatially by a factor of `rates`. This is equivalent to `rate` in
dilated (a.k.a. Atrous) convolutions.
padding: The type of padding algorithm to use.
We specify the size-related attributes as: ```python ksizes = [1,
ksize_rows, ksize_cols, 1] strides = [1, strides_rows, strides_cols, 1]
rates = [1, rates_rows, rates_cols, 1]
name: A name for the operation (optional).
Returns:
A 4-D Tensor. Has the same type as `images`, and with shape `[batch,
out_rows, out_cols, ksize_rows * ksize_cols * depth]` containing image
patches with size `ksize_rows x ksize_cols x depth` vectorized in the
\"depth\" dimension. Note `out_rows` and `out_cols` are the dimensions of
the output patches.
"""
# pylint: enable=line-too-long
return gen_array_ops.extract_image_patches(images, sizes, strides, rates,
padding, name)
@tf_export(v1=["image.extract_image_patches", "extract_image_patches"])
@deprecation.deprecated_args(None, "ksizes is deprecated, use sizes instead",
"ksizes")
def extract_image_patches( # pylint: disable=missing-docstring
images,
ksizes=None,
strides=None,
rates=None,
padding=None,
name=None,
sizes=None):
ksizes = deprecation.deprecated_argument_lookup("sizes", sizes, "ksizes",
ksizes)
return gen_array_ops.extract_image_patches(images, ksizes, strides, rates,
padding, name)
extract_image_patches.__doc__ = gen_array_ops.extract_image_patches.__doc__
@tf_export("fingerprint")
def fingerprint(data, method="farmhash64", name=None):
r"""Generates fingerprint values.
Generates fingerprint values of `data`.
Fingerprint op considers the first dimension of `data` as the batch dimension,
and `output[i]` contains the fingerprint value generated from contents in
`data[i, ...]` for all `i`.
Fingerprint op writes fingerprint values as byte arrays. For example, the
default method `farmhash64` generates a 64-bit fingerprint value at a time.
This 8-byte value is written out as an `tf.uint8` array of size 8, in
little-endian order.
For example, suppose that `data` has data type `tf.int32` and shape (2, 3, 4),
and that the fingerprint method is `farmhash64`. In this case, the output
shape is (2, 8), where 2 is the batch dimension size of `data`, and 8 is the
size of each fingerprint value in bytes. `output[0, :]` is generated from
12 integers in `data[0, :, :]` and similarly `output[1, :]` is generated from
other 12 integers in `data[1, :, :]`.
Note that this op fingerprints the raw underlying buffer, and it does not
fingerprint Tensor's metadata such as data type and/or shape. For example, the
fingerprint values are invariant under reshapes and bitcasts as long as the
batch dimension remain the same:
```python
tf.fingerprint(data) == tf.fingerprint(tf.reshape(data, ...))
tf.fingerprint(data) == tf.fingerprint(tf.bitcast(data, ...))
```
For string data, one should expect `tf.fingerprint(data) !=
tf.fingerprint(tf.string.reduce_join(data))` in general.
Args:
data: A `Tensor`. Must have rank 1 or higher.
method: A `Tensor` of type `tf.string`. Fingerprint method used by this op.
Currently available method is `farmhash64`.
name: A name for the operation (optional).
Returns:
A two-dimensional `Tensor` of type `tf.uint8`. The first dimension equals to
`data`'s first dimension, and the second dimension size depends on the
fingerprint algorithm.
"""
return gen_array_ops.fingerprint(data, method, name)
|
tensorflow-master
|
tensorflow/python/ops/array_ops.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for tensorflow.ops.math_ops."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.eager import context
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import errors_impl
from tensorflow.python.framework import ops
from tensorflow.python.framework import test_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gradients
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import resource_variable_ops
from tensorflow.python.ops import variables
from tensorflow.python.platform import googletest
exp = np.exp
log = np.log
class ReduceTest(test_util.TensorFlowTestCase):
@test_util.run_in_graph_and_eager_modes
def testReduceAllDims(self):
x = np.array([[1, 2, 3], [4, 5, 6]], dtype=np.int32)
with test_util.device(use_gpu=True):
y_tf = self.evaluate(math_ops.reduce_sum(x))
self.assertEqual(y_tf, 21)
@test_util.run_in_graph_and_eager_modes
def testReduceExplicitAxes(self):
x = np.array([[1, 2, 3], [4, 5, 6]], dtype=np.int32)
with test_util.device(use_gpu=True):
for axis in (0, -2, (0, 0), (0, -2)):
self.assertAllEqual(self.evaluate(math_ops.reduce_sum(x, axis=axis)),
[5, 7, 9])
for axis in (1, -1, (1, 1), (1, -1)):
self.assertAllEqual(self.evaluate(math_ops.reduce_sum(x, axis=axis)),
[6, 15])
for axis in (None, (0, 1), (-1, -2), (-2, -1, 0, 1)):
self.assertEqual(self.evaluate(math_ops.reduce_sum(x, axis=axis)), 21)
@test_util.run_in_graph_and_eager_modes
def testReduceInvalidAxis(self):
if context.executing_eagerly():
# The shape check is in run a graph construction time. In eager mode,
# it misses the check, magically return result given wrong shape.
return
x = np.array([[1, 2, 3], [4, 5, 6]], dtype=np.int32)
axis = np.array([[0], [1]])
with self.assertRaisesRegexp(ValueError, "must be at most rank 1"):
math_ops.reduce_sum(x, axis)
@test_util.run_in_graph_and_eager_modes
def testReduceVar(self):
x = np.array([[0, 0, 0], [0, 0, 0]], "float32")
self.assertAllClose(self.evaluate(math_ops.reduce_variance(x)), 0)
self.assertAllClose(
self.evaluate(math_ops.reduce_variance(x, axis=0)), [0, 0, 0])
x = np.array([[0, 2, 1, 1], [1, 2, 0, 1]], "float32")
self.assertAllClose(self.evaluate(math_ops.reduce_variance(x)), 0.5)
@test_util.run_in_graph_and_eager_modes
def testReduceStd(self):
x = np.array([[0, 0, 0], [0, 0, 0]], "float32")
self.assertAllClose(self.evaluate(math_ops.reduce_std(x)), 0)
self.assertAllClose(
self.evaluate(math_ops.reduce_std(x, axis=0)), [0, 0, 0])
x = np.array([[1, 2, 1, 1], [1, 1, 0, 1]], "float32")
self.assertAllClose(self.evaluate(math_ops.reduce_std(x)), 0.5)
class LogSumExpTest(test_util.TensorFlowTestCase):
@test_util.run_deprecated_v1
def testReduceLogSumExp(self):
for dtype in [np.float16, np.float32, np.double]:
x_np = np.random.rand(5, 5).astype(dtype)
with self.cached_session(use_gpu=True):
y_tf_np = math_ops.reduce_logsumexp(x_np).eval()
y_np = log(np.sum(exp(x_np)))
self.assertAllClose(y_tf_np, y_np)
def testReductionIndices(self):
for dtype in [np.float16, np.float32, np.double]:
x_np = np.random.rand(5, 5).astype(dtype)
with self.cached_session(use_gpu=True):
y_tf = math_ops.reduce_logsumexp(x_np, axis=[0])
y_np = log(np.sum(exp(x_np), axis=0))
self.assertShapeEqual(y_np, y_tf)
y_tf_np = self.evaluate(y_tf)
self.assertAllClose(y_tf_np, y_np)
def testReductionIndices2(self):
for dtype in [np.float16, np.float32, np.double]:
x_np = np.random.rand(5, 5).astype(dtype)
with self.cached_session(use_gpu=True):
y_tf = math_ops.reduce_logsumexp(x_np, axis=0)
y_np = log(np.sum(exp(x_np), axis=0))
self.assertShapeEqual(y_np, y_tf)
y_tf_np = self.evaluate(y_tf)
self.assertAllClose(y_tf_np, y_np)
@test_util.run_deprecated_v1
def testKeepDims(self):
for dtype in [np.float16, np.float32, np.double]:
x_np = np.random.rand(5, 5).astype(dtype)
with self.cached_session(use_gpu=True):
y_tf_np = math_ops.reduce_logsumexp(x_np, keepdims=True).eval()
self.assertEqual(y_tf_np.ndim, x_np.ndim)
y_np = log(np.sum(exp(x_np), keepdims=True))
self.assertAllClose(y_tf_np, y_np)
@test_util.run_deprecated_v1
def testOverflow(self):
x = [1000, 1001, 1002, 1003]
for dtype in [np.float16, np.float32, np.double]:
x_np = np.array(x, dtype=dtype)
max_np = np.max(x_np)
with self.assertRaisesRegexp(RuntimeWarning,
"overflow encountered in exp"):
out = log(np.sum(exp(x_np)))
if out == np.inf:
raise RuntimeWarning("overflow encountered in exp")
with self.cached_session(use_gpu=True):
x_tf = constant_op.constant(x_np, shape=x_np.shape)
y_tf_np = math_ops.reduce_logsumexp(x_tf).eval()
y_np = log(np.sum(exp(x_np - max_np))) + max_np
self.assertAllClose(y_tf_np, y_np)
@test_util.run_deprecated_v1
def testUnderflow(self):
x = [-1000, -1001, -1002, -1003]
for dtype in [np.float16, np.float32, np.double]:
x_np = np.array(x, dtype=dtype)
max_np = np.max(x_np)
with self.assertRaisesRegexp(RuntimeWarning,
"divide by zero encountered in log"):
out = log(np.sum(exp(x_np)))
if out == -np.inf:
raise RuntimeWarning("divide by zero encountered in log")
with self.cached_session(use_gpu=True):
x_tf = constant_op.constant(x_np, shape=x_np.shape)
y_tf_np = math_ops.reduce_logsumexp(x_tf).eval()
y_np = log(np.sum(exp(x_np - max_np))) + max_np
self.assertAllClose(y_tf_np, y_np)
@test_util.run_deprecated_v1
def testInfinity(self):
with self.session(use_gpu=True):
res = math_ops.reduce_logsumexp(-np.inf).eval()
self.assertEqual(-np.inf, res)
class RoundTest(test_util.TensorFlowTestCase):
@test_util.run_in_graph_and_eager_modes
def testRounding(self):
x = np.arange(-5.0, 5.0, .25)
for dtype in [np.float32, np.double, np.int32]:
x_np = np.array(x, dtype=dtype)
with test_util.device(use_gpu=True):
x_tf = constant_op.constant(x_np, shape=x_np.shape)
y_tf = math_ops.round(x_tf)
y_tf_np = self.evaluate(y_tf)
y_np = np.round(x_np)
self.assertAllClose(y_tf_np, y_np, atol=1e-2)
class ModTest(test_util.TensorFlowTestCase):
@test_util.run_deprecated_v1
def testFloat(self):
x = [0.5, 0.7, 0.3]
for dtype in [np.float32, np.double]:
# Test scalar and vector versions.
for denom in [x[0], [x[0]] * 3]:
x_np = np.array(x, dtype=dtype)
with self.cached_session(use_gpu=True):
x_tf = constant_op.constant(x_np, shape=x_np.shape)
y_tf = math_ops.mod(x_tf, denom)
y_tf_np = self.evaluate(y_tf)
y_np = np.fmod(x_np, denom)
self.assertAllClose(y_tf_np, y_np, atol=1e-2)
def testFixed(self):
x = [5, 10, 23]
for dtype in [np.int32, np.int64]:
# Test scalar and vector versions.
for denom in [x[0], x]:
x_np = np.array(x, dtype=dtype)
with self.cached_session(use_gpu=True):
x_tf = constant_op.constant(x_np, shape=x_np.shape)
y_tf = math_ops.mod(x_tf, denom)
y_tf_np = self.evaluate(y_tf)
y_np = np.mod(x_np, denom)
self.assertAllClose(y_tf_np, y_np)
class SquaredDifferenceTest(test_util.TensorFlowTestCase):
@test_util.run_in_graph_and_eager_modes
def testSquaredDifference(self):
for dtype in [np.float16, np.float32, np.float64, np.int32, np.int64]:
x = np.array([[1, 2, 3], [4, 5, 6]], dtype=dtype)
y = np.array([-3, -2, -1], dtype=dtype)
z = (x - y) * (x - y)
with test_util.device(use_gpu=True):
z_tf = self.evaluate(math_ops.squared_difference(x, y))
self.assertAllClose(z, z_tf)
@test_util.run_in_graph_and_eager_modes()
def testComplexSquaredDifference(self):
for dtype in [np.complex64, np.complex128]:
x = np.array([[1 + 3j, 2 + 2j, 3 + 1j], [4 - 1j, 5 - 2j, 6 - 3j]],
dtype=dtype)
y = np.array([-3 + 1j, -2 + 2j, -1 + 3j], dtype=dtype)
z = np.conj(x - y) * (x - y)
with test_util.device(use_gpu=False):
z_tf = self.evaluate(math_ops.squared_difference(x, y))
self.assertAllClose(z, z_tf)
class ApproximateEqualTest(test_util.TensorFlowTestCase):
@test_util.run_in_graph_and_eager_modes
def testApproximateEqual(self):
for dtype in [np.float32, np.double]:
x = dtype(1)
y = dtype(1.00009)
z = False
with test_util.device(use_gpu=True):
# Default tolerance is 0.00001
z_tf = self.evaluate(math_ops.approximate_equal(x, y))
self.assertAllEqual(z, z_tf)
for dtype in [np.float32, np.double]:
x = dtype(1)
y = dtype(1.000009)
z = True
with test_util.device(use_gpu=True):
# Default tolerance is 0.00001
z_tf = self.evaluate(math_ops.approximate_equal(x, y))
self.assertAllEqual(z, z_tf)
for dtype in [np.float32, np.double]:
x = np.array([[[[-1, 2.00009999], [-3, 4.01]]]], dtype=dtype)
y = np.array([[[[-1.001, 2], [-3.00009, 4]]]], dtype=dtype)
z = np.array([[[[False, True], [True, False]]]], dtype=np.bool)
with test_util.device(use_gpu=True):
z_tf = self.evaluate(math_ops.approximate_equal(x, y, tolerance=0.0001))
self.assertAllEqual(z, z_tf)
@test_util.run_deprecated_v1
def testApproximateEqualShape(self):
for dtype in [np.float32, np.double]:
x = np.array([1, 2], dtype=dtype)
y = np.array([[1, 2]], dtype=dtype)
# The inputs 'x' and 'y' must have the same shape.
with self.assertRaisesRegexp(
ValueError, "Shapes must be equal rank, but are 1 and 2"):
math_ops.approximate_equal(x, y)
class ScalarMulTest(test_util.TensorFlowTestCase):
@test_util.run_in_graph_and_eager_modes
def testAcceptsRefs(self):
if context.executing_eagerly():
var = resource_variable_ops.ResourceVariable(10, name="var")
else:
var = variables.Variable(10)
result = math_ops.scalar_mul(3, var)
init = variables.global_variables_initializer()
with test_util.device(use_gpu=True):
self.evaluate(init)
self.assertEqual(30, self.evaluate(result))
@test_util.run_in_graph_and_eager_modes
def testAcceptsConstant(self):
const = constant_op.constant(10)
result = math_ops.scalar_mul(3, const)
with test_util.device(use_gpu=True):
self.assertEqual(30, self.evaluate(result))
@test_util.run_in_graph_and_eager_modes
def testAcceptsTensor(self):
tensor = array_ops.ones([10, 10])
result = math_ops.scalar_mul(3, tensor)
expected = array_ops.ones([10, 10]) * 3
with test_util.device(use_gpu=True):
self.assertAllEqual(self.evaluate(expected), self.evaluate(result))
@test_util.run_in_graph_and_eager_modes
def testAcceptsIndexedSlices(self):
values = constant_op.constant([2, 3, 5, 7, 0, -1], shape=[3, 2])
indices = constant_op.constant([0, 2, 5])
x = math_ops.scalar_mul(-3, ops.IndexedSlices(values, indices))
with test_util.device(use_gpu=True):
self.assertAllEqual(self.evaluate(x.values),
[[-6, -9], [-15, -21], [0, 3]])
self.assertAllEqual(self.evaluate(x.indices), [0, 2, 5])
class AccumulateNTest(test_util.TensorFlowTestCase):
@test_util.run_deprecated_v1
def testFloat(self):
np.random.seed(12345)
x = [np.random.random((1, 2, 3, 4, 5)) - 0.5 for _ in range(5)]
tf_x = ops.convert_n_to_tensor(x)
with self.session(use_gpu=True):
self.assertAllClose(sum(x), math_ops.accumulate_n(tf_x).eval())
self.assertAllClose(x[0] * 5, math_ops.accumulate_n([tf_x[0]] * 5).eval())
@test_util.run_deprecated_v1
def testInt(self):
np.random.seed(54321)
x = [np.random.randint(-128, 128, (5, 4, 3, 2, 1)) for _ in range(6)]
tf_x = ops.convert_n_to_tensor(x)
with self.session(use_gpu=True):
self.assertAllEqual(sum(x), math_ops.accumulate_n(tf_x).eval())
self.assertAllEqual(x[0] * 6, math_ops.accumulate_n([tf_x[0]] * 6).eval())
class AddNTest(test_util.TensorFlowTestCase):
@test_util.run_deprecated_v1
def testPartials(self):
"""Test that previously revealed a bug in buffer forwarding for AddN."""
partials = []
for _ in range(98):
partials.append(math_ops.add_n([constant_op.constant(1)]))
partials.append(
math_ops.add_n([constant_op.constant(1),
constant_op.constant(1)]))
res = math_ops.add_n(partials) + constant_op.constant(0)
with self.session(use_gpu=True):
self.assertAllEqual(res.eval(), 100)
@test_util.run_deprecated_v1
def testFloat(self):
np.random.seed(12345)
for num_inputs in range(1, 10):
x = [np.random.random((1, 2, 3, 4, 5)) - 0.5 for _ in range(num_inputs)]
tf_x = ops.convert_n_to_tensor(x)
with self.cached_session(use_gpu=True):
self.assertAllClose(sum(x), math_ops.add_n(tf_x).eval())
self.assertAllClose(x[0] * num_inputs,
math_ops.add_n([tf_x[0]] * num_inputs).eval())
@test_util.run_deprecated_v1
def testInt(self):
np.random.seed(54321)
for num_inputs in range(1, 10):
x = [
np.random.randint(-128, 128, (5, 4, 3, 2, 1))
for _ in range(num_inputs)
]
tf_x = ops.convert_n_to_tensor(x)
with self.cached_session(use_gpu=True):
self.assertAllEqual(sum(x), math_ops.add_n(tf_x).eval())
self.assertAllEqual(x[0] * num_inputs,
math_ops.add_n([tf_x[0]] * num_inputs).eval())
@test_util.run_deprecated_v1
def testGrad(self):
np.random.seed(42)
for num_inputs in range(1, 10):
with self.cached_session(use_gpu=True) as sess:
input_vars = [
variables.Variable(10.0 * np.random.random())
for i in range(0, num_inputs)
]
addn = math_ops.add_n(input_vars)
self.evaluate(variables.global_variables_initializer())
add_n_grad = gradients.gradients(addn, input_vars)
self.assertAllEqual(np.repeat(1.0, num_inputs), # d/dx (x + y + ...) = 1
[g.eval() for g in add_n_grad])
@test_util.run_deprecated_v1
def testIndexedSlices(self):
slc = ops.IndexedSlices(
array_ops.constant([1, 2], shape=[1, 2]), array_ops.constant([1]),
array_ops.constant([2, 2]))
slc_as_dense = np.array([[0, 0], [1, 2]])
with self.test_session(use_gpu=True):
# add_n currently always converts IndexedSlices to dense
self.assertAllEqual(slc_as_dense, math_ops.add_n([slc]).eval())
self.assertAllEqual(2 * slc_as_dense, math_ops.add_n([slc, slc]).eval())
class DivAndModTest(test_util.TensorFlowTestCase):
# TODO(aselle): Test more types before exposing new division operators.
def intTestData(self):
nums = np.arange(-10, 10, 1).reshape(20, 1)
divs = np.arange(-3, 4, 2).reshape(1, 4)
return nums, divs
def floatTestData(self):
nums = np.arange(-10, 10, .25).reshape(80, 1)
divs = np.arange(-3, 0, .25).reshape(1, 12)
return nums, divs
@test_util.run_deprecated_v1
def testFloorModInt(self):
nums, divs = self.intTestData()
with self.cached_session():
# TODO(aselle): Change test to use % after switch
# tf_result = math_ops.floor_mod(nums, divs).eval()
tf_result = math_ops.floormod(nums, divs).eval()
np_result = nums % divs
self.assertAllEqual(tf_result, np_result)
@test_util.run_deprecated_v1
def testFloorModFloat(self):
nums, divs = self.floatTestData()
with self.cached_session():
tf_result = math_ops.floormod(nums, divs).eval()
np_result = nums % divs
self.assertAllEqual(tf_result, np_result)
# TODO(aselle): put this test in once % switched to floormod
# tf2_result = (array_ops.constant(nums)
# % array_ops.constant(divs)).eval()
# self.assertAllEqual(tf2_result, tf_result)
@test_util.run_deprecated_v1
def testTruncateModInt(self):
nums, divs = self.intTestData()
with self.cached_session():
tf_result = math_ops.truncatemod(nums, divs).eval()
np_result = np.fmod(nums, divs)
self.assertAllEqual(tf_result, np_result)
@test_util.run_deprecated_v1
def testTruncateModFloat(self):
nums, divs = self.floatTestData()
with self.cached_session():
tf_result = math_ops.truncatemod(nums, divs).eval()
np_result = np.fmod(nums, divs)
self.assertAllEqual(tf_result, np_result)
@test_util.run_deprecated_v1
def testDivideInt(self):
nums, divs = self.intTestData()
with self.cached_session():
tf_result = math_ops.floor_div(nums, divs).eval()
np_result = nums // divs
self.assertAllEqual(tf_result, np_result)
# TODO(aselle): Put this test in once // is switched to floordiv
# tf2_result = (array_ops.constant(nums)
# // array_ops.constant(divs)).eval()
# self.assertAllEqual(tf2_result, tf_result)
@test_util.run_deprecated_v1
def testDivideName(self):
with self.cached_session():
op = math_ops.divide(
array_ops.constant(3), array_ops.constant(4), name="my_cool_divide")
self.assertEqual(op.name, "my_cool_divide:0")
@test_util.run_deprecated_v1
def testRealDiv(self):
nums, divs = self.floatTestData()
with self.cached_session():
tf_result = math_ops.realdiv(nums, divs).eval()
np_result = np.divide(nums, divs)
self.assertAllEqual(tf_result, np_result)
@test_util.run_deprecated_v1
def testComplexDiv(self):
foo = array_ops.constant([1. + 3.j])
with self.cached_session():
_ = math_ops.divide(foo, 1.).eval()
_ = math_ops.div(foo, 2.).eval()
@test_util.run_deprecated_v1
def testFloorDivGrad(self):
with self.cached_session():
a = variables.Variable(2.)
b = variables.Variable(4.)
with self.cached_session() as sess:
self.evaluate(variables.global_variables_initializer())
c_grad = gradients.gradients(math_ops.divide(a, b), [a, b])
self.assertAllEqual([x.eval() for x in c_grad], [.25, -.125])
c_grad = gradients.gradients(math_ops.div(a, b), [a, b])
self.assertAllEqual([x.eval() for x in c_grad], [.25, -.125])
c_grad = gradients.gradients(math_ops.floordiv(a, b), [a, b])
self.assertAllEqual(
[None if x is None else self.evaluate(x) for x in c_grad],
[None, None])
@test_util.run_deprecated_v1
def testConsistent(self):
nums, divs = self.intTestData()
with self.cached_session():
tf_result = (math_ops.floor_div(nums, divs) * divs + math_ops.floormod(
nums, divs)).eval()
tf_nums = array_ops.constant(nums)
tf_divs = array_ops.constant(divs)
tf2_result = (tf_nums // tf_divs * tf_divs + tf_nums % tf_divs).eval()
np_result = (nums // divs) * divs + (nums % divs)
# Consistent with numpy
self.assertAllEqual(tf_result, np_result)
# Consistent with two forms of divide
self.assertAllEqual(tf_result, tf2_result)
# consistency for truncation form
tf3_result = (math_ops.truncatediv(nums, divs) * divs +
math_ops.truncatemod(nums, divs)).eval()
expanded_nums = np.reshape(
np.tile(nums, divs.shape[1]), (nums.shape[0], divs.shape[1]))
# Consistent with desire to get numerator
self.assertAllEqual(tf3_result, expanded_nums)
# Consistent with desire to get numerator
self.assertAllEqual(tf_result, expanded_nums)
class DivNoNanTest(test_util.TensorFlowTestCase):
@test_util.run_deprecated_v1
def testBasic(self):
for dtype in [np.float32, np.float64]:
nums = np.arange(-10, 10, .25, dtype=dtype).reshape(80, 1)
divs = np.arange(-3, 3, .25, dtype=dtype).reshape(1, 24)
np_result = np.true_divide(nums, divs)
np_result[:, divs[0] == 0] = 0
with self.cached_session(use_gpu=True):
tf_result = math_ops.div_no_nan(nums, divs).eval()
self.assertAllEqual(tf_result, np_result)
class MultiplyNoNanTest(test_util.TensorFlowTestCase):
@test_util.run_deprecated_v1
def testBasic(self):
for dtype in [np.float32, np.float64]:
values = [0, 1, np.nan, np.inf, np.NINF]
x = constant_op.constant(values, dtype=dtype)
zeros = constant_op.constant(np.zeros((5,)), dtype=dtype)
ones = constant_op.constant(np.ones((5,)), dtype=dtype)
with self.cached_session(use_gpu=True):
tf_result_zeros = math_ops.multiply_no_nan(x, zeros).eval()
self.assertAllEqual(tf_result_zeros, zeros)
tf_result_ones = math_ops.multiply_no_nan(x, ones).eval()
self.assertAllEqual(tf_result_ones, x)
class XlogyTest(test_util.TensorFlowTestCase):
@test_util.run_in_graph_and_eager_modes
def testXlogyNoZero(self):
for dtype in [dtypes.float16, dtypes.float32, dtypes.float64]:
x = constant_op.constant([[0.1, 0.2, 3.5], [-2., -5., 30.]], dtype=dtype)
y = constant_op.constant([[0.1, 0.2, 3.5], [3.1, 4., 2.]], dtype=dtype)
with self.cached_session(use_gpu=True):
xlogy = self.evaluate(math_ops.xlogy(x, y))
xtimeslogy = self.evaluate(x * math_ops.log(y))
self.assertAllClose(xlogy, xtimeslogy)
@test_util.run_in_graph_and_eager_modes
def testXlogyWithZero(self):
for dtype in [dtypes.float16, dtypes.float32, dtypes.float64]:
x = constant_op.constant(np.zeros((2, 3)), dtype=dtype)
y = constant_op.constant([[0.1, 0.2, 3.5], [0., 1., 2.]], dtype=dtype)
with self.cached_session(use_gpu=True):
xlogy_tf_np = self.evaluate(math_ops.xlogy(x, y))
zeros_np = self.evaluate(array_ops.zeros_like(y))
self.assertAllClose(xlogy_tf_np, zeros_np)
@test_util.run_in_graph_and_eager_modes
def testXlogyWithZeroBroadcast(self):
for dtype in [dtypes.float16, dtypes.float32, dtypes.float64]:
x = constant_op.constant([[0.], [1.]], dtype=dtype)
y = constant_op.constant([[0.1, 0.2, 3.5], [0., 1., 2.]], dtype=dtype)
with self.cached_session(use_gpu=True):
xlogy_tf_np = self.evaluate(math_ops.xlogy(x, y))
zeros_np = self.evaluate(array_ops.zeros_like(y[0]))
xtimes_logy = self.evaluate(math_ops.log(y[1]))
self.assertAllClose(zeros_np, xlogy_tf_np[0])
self.assertAllClose(xtimes_logy, xlogy_tf_np[1])
class XdivyTest(test_util.TensorFlowTestCase):
@test_util.run_in_graph_and_eager_modes
def testXdivyNoZero(self):
for dtype in [dtypes.float16, dtypes.float32, dtypes.float64]:
x = constant_op.constant([[0.1, 0.2, 3.5], [-2., -5., 30.]], dtype=dtype)
y = constant_op.constant([[0.1, 0.2, 3.5], [3.1, 4., 2.]], dtype=dtype)
with self.cached_session(use_gpu=True):
xdivy = self.evaluate(math_ops.xdivy(x, y))
x_over_y = self.evaluate(x / y)
self.assertAllClose(xdivy, x_over_y)
@test_util.run_in_graph_and_eager_modes
def testXdivyWithZero(self):
for dtype in [dtypes.float16, dtypes.float32, dtypes.float64]:
x = constant_op.constant(np.zeros((2, 3)), dtype=dtype)
y = constant_op.constant([[0.1, 0.2, 3.5], [0., 1., 2.]], dtype=dtype)
with self.cached_session(use_gpu=True):
xdivy_tf_np = self.evaluate(math_ops.xdivy(x, y))
zeros_np = self.evaluate(array_ops.zeros_like(y))
self.assertAllClose(xdivy_tf_np, zeros_np)
@test_util.run_in_graph_and_eager_modes
def testXdivyWithZeroBroadcast(self):
for dtype in [dtypes.float16, dtypes.float32, dtypes.float64]:
x = constant_op.constant([[0.], [1.]], dtype=dtype)
y = constant_op.constant([[0.1, 0.2, 3.5], [0., 1., 2.]], dtype=dtype)
with self.cached_session(use_gpu=True):
xdivy_tf_np = self.evaluate(math_ops.xdivy(x, y))
zeros_np = self.evaluate(array_ops.zeros_like(y[0]))
x_over_y = self.evaluate(1 / y[1])
self.assertAllClose(zeros_np, xdivy_tf_np[0])
self.assertAllClose(x_over_y, xdivy_tf_np[1])
class NextAfterTest(test_util.TensorFlowTestCase):
# Basic NextAfter tests that replicate numpy nextafter tests.
@test_util.run_in_graph_and_eager_modes
def testBasic(self):
for dtype in [dtypes.float32, dtypes.float64]:
one = constant_op.constant([1], dtype=dtype)
two = constant_op.constant([2], dtype=dtype)
zero = constant_op.constant([0], dtype=dtype)
nan = constant_op.constant([np.nan], dtype=dtype)
eps = constant_op.constant([np.finfo(dtype.as_numpy_dtype).eps],
dtype=dtype)
self.assertAllEqual(math_ops.nextafter(one, two) - one, eps)
self.assertAllLess(math_ops.nextafter(one, zero) - one, 0)
self.assertAllEqual(
math_ops.is_nan(math_ops.nextafter(nan, one)), [True])
self.assertAllEqual(
math_ops.is_nan(math_ops.nextafter(one, nan)), [True])
self.assertAllEqual(math_ops.nextafter(one, one), one)
@test_util.run_in_graph_and_eager_modes
def testBroadcasting(self):
for dtype in [dtypes.float32, dtypes.float64]:
one = constant_op.constant([1, 1], dtype=dtype)
two = constant_op.constant([2], dtype=dtype)
eps = np.finfo(dtype.as_numpy_dtype).eps
eps_const = constant_op.constant([eps, eps], dtype=dtype)
self.assertAllEqual(math_ops.nextafter(one, two) - one, eps_const)
class BinaryOpsTest(test_util.TensorFlowTestCase):
@test_util.run_in_graph_and_eager_modes
def testErrorReceivedIfDtypeMismatchFromOp(self):
if context.executing_eagerly():
error = errors_impl.InvalidArgumentError
error_message = (
r"cannot compute Add(V2)? as input #1\(zero-based\) was expected to "
r"be a int32 tensor but is a float tensor \[Op:Add(V2)?\] name: add/")
else:
error = TypeError
error_message = (
"Input 'y' of 'Add(V2)?' Op has type float32 that does not "
"match type int32 of argument 'x'.")
with self.assertRaisesRegexp(error, error_message):
a = array_ops.ones([1], dtype=dtypes.int32) + 1.0
self.evaluate(a)
if __name__ == "__main__":
googletest.main()
|
tensorflow-master
|
tensorflow/python/ops/math_ops_test.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for bitwise operations."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import six
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import test_util
from tensorflow.python.ops import bitwise_ops
from tensorflow.python.ops import gen_bitwise_ops
from tensorflow.python.platform import googletest
class BitwiseOpTest(test_util.TensorFlowTestCase):
def __init__(self, method_name="runTest"):
super(BitwiseOpTest, self).__init__(method_name)
@test_util.run_deprecated_v1
def testBinaryOps(self):
dtype_list = [dtypes.int8, dtypes.int16, dtypes.int32, dtypes.int64,
dtypes.uint8, dtypes.uint16, dtypes.uint32, dtypes.uint64]
with self.session(use_gpu=True) as sess:
for dtype in dtype_list:
lhs = constant_op.constant([0, 5, 3, 14], dtype=dtype)
rhs = constant_op.constant([5, 0, 7, 11], dtype=dtype)
and_result, or_result, xor_result = sess.run(
[bitwise_ops.bitwise_and(lhs, rhs),
bitwise_ops.bitwise_or(lhs, rhs),
bitwise_ops.bitwise_xor(lhs, rhs)])
self.assertAllEqual(and_result, [0, 0, 3, 10])
self.assertAllEqual(or_result, [5, 5, 7, 15])
self.assertAllEqual(xor_result, [5, 5, 4, 5])
def testPopulationCountOp(self):
dtype_list = [dtypes.int8, dtypes.int16,
dtypes.int32, dtypes.int64,
dtypes.uint8, dtypes.uint16]
raw_inputs = [0, 1, -1, 3, -3, 5, -5, 14, -14,
127, 128, 255, 256, 65535, 65536,
2**31 - 1, 2**31, 2**32 - 1, 2**32, -2**32 + 1, -2**32,
-2**63 + 1, 2**63 - 1]
def count_bits(x):
return sum(bin(z).count("1") for z in six.iterbytes(x.tobytes()))
for dtype in dtype_list:
with self.cached_session(use_gpu=True) as sess:
print("PopulationCount test: ", dtype)
inputs = np.array(raw_inputs, dtype=dtype.as_numpy_dtype)
truth = [count_bits(x) for x in inputs]
input_tensor = constant_op.constant(inputs, dtype=dtype)
popcnt_result = self.evaluate(
gen_bitwise_ops.population_count(input_tensor))
self.assertAllEqual(truth, popcnt_result)
@test_util.run_deprecated_v1
def testInvertOp(self):
dtype_list = [dtypes.int8, dtypes.int16, dtypes.int32, dtypes.int64,
dtypes.uint8, dtypes.uint16, dtypes.uint32, dtypes.uint64]
inputs = [0, 5, 3, 14]
with self.session(use_gpu=True) as sess:
for dtype in dtype_list:
# Because of issues with negative numbers, let's test this indirectly.
# 1. invert(a) and a = 0
# 2. invert(a) or a = invert(0)
input_tensor = constant_op.constant(inputs, dtype=dtype)
not_a_and_a, not_a_or_a, not_0 = sess.run(
[bitwise_ops.bitwise_and(
input_tensor, bitwise_ops.invert(input_tensor)),
bitwise_ops.bitwise_or(
input_tensor, bitwise_ops.invert(input_tensor)),
bitwise_ops.invert(constant_op.constant(0, dtype=dtype))])
self.assertAllEqual(not_a_and_a, [0, 0, 0, 0])
self.assertAllEqual(not_a_or_a, [not_0] * 4)
# For unsigned dtypes let's also check the result directly.
if dtype.is_unsigned:
inverted = self.evaluate(bitwise_ops.invert(input_tensor))
expected = [dtype.max - x for x in inputs]
self.assertAllEqual(inverted, expected)
@test_util.run_deprecated_v1
def testShiftsWithPositiveLHS(self):
dtype_list = [np.int8, np.int16, np.int32, np.int64,
np.uint8, np.uint16, np.uint32, np.uint64]
with self.session(use_gpu=True) as sess:
for dtype in dtype_list:
lhs = np.array([0, 5, 3, 14], dtype=dtype)
rhs = np.array([5, 0, 7, 3], dtype=dtype)
left_shift_result, right_shift_result = sess.run(
[bitwise_ops.left_shift(lhs, rhs),
bitwise_ops.right_shift(lhs, rhs)])
self.assertAllEqual(left_shift_result, np.left_shift(lhs, rhs))
self.assertAllEqual(right_shift_result, np.right_shift(lhs, rhs))
@test_util.run_deprecated_v1
def testShiftsWithNegativeLHS(self):
dtype_list = [np.int8, np.int16, np.int32, np.int64]
with self.session(use_gpu=True) as sess:
for dtype in dtype_list:
lhs = np.array([-1, -5, -3, -14], dtype=dtype)
rhs = np.array([5, 0, 7, 11], dtype=dtype)
left_shift_result, right_shift_result = sess.run(
[bitwise_ops.left_shift(lhs, rhs),
bitwise_ops.right_shift(lhs, rhs)])
self.assertAllEqual(left_shift_result, np.left_shift(lhs, rhs))
self.assertAllEqual(right_shift_result, np.right_shift(lhs, rhs))
@test_util.run_deprecated_v1
def testImplementationDefinedShiftsDoNotCrash(self):
dtype_list = [np.int8, np.int16, np.int32, np.int64]
with self.session(use_gpu=True) as sess:
for dtype in dtype_list:
lhs = np.array([-1, -5, -3, -14], dtype=dtype)
rhs = np.array([-2, 64, 101, 32], dtype=dtype)
# We intentionally do not test for specific values here since the exact
# outputs are implementation-defined. However, we should not crash or
# trigger an undefined-behavior error from tools such as
# AddressSanitizer.
sess.run([bitwise_ops.left_shift(lhs, rhs),
bitwise_ops.right_shift(lhs, rhs)])
@test_util.run_deprecated_v1
def testShapeInference(self):
dtype_list = [dtypes.int8, dtypes.int16, dtypes.int32, dtypes.int64,
dtypes.uint8, dtypes.uint16]
with self.session(use_gpu=True) as sess:
for dtype in dtype_list:
lhs = constant_op.constant([[0], [3], [5]], dtype=dtype)
rhs = constant_op.constant([[1, 2, 4]], dtype=dtype)
and_tensor = bitwise_ops.bitwise_and(lhs, rhs)
or_tensor = bitwise_ops.bitwise_or(lhs, rhs)
xor_tensor = bitwise_ops.bitwise_xor(lhs, rhs)
ls_tensor = bitwise_ops.left_shift(lhs, rhs)
rs_tensor = bitwise_ops.right_shift(lhs, rhs)
and_result, or_result, xor_result, ls_result, rs_result = sess.run(
[and_tensor, or_tensor, xor_tensor, ls_tensor, rs_tensor])
# Compare shape inference with result
self.assertAllEqual(and_tensor.get_shape().as_list(), and_result.shape)
self.assertAllEqual(and_tensor.get_shape().as_list(), [3, 3])
self.assertAllEqual(or_tensor.get_shape().as_list(), or_result.shape)
self.assertAllEqual(or_tensor.get_shape().as_list(), [3, 3])
self.assertAllEqual(xor_tensor.get_shape().as_list(), xor_result.shape)
self.assertAllEqual(xor_tensor.get_shape().as_list(), [3, 3])
self.assertAllEqual(ls_tensor.get_shape().as_list(), ls_result.shape)
self.assertAllEqual(ls_tensor.get_shape().as_list(), [3, 3])
self.assertAllEqual(rs_tensor.get_shape().as_list(), rs_result.shape)
self.assertAllEqual(rs_tensor.get_shape().as_list(), [3, 3])
if __name__ == "__main__":
googletest.main()
|
tensorflow-master
|
tensorflow/python/ops/bitwise_ops_test.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Functional tests for quantized convolutional operations."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.ops import nn_ops
from tensorflow.python.platform import test
class Conv2DTest(test.TestCase):
def __init__(self, method_name="runTest"):
super(Conv2DTest, self).__init__(method_name)
def _VerifyValues(self, tensor_in_sizes, filter_in_sizes, stride, padding,
expected):
"""Verifies the output values of the convolution function.
Args:
tensor_in_sizes: Input tensor dimensions in
[batch, input_rows, input_cols, input_depth].
filter_in_sizes: Filter tensor dimensions in
[kernel_rows, kernel_cols, input_depth, output_depth].
stride: Stride.
padding: Padding type.
expected: An array containing the expected operation outputs.
"""
total_size_1 = 1
total_size_2 = 1
for s in tensor_in_sizes:
total_size_1 *= s
for s in filter_in_sizes:
total_size_2 *= s
# Initializes the input tensor with array containing incrementing
# numbers from 1.
x1 = np.array([f for f in range(1, total_size_1 + 1)])
x1 = x1.astype(np.uint8).reshape(tensor_in_sizes)
x1_min = 0.0
x1_max = 255.0
x2 = np.array([f for f in range(1, total_size_2 + 1)]).astype(np.uint8)
x2 = x2.astype(np.uint8).reshape(filter_in_sizes)
x2_min = 0.0
x2_max = 255.0
with self.cached_session(use_gpu=False) as sess:
t1 = constant_op.constant(x1, shape=tensor_in_sizes, dtype=dtypes.quint8)
t2 = constant_op.constant(x2, shape=filter_in_sizes, dtype=dtypes.quint8)
conv = nn_ops.quantized_conv2d(
t1,
t2,
out_type=dtypes.qint32,
strides=[1, stride, stride, 1],
padding=padding,
min_input=x1_min,
max_input=x1_max,
min_filter=x2_min,
max_filter=x2_max)
value = self.evaluate(conv)
quantized_output = value[0]
output_min = value[1]
output_max = value[2]
float_output = self._QuantizedOutputToFloat(quantized_output, output_min,
output_max)
self.assertArrayNear(expected, float_output.flatten(), 1.0)
self.assertEqual(value[0].shape, conv[0].get_shape())
def _assertQuantizedArrayEquals(self, iarray1, iarray2):
for i1, i2 in zip(iarray1, iarray2):
self.assertTrue(i1 == i2)
def _QuantizedOutputToFloat(self, quantized, quantized_min, quantized_max):
number_of_bits = 32
number_of_steps = 1 << number_of_bits
range_adjust = (number_of_steps / (number_of_steps - 1.0))
quantized_range = ((quantized_max - quantized_min) * range_adjust)
range_scale = (quantized_range / number_of_steps)
lowest_quantized = -(1 << (number_of_bits - 1))
result = np.array([(quantized_min +
((float(x) - lowest_quantized) * range_scale))
for x in quantized.flatten()])
return result
def testConv2D1x1Filter(self):
# Our generated input is [batch, rows, cols, depth], and looks like this:
# (1,2,3) (4,5,6) (7,8,9)
# (10,11,12) (13,14,15) (16,17,18)
# The filter data is:
# (1,4,7) (2,5,8) (3,6,9)
# That means the calculations are:
# 1*1+2*4+3*7=30
# 1*2+2*5+3*8=36
# 1*3+2*6+3*9=42
# 4*1+5*4+6*7=66
# 4*2+5*5+6*8=81
# 4*3+5*6+6*9=96
# 7*1+5*8+6*9=102
# 7*2+8*5+9*8=126
# 7*3+8*6+9*9=150
# 10*1+11*4+12*7=138
# 10*2+11*5+12*8=171
# 10*3+11*6+12*9=204
# 13*1+14*4+15*7=174
# 13*2+14*5+15*8=216
# 13*3+14*6+15*9=258, clamped to 255
# 16*1+17*4+18*7=210
# 16*2+17*5+18*8=261, clamped to 255
# 16*3+17*6+18*9=312, clamped to 255
# Because the output shift is zero, we call the non-optimized reference
# path for the convolution.
expected_output = [
30, 36, 42, 66, 81, 96, 102, 126, 150, 138, 171, 204, 174, 216, 258,
210, 261, 312
]
self._VerifyValues(
tensor_in_sizes=[1, 2, 3, 3],
filter_in_sizes=[1, 1, 3, 3],
stride=1,
padding="VALID",
expected=expected_output)
def testConv2D2x2Filter(self):
# Our generated input is [batch, rows, cols, depth], and looks like this:
# (1,2,3) (4,5,6) (7,8,9)
# (10,11,12) (13,14,15) (16,17,18)
# The filter data is [filter_height, filter_width, depth, filter_count]:
# ( 1, 4, 7) (10, 13, 16)
# (19,22,25) (28, 31, 34)
# -
# ( 2, 5, 8) (11, 14, 17)
# (20,23,26) (29, 32, 35)
# -
# ( 3, 6, 9) (12, 15, 18)
# (21,24,27) (30, 33, 36)
# The raw accumulated totals are:
# 1*1+2*4+3*7+4*10+5*13+6*16+10*19+11*22+12*25+13*28+14*31+15*34=2271
# 1*2+2*5+3*8+4*11+5*14+6*17+10*20+11*23+12*26+13*29+14*32+15*35=2367
# 1*3+2*6+3*9+4*12+5*15+6*18+10*21+11*24+12*27+13*30+14*33+15*36=2463
# 4*1+5*4+6*7+7*10+8*13+9*16+13*19+14*22+15*25+16*28+17*31+18*34=2901
# 4*2+5*5+6*8+7*11+8*14+9*17+13*20+14*23+15*26+16*29+17*32+18*35=3033
# 4*3+5*6+6*9+7*12+8*15+9*18+13*21+14*24+15*27+16*30+17*33+18*36=3165
# The expected values are taken from the raw totals and rescaled to fit into
# eight bits.
expected_output = [2271.0, 2367.0, 2463.0, 2901.0, 3033.0, 3165.0]
self._VerifyValues(
tensor_in_sizes=[1, 2, 3, 3],
filter_in_sizes=[2, 2, 3, 3],
stride=1,
padding="VALID",
expected=expected_output)
def testConv2D1x2Filter(self):
# The outputs are computed using third_party/py/IPython/notebook.
# With a shift of 21, we should execute the optimized path here.
expected_output = [
231.0, 252.0, 273.0, 384.0, 423.0, 462.0, 690.0, 765.0, 840.0, 843.0,
936.0, 1029.0
]
self._VerifyValues(
tensor_in_sizes=[1, 2, 3, 3],
filter_in_sizes=[1, 2, 3, 3],
stride=1,
padding="VALID",
expected=expected_output)
def testConv2D2x2FilterStride2(self):
# With a shift of 21, we should execute the optimized path here.
expected_output = [2271.0, 2367.0, 2463.0]
self._VerifyValues(
tensor_in_sizes=[1, 2, 3, 3],
filter_in_sizes=[2, 2, 3, 3],
stride=2,
padding="VALID",
expected=expected_output)
def testConv2D2x2FilterStride2Same(self):
# With a shift of 21, we should execute the optimized path here.
expected_output = [2271.0, 2367.0, 2463.0, 1230.0, 1305.0, 1380.0]
self._VerifyValues(
tensor_in_sizes=[1, 2, 3, 3],
filter_in_sizes=[2, 2, 3, 3],
stride=2,
padding="SAME",
expected=expected_output)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/ops/quantized_conv_ops_test.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Operations for automatic batching and unbatching."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.eager import function
from tensorflow.python.framework import ops
from tensorflow.python.framework import tensor_spec
from tensorflow.python.ops import gen_batch_ops
# pylint: disable=wildcard-import
from tensorflow.python.ops.gen_batch_ops import *
# pylint: enable=wildcard-import
from tensorflow.python.util.tf_export import tf_export
@tf_export("nondifferentiable_batch_function")
def batch_function(num_batch_threads,
max_batch_size,
batch_timeout_micros,
allowed_batch_sizes=None,
max_enqueued_batches=10,
autograph=True):
"""Batches the computation done by the decorated function.
So, for example, in the following code
```python
@batch_function(1, 2, 3)
def layer(a):
return tf.matmul(a, a)
b = layer(w)
```
if more than one session.run call is simultaneously trying to compute `b`
the values of `w` will be gathered, non-deterministically concatenated
along the first axis, and only one thread will run the computation. See the
documentation of the `Batch` op for more details.
Assumes that all arguments of the decorated function are Tensors which will
be batched along their first dimension.
SparseTensor is not supported. The return value of the decorated function
must be a Tensor or a list/tuple of Tensors.
Args:
num_batch_threads: Number of scheduling threads for processing batches
of work. Determines the number of batches processed in parallel.
max_batch_size: Batch sizes will never be bigger than this.
batch_timeout_micros: Maximum number of microseconds to wait before
outputting an incomplete batch.
allowed_batch_sizes: Optional list of allowed batch sizes. If left empty,
does nothing. Otherwise, supplies a list of batch sizes, causing the op
to pad batches up to one of those sizes. The entries must increase
monotonically, and the final entry must equal max_batch_size.
max_enqueued_batches: The maximum depth of the batch queue. Defaults to 10.
autograph: Whether to use autograph to compile python and eager style code
for efficient graph-mode execution.
Returns:
The decorated function will return the unbatched computation output Tensors.
"""
def decorator(fn): # pylint: disable=missing-docstring
def decorated(*args): # pylint: disable=missing-docstring
@function.defun(autograph=autograph)
def computation(*computation_args):
return fn(*computation_args)
computation = computation.get_concrete_function(
*[tensor_spec.TensorSpec(dtype=x.dtype, shape=x.shape, name=str(i))
for i, x in enumerate(args)])
with ops.name_scope("batch") as name:
for a in args:
if not isinstance(a, ops.Tensor):
raise ValueError("All arguments to functions decorated with "
"`batch_function` are supposed to be Tensors; "
"found %s" % repr(a))
return gen_batch_ops.batch_function(
num_batch_threads=num_batch_threads,
max_batch_size=max_batch_size,
batch_timeout_micros=batch_timeout_micros,
allowed_batch_sizes=allowed_batch_sizes,
max_enqueued_batches=max_enqueued_batches,
shared_name=name,
f=computation,
in_tensors=list(args),
captured_tensors=computation.captured_inputs,
Tout=[o.dtype for o in computation.outputs])
return decorated
return decorator
|
tensorflow-master
|
tensorflow/python/ops/batch_ops.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Operations to emit summaries."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import abc
import collections
import functools
import getpass
import os
import re
import threading
import time
import six
from tensorflow.core.framework import graph_pb2
from tensorflow.core.framework import summary_pb2
from tensorflow.core.protobuf import config_pb2
from tensorflow.python.eager import context
from tensorflow.python.eager import profiler as _profiler
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import smart_cond
from tensorflow.python.framework import tensor_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import gen_summary_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import resource_variable_ops
from tensorflow.python.ops import summary_op_util
from tensorflow.python.platform import tf_logging as logging
from tensorflow.python.training import training_util
from tensorflow.python.util import deprecation
from tensorflow.python.util import tf_contextlib
from tensorflow.python.util.tf_export import tf_export
# Name for graph collection of summary writer init ops, which is only exposed
# as a legacy API for tf.contrib.summary in TF 1.x.
_SUMMARY_WRITER_INIT_COLLECTION_NAME = "_SUMMARY_WRITER_V2"
_EXPERIMENT_NAME_PATTERNS = re.compile(r"^[^\x00-\x1F<>]{0,256}$")
_RUN_NAME_PATTERNS = re.compile(r"^[^\x00-\x1F<>]{0,512}$")
_USER_NAME_PATTERNS = re.compile(r"^[a-z]([-a-z0-9]{0,29}[a-z0-9])?$", re.I)
def _should_record_summaries_internal(default_state):
"""Returns boolean Tensor if summaries should/shouldn't be recorded.
Now the summary condition is decided by logical "and" of two conditions:
ctx.summary_recording and ctx.summary_recording_distribution_strategy. The
former one is usually set by user, and the latter one is controlled by
DistributionStrategy (tf.distribute.ReplicaContext).
Args:
default_state: can be True or False. The default summary behavior when user
does not specify ctx.summary_recording and
ctx.summary_recording_distribution_strategy is True.
"""
ctx = context.context()
resolve = lambda x: x() if callable(x) else x
cond_distributed = resolve(ctx.summary_recording_distribution_strategy)
cond = resolve(ctx.summary_recording)
if cond is None:
cond = default_state
return math_ops.logical_and(cond_distributed, cond)
def _should_record_summaries_v2():
"""Returns boolean Tensor which is true if summaries should be recorded.
If no recording status has been set, this defaults to True, unlike the public
should_record_summaries().
"""
return _should_record_summaries_internal(default_state=True)
def should_record_summaries():
"""Returns boolean Tensor which is true if summaries should be recorded."""
return _should_record_summaries_internal(default_state=False)
@tf_export("summary.record_if", v1=[])
@tf_contextlib.contextmanager
def record_if(condition):
"""Sets summary recording on or off per the provided boolean value.
The provided value can be a python boolean, a scalar boolean Tensor, or
or a callable providing such a value; if a callable is passed it will be
invoked on-demand to determine whether summary writing will occur.
Args:
condition: can be True, False, a bool Tensor, or a callable providing such.
Yields:
Returns a context manager that sets this value on enter and restores the
previous value on exit.
"""
old = context.context().summary_recording
try:
context.context().summary_recording = condition
yield
finally:
context.context().summary_recording = old
# TODO(apassos) consider how to handle local step here.
def record_summaries_every_n_global_steps(n, global_step=None):
"""Sets the should_record_summaries Tensor to true if global_step % n == 0."""
if global_step is None:
global_step = training_util.get_or_create_global_step()
with ops.device("cpu:0"):
should = lambda: math_ops.equal(global_step % n, 0)
if not context.executing_eagerly():
should = should()
return record_if(should)
def always_record_summaries():
"""Sets the should_record_summaries Tensor to always true."""
return record_if(True)
def never_record_summaries():
"""Sets the should_record_summaries Tensor to always false."""
return record_if(False)
@tf_export("summary.experimental.get_step", v1=[])
def get_step():
"""Returns the default summary step for the current thread.
Returns:
The step set by `tf.summary.experimental.set_step()` if one has been set,
otherwise None.
"""
return context.context().summary_step
@tf_export("summary.experimental.set_step", v1=[])
def set_step(step):
"""Sets the default summary step for the current thread.
For convenience, this function sets a default value for the `step` parameter
used in summary-writing functions elsewhere in the API so that it need not
be explicitly passed in every such invocation. The value can be a constant
or a variable, and can be retrieved via `tf.summary.experimental.get_step()`.
Note: when using this with @tf.functions, the step value will be captured at
the time the function is traced, so changes to the step outside the function
will not be reflected inside the function unless using a `tf.Variable` step.
Args:
step: An `int64`-castable default step value, or None to unset.
"""
context.context().summary_step = step
@tf_export("summary.SummaryWriter", v1=[])
@six.add_metaclass(abc.ABCMeta)
class SummaryWriter(object):
"""Interface representing a stateful summary writer object."""
@abc.abstractmethod
def set_as_default(self):
"""Enables this summary writer for the current thread."""
raise NotImplementedError()
@abc.abstractmethod
@tf_contextlib.contextmanager
def as_default(self):
"""Returns a context manager that enables summary writing."""
raise NotImplementedError()
def init(self):
"""Initializes the summary writer."""
raise NotImplementedError()
def flush(self):
"""Flushes any buffered data."""
raise NotImplementedError()
def close(self):
"""Flushes and closes the summary writer."""
raise NotImplementedError()
class ResourceSummaryWriter(SummaryWriter):
"""Implementation of SummaryWriter using a SummaryWriterInterface resource."""
def __init__(self, shared_name, init_op_fn, name=None, v2=False):
self._resource = gen_summary_ops.summary_writer(
shared_name=shared_name, name=name)
# TODO(nickfelt): cache other constructed ops in graph mode
self._init_op_fn = init_op_fn
self._init_op = init_op_fn(self._resource)
self._v2 = v2
self._closed = False
if context.executing_eagerly():
self._resource_deleter = resource_variable_ops.EagerResourceDeleter(
handle=self._resource, handle_device="cpu:0")
else:
ops.add_to_collection(_SUMMARY_WRITER_INIT_COLLECTION_NAME, self._init_op)
def set_as_default(self):
"""Enables this summary writer for the current thread."""
if self._v2 and context.executing_eagerly() and self._closed:
raise RuntimeError("SummaryWriter is already closed")
context.context().summary_writer = self
@tf_contextlib.contextmanager
def as_default(self):
"""Returns a context manager that enables summary writing."""
if self._v2 and context.executing_eagerly() and self._closed:
raise RuntimeError("SummaryWriter is already closed")
old = context.context().summary_writer
try:
context.context().summary_writer = self
yield self
# Flushes the summary writer in eager mode or in graph functions, but
# not in legacy graph mode (you're on your own there).
self.flush()
finally:
context.context().summary_writer = old
def init(self):
"""Initializes the summary writer."""
if self._v2:
if context.executing_eagerly() and self._closed:
raise RuntimeError("SummaryWriter is already closed")
return self._init_op
# Legacy behavior allows re-initializing the resource.
return self._init_op_fn(self._resource)
def flush(self):
"""Flushes any buffered data."""
if self._v2 and context.executing_eagerly() and self._closed:
return
return _flush_fn(writer=self)
def close(self):
"""Flushes and closes the summary writer."""
if self._v2 and context.executing_eagerly() and self._closed:
return
try:
with ops.control_dependencies([self.flush()]):
with ops.device("cpu:0"):
return gen_summary_ops.close_summary_writer(self._resource)
finally:
if self._v2 and context.executing_eagerly():
self._closed = True
class NoopSummaryWriter(SummaryWriter):
"""A summary writer that does nothing, for create_noop_writer()."""
def set_as_default(self):
pass
@tf_contextlib.contextmanager
def as_default(self):
yield
def init(self):
pass
def flush(self):
pass
def close(self):
pass
@tf_export(v1=["summary.initialize"])
def initialize(
graph=None, # pylint: disable=redefined-outer-name
session=None):
"""Initializes summary writing for graph execution mode.
This operation is a no-op when executing eagerly.
This helper method provides a higher-level alternative to using
`tf.contrib.summary.summary_writer_initializer_op` and
`tf.contrib.summary.graph`.
Most users will also want to call `tf.compat.v1.train.create_global_step`
which can happen before or after this function is called.
Args:
graph: A `tf.Graph` or `tf.compat.v1.GraphDef` to output to the writer.
This function will not write the default graph by default. When
writing to an event log file, the associated step will be zero.
session: So this method can call `tf.Session.run`. This defaults
to `tf.compat.v1.get_default_session`.
Raises:
RuntimeError: If the current thread has no default
`tf.contrib.summary.SummaryWriter`.
ValueError: If session wasn't passed and no default session.
"""
if context.executing_eagerly():
return
if context.context().summary_writer is None:
raise RuntimeError("No default tf.contrib.summary.SummaryWriter found")
if session is None:
session = ops.get_default_session()
if session is None:
raise ValueError("session must be passed if no default session exists")
session.run(summary_writer_initializer_op())
if graph is not None:
data = _serialize_graph(graph)
x = array_ops.placeholder(dtypes.string)
session.run(_graph(x, 0), feed_dict={x: data})
@tf_export("summary.create_file_writer", v1=[])
def create_file_writer_v2(logdir,
max_queue=None,
flush_millis=None,
filename_suffix=None,
name=None):
"""Creates a summary file writer for the given log directory.
Args:
logdir: a string specifying the directory in which to write an event file.
max_queue: the largest number of summaries to keep in a queue; will
flush once the queue gets bigger than this. Defaults to 10.
flush_millis: the largest interval between flushes. Defaults to 120,000.
filename_suffix: optional suffix for the event file name. Defaults to `.v2`.
name: a name for the op that creates the writer.
Returns:
A SummaryWriter object.
"""
if logdir is None:
raise ValueError("logdir cannot be None")
inside_function = ops.inside_function()
with ops.name_scope(name, "create_file_writer") as scope, ops.device("cpu:0"):
# Run init inside an init_scope() to hoist it out of tf.functions.
with ops.init_scope():
if context.executing_eagerly():
_check_create_file_writer_args(
inside_function,
logdir=logdir,
max_queue=max_queue,
flush_millis=flush_millis,
filename_suffix=filename_suffix)
logdir = ops.convert_to_tensor(logdir, dtype=dtypes.string)
if max_queue is None:
max_queue = constant_op.constant(10)
if flush_millis is None:
flush_millis = constant_op.constant(2 * 60 * 1000)
if filename_suffix is None:
filename_suffix = constant_op.constant(".v2")
# Prepend the PID and a process-local UID to the filename suffix to avoid
# filename collisions within the machine (the filename already contains
# the hostname to avoid cross-machine collisions).
unique_prefix = constant_op.constant(".%s.%s" % (os.getpid(), ops.uid()))
filename_suffix = unique_prefix + filename_suffix
# Use a unique shared_name to prevent resource sharing.
if context.executing_eagerly():
shared_name = context.shared_name()
else:
shared_name = ops.name_from_scope_name(scope) # pylint: disable=protected-access
return ResourceSummaryWriter(
shared_name=shared_name,
init_op_fn=functools.partial(
gen_summary_ops.create_summary_file_writer,
logdir=logdir,
max_queue=max_queue,
flush_millis=flush_millis,
filename_suffix=filename_suffix),
name=name,
v2=True)
def create_file_writer(logdir,
max_queue=None,
flush_millis=None,
filename_suffix=None,
name=None):
"""Creates a summary file writer in the current context under the given name.
Args:
logdir: a string, or None. If a string, creates a summary file writer
which writes to the directory named by the string. If None, returns
a mock object which acts like a summary writer but does nothing,
useful to use as a context manager.
max_queue: the largest number of summaries to keep in a queue; will
flush once the queue gets bigger than this. Defaults to 10.
flush_millis: the largest interval between flushes. Defaults to 120,000.
filename_suffix: optional suffix for the event file name. Defaults to `.v2`.
name: Shared name for this SummaryWriter resource stored to default
Graph. Defaults to the provided logdir prefixed with `logdir:`. Note: if a
summary writer resource with this shared name already exists, the returned
SummaryWriter wraps that resource and the other arguments have no effect.
Returns:
Either a summary writer or an empty object which can be used as a
summary writer.
"""
if logdir is None:
return NoopSummaryWriter()
logdir = str(logdir)
with ops.device("cpu:0"):
if max_queue is None:
max_queue = constant_op.constant(10)
if flush_millis is None:
flush_millis = constant_op.constant(2 * 60 * 1000)
if filename_suffix is None:
filename_suffix = constant_op.constant(".v2")
if name is None:
name = "logdir:" + logdir
return ResourceSummaryWriter(
shared_name=name,
init_op_fn=functools.partial(
gen_summary_ops.create_summary_file_writer,
logdir=logdir,
max_queue=max_queue,
flush_millis=flush_millis,
filename_suffix=filename_suffix))
def create_db_writer(db_uri,
experiment_name=None,
run_name=None,
user_name=None,
name=None):
"""Creates a summary database writer in the current context.
This can be used to write tensors from the execution graph directly
to a database. Only SQLite is supported right now. This function
will create the schema if it doesn't exist. Entries in the Users,
Experiments, and Runs tables will be created automatically if they
don't already exist.
Args:
db_uri: For example "file:/tmp/foo.sqlite".
experiment_name: Defaults to YYYY-MM-DD in local time if None.
Empty string means the Run will not be associated with an
Experiment. Can't contain ASCII control characters or <>. Case
sensitive.
run_name: Defaults to HH:MM:SS in local time if None. Empty string
means a Tag will not be associated with any Run. Can't contain
ASCII control characters or <>. Case sensitive.
user_name: Defaults to system username if None. Empty means the
Experiment will not be associated with a User. Must be valid as
both a DNS label and Linux username.
name: Shared name for this SummaryWriter resource stored to default
`tf.Graph`.
Returns:
A `tf.summary.SummaryWriter` instance.
"""
with ops.device("cpu:0"):
if experiment_name is None:
experiment_name = time.strftime("%Y-%m-%d", time.localtime(time.time()))
if run_name is None:
run_name = time.strftime("%H:%M:%S", time.localtime(time.time()))
if user_name is None:
user_name = getpass.getuser()
experiment_name = _cleanse_string(
"experiment_name", _EXPERIMENT_NAME_PATTERNS, experiment_name)
run_name = _cleanse_string("run_name", _RUN_NAME_PATTERNS, run_name)
user_name = _cleanse_string("user_name", _USER_NAME_PATTERNS, user_name)
return ResourceSummaryWriter(
shared_name=name,
init_op_fn=functools.partial(
gen_summary_ops.create_summary_db_writer,
db_uri=db_uri,
experiment_name=experiment_name,
run_name=run_name,
user_name=user_name))
@tf_export("summary.create_noop_writer", v1=[])
def create_noop_writer():
"""Returns a summary writer that does nothing.
This is useful as a placeholder in code that expects a context manager.
"""
return NoopSummaryWriter()
def _cleanse_string(name, pattern, value):
if isinstance(value, six.string_types) and pattern.search(value) is None:
raise ValueError("%s (%s) must match %s" % (name, value, pattern.pattern))
return ops.convert_to_tensor(value, dtypes.string)
def _nothing():
"""Convenient else branch for when summaries do not record."""
return constant_op.constant(False)
@tf_export(v1=["summary.all_v2_summary_ops"])
def all_v2_summary_ops():
"""Returns all V2-style summary ops defined in the current default graph.
This includes ops from TF 2.0 tf.summary and TF 1.x tf.contrib.summary (except
for `tf.contrib.summary.graph` and `tf.contrib.summary.import_event`), but
does *not* include TF 1.x tf.summary ops.
Returns:
List of summary ops, or None if called under eager execution.
"""
if context.executing_eagerly():
return None
return ops.get_collection(ops.GraphKeys._SUMMARY_COLLECTION) # pylint: disable=protected-access
def summary_writer_initializer_op():
"""Graph-mode only. Returns the list of ops to create all summary writers.
Returns:
The initializer ops.
Raises:
RuntimeError: If in Eager mode.
"""
if context.executing_eagerly():
raise RuntimeError(
"tf.contrib.summary.summary_writer_initializer_op is only "
"supported in graph mode.")
return ops.get_collection(_SUMMARY_WRITER_INIT_COLLECTION_NAME)
_INVALID_SCOPE_CHARACTERS = re.compile(r"[^-_/.A-Za-z0-9]")
@tf_export("summary.experimental.summary_scope", v1=[])
@tf_contextlib.contextmanager
def summary_scope(name, default_name="summary", values=None):
"""Experimental context manager for use when defining a custom summary op.
This behaves similarly to `tf.name_scope`, except that it returns a generated
summary tag in addition to the scope name. The tag is structurally similar to
the scope name - derived from the user-provided name, prefixed with enclosing
name scopes if any - but we relax the constraint that it be uniquified, as
well as the character set limitation (so the user-provided name can contain
characters not legal for scope names; in the scope name these are removed).
This makes the summary tag more predictable and consistent for the user.
For example, to define a new summary op called `my_op`:
```python
def my_op(name, my_value, step):
with tf.summary.summary_scope(name, "MyOp", [my_value]) as (tag, scope):
my_value = tf.convert_to_tensor(my_value)
return tf.summary.write(tag, my_value, step=step)
```
Args:
name: string name for the summary.
default_name: Optional; if provided, used as default name of the summary.
values: Optional; passed as `values` parameter to name_scope.
Yields:
A tuple `(tag, scope)` as described above.
"""
name = name or default_name
current_scope = ops.get_name_scope()
tag = current_scope + "/" + name if current_scope else name
# Strip illegal characters from the scope name, and if that leaves nothing,
# use None instead so we pick up the default name.
name = _INVALID_SCOPE_CHARACTERS.sub("", name) or None
with ops.name_scope(name, default_name, values) as scope:
yield tag, scope
@tf_export("summary.write", v1=[])
def write(tag, tensor, step=None, metadata=None, name=None):
"""Writes a generic summary to the default SummaryWriter if one exists.
This exists primarily to support the definition of type-specific summary ops
like scalar() and image(), and is not intended for direct use unless defining
a new type-specific summary op.
Args:
tag: string tag used to identify the summary (e.g. in TensorBoard), usually
generated with `tf.summary.summary_scope`
tensor: the Tensor holding the summary data to write
step: Explicit `int64`-castable monotonic step value for this summary. If
omitted, this defaults to `tf.summary.experimental.get_step()`, which must
not be None.
metadata: Optional SummaryMetadata, as a proto or serialized bytes
name: Optional string name for this op.
Returns:
True on success, or false if no summary was written because no default
summary writer was available.
Raises:
ValueError: if a default writer exists, but no step was provided and
`tf.summary.experimental.get_step()` is None.
"""
with ops.name_scope(name, "write_summary") as scope:
if context.context().summary_writer is None:
return constant_op.constant(False)
if step is None:
step = get_step()
if step is None:
raise ValueError("No step set via 'step' argument or "
"tf.summary.experimental.set_step()")
if metadata is None:
serialized_metadata = b""
elif hasattr(metadata, "SerializeToString"):
serialized_metadata = metadata.SerializeToString()
else:
serialized_metadata = metadata
def record():
"""Record the actual summary and return True."""
# Note the identity to move the tensor to the CPU.
with ops.device("cpu:0"):
write_summary_op = gen_summary_ops.write_summary(
context.context().summary_writer._resource, # pylint: disable=protected-access
step,
array_ops.identity(tensor),
tag,
serialized_metadata,
name=scope)
with ops.control_dependencies([write_summary_op]):
return constant_op.constant(True)
with ops.device("cpu:0"):
op = smart_cond.smart_cond(
_should_record_summaries_v2(), record, _nothing, name="summary_cond")
if not context.executing_eagerly():
ops.add_to_collection(ops.GraphKeys._SUMMARY_COLLECTION, op) # pylint: disable=protected-access
return op
@tf_export("summary.experimental.write_raw_pb", v1=[])
def write_raw_pb(tensor, step=None, name=None):
"""Writes a summary using raw `tf.compat.v1.Summary` protocol buffers.
Experimental: this exists to support the usage of V1-style manual summary
writing (via the construction of a `tf.compat.v1.Summary` protocol buffer)
with the V2 summary writing API.
Args:
tensor: the string Tensor holding one or more serialized `Summary` protobufs
step: Explicit `int64`-castable monotonic step value for this summary. If
omitted, this defaults to `tf.summary.experimental.get_step()`, which must
not be None.
name: Optional string name for this op.
Returns:
True on success, or false if no summary was written because no default
summary writer was available.
Raises:
ValueError: if a default writer exists, but no step was provided and
`tf.summary.experimental.get_step()` is None.
"""
with ops.name_scope(name, "write_raw_pb") as scope:
if context.context().summary_writer is None:
return constant_op.constant(False)
if step is None:
step = get_step()
if step is None:
raise ValueError("No step set via 'step' argument or "
"tf.summary.experimental.set_step()")
def record():
"""Record the actual summary and return True."""
# Note the identity to move the tensor to the CPU.
with ops.device("cpu:0"):
raw_summary_op = gen_summary_ops.write_raw_proto_summary(
context.context().summary_writer._resource, # pylint: disable=protected-access
step,
array_ops.identity(tensor),
name=scope)
with ops.control_dependencies([raw_summary_op]):
return constant_op.constant(True)
with ops.device("cpu:0"):
op = smart_cond.smart_cond(
_should_record_summaries_v2(), record, _nothing, name="summary_cond")
if not context.executing_eagerly():
ops.add_to_collection(ops.GraphKeys._SUMMARY_COLLECTION, op) # pylint: disable=protected-access
return op
def summary_writer_function(name, tensor, function, family=None):
"""Helper function to write summaries.
Args:
name: name of the summary
tensor: main tensor to form the summary
function: function taking a tag and a scope which writes the summary
family: optional, the summary's family
Returns:
The result of writing the summary.
"""
name_scope = ops.get_name_scope()
if name_scope:
# Add a slash to allow reentering the name scope.
name_scope += "/"
def record():
with ops.name_scope(name_scope), summary_op_util.summary_scope(
name, family, values=[tensor]) as (tag, scope):
with ops.control_dependencies([function(tag, scope)]):
return constant_op.constant(True)
if context.context().summary_writer is None:
return control_flow_ops.no_op()
with ops.device("cpu:0"):
op = smart_cond.smart_cond(
should_record_summaries(), record, _nothing, name="")
if not context.executing_eagerly():
ops.add_to_collection(ops.GraphKeys._SUMMARY_COLLECTION, op) # pylint: disable=protected-access
return op
def generic(name, tensor, metadata=None, family=None, step=None):
"""Writes a tensor summary if possible."""
def function(tag, scope):
if metadata is None:
serialized_metadata = constant_op.constant("")
elif hasattr(metadata, "SerializeToString"):
serialized_metadata = constant_op.constant(metadata.SerializeToString())
else:
serialized_metadata = metadata
# Note the identity to move the tensor to the CPU.
return gen_summary_ops.write_summary(
context.context().summary_writer._resource, # pylint: disable=protected-access
_choose_step(step),
array_ops.identity(tensor),
tag,
serialized_metadata,
name=scope)
return summary_writer_function(name, tensor, function, family=family)
def scalar(name, tensor, family=None, step=None):
"""Writes a scalar summary if possible.
Unlike `tf.contrib.summary.generic` this op may change the dtype
depending on the writer, for both practical and efficiency concerns.
Args:
name: An arbitrary name for this summary.
tensor: A `tf.Tensor` Must be one of the following types:
`float32`, `float64`, `int32`, `int64`, `uint8`, `int16`,
`int8`, `uint16`, `half`, `uint32`, `uint64`.
family: Optional, the summary's family.
step: The `int64` monotonic step variable, which defaults
to `tf.compat.v1.train.get_global_step`.
Returns:
The created `tf.Operation` or a `tf.no_op` if summary writing has
not been enabled for this context.
"""
def function(tag, scope):
# Note the identity to move the tensor to the CPU.
return gen_summary_ops.write_scalar_summary(
context.context().summary_writer._resource, # pylint: disable=protected-access
_choose_step(step),
tag,
array_ops.identity(tensor),
name=scope)
return summary_writer_function(name, tensor, function, family=family)
def histogram(name, tensor, family=None, step=None):
"""Writes a histogram summary if possible."""
def function(tag, scope):
# Note the identity to move the tensor to the CPU.
return gen_summary_ops.write_histogram_summary(
context.context().summary_writer._resource, # pylint: disable=protected-access
_choose_step(step),
tag,
array_ops.identity(tensor),
name=scope)
return summary_writer_function(name, tensor, function, family=family)
def image(name, tensor, bad_color=None, max_images=3, family=None, step=None):
"""Writes an image summary if possible."""
def function(tag, scope):
bad_color_ = (constant_op.constant([255, 0, 0, 255], dtype=dtypes.uint8)
if bad_color is None else bad_color)
# Note the identity to move the tensor to the CPU.
return gen_summary_ops.write_image_summary(
context.context().summary_writer._resource, # pylint: disable=protected-access
_choose_step(step),
tag,
array_ops.identity(tensor),
bad_color_,
max_images,
name=scope)
return summary_writer_function(name, tensor, function, family=family)
def audio(name, tensor, sample_rate, max_outputs, family=None, step=None):
"""Writes an audio summary if possible."""
def function(tag, scope):
# Note the identity to move the tensor to the CPU.
return gen_summary_ops.write_audio_summary(
context.context().summary_writer._resource, # pylint: disable=protected-access
_choose_step(step),
tag,
array_ops.identity(tensor),
sample_rate=sample_rate,
max_outputs=max_outputs,
name=scope)
return summary_writer_function(name, tensor, function, family=family)
def graph(param, step=None, name=None):
"""Writes a TensorFlow graph to the summary interface.
The graph summary is, strictly speaking, not a summary. Conditions
like `tf.summary.should_record_summaries` do not apply. Only
a single graph can be associated with a particular run. If multiple
graphs are written, then only the last one will be considered by
TensorBoard.
When not using eager execution mode, the user should consider passing
the `graph` parameter to `tf.compat.v1.summary.initialize` instead of
calling this function. Otherwise special care needs to be taken when
using the graph to record the graph.
Args:
param: A `tf.Tensor` containing a serialized graph proto. When
eager execution is enabled, this function will automatically
coerce `tf.Graph`, `tf.compat.v1.GraphDef`, and string types.
step: The global step variable. This doesn't have useful semantics
for graph summaries, but is used anyway, due to the structure of
event log files. This defaults to the global step.
name: A name for the operation (optional).
Returns:
The created `tf.Operation` or a `tf.no_op` if summary writing has
not been enabled for this context.
Raises:
TypeError: If `param` isn't already a `tf.Tensor` in graph mode.
"""
if not context.executing_eagerly() and not isinstance(param, ops.Tensor):
raise TypeError("graph() needs a tf.Tensor (e.g. tf.placeholder) in graph "
"mode, but was: %s" % type(param))
writer = context.context().summary_writer
if writer is None:
return control_flow_ops.no_op()
with ops.device("cpu:0"):
if isinstance(param, (ops.Graph, graph_pb2.GraphDef)):
tensor = ops.convert_to_tensor(_serialize_graph(param), dtypes.string)
else:
tensor = array_ops.identity(param)
return gen_summary_ops.write_graph_summary(
writer._resource, _choose_step(step), tensor, name=name) # pylint: disable=protected-access
_graph = graph # for functions with a graph parameter
def import_event(tensor, name=None):
"""Writes a `tf.compat.v1.Event` binary proto.
This can be used to import existing event logs into a new summary writer sink.
Please note that this is lower level than the other summary functions and
will ignore the `tf.summary.should_record_summaries` setting.
Args:
tensor: A `tf.Tensor` of type `string` containing a serialized
`tf.compat.v1.Event` proto.
name: A name for the operation (optional).
Returns:
The created `tf.Operation`.
"""
return gen_summary_ops.import_event(
context.context().summary_writer._resource, tensor, name=name) # pylint: disable=protected-access
@tf_export("summary.flush", v1=[])
def flush(writer=None, name=None):
"""Forces summary writer to send any buffered data to storage.
This operation blocks until that finishes.
Args:
writer: The `tf.summary.SummaryWriter` resource to flush.
The thread default will be used if this parameter is None.
Otherwise a `tf.no_op` is returned.
name: A name for the operation (optional).
Returns:
The created `tf.Operation`.
"""
if writer is None:
writer = context.context().summary_writer
if writer is None:
return control_flow_ops.no_op()
if isinstance(writer, ResourceSummaryWriter):
resource = writer._resource # pylint: disable=protected-access
else:
# Assume we were passed a raw resource tensor.
resource = writer
with ops.device("cpu:0"):
return gen_summary_ops.flush_summary_writer(resource, name=name)
_flush_fn = flush # for within SummaryWriter.flush()
def eval_dir(model_dir, name=None):
"""Construct a logdir for an eval summary writer."""
return os.path.join(model_dir, "eval" if not name else "eval_" + name)
@deprecation.deprecated(date=None,
instructions="Renamed to create_file_writer().")
def create_summary_file_writer(*args, **kwargs):
"""Please use `tf.contrib.summary.create_file_writer`."""
logging.warning("Deprecation Warning: create_summary_file_writer was renamed "
"to create_file_writer")
return create_file_writer(*args, **kwargs)
def _serialize_graph(arbitrary_graph):
if isinstance(arbitrary_graph, ops.Graph):
return arbitrary_graph.as_graph_def(add_shapes=True).SerializeToString()
else:
return arbitrary_graph.SerializeToString()
def _choose_step(step):
if step is None:
return training_util.get_or_create_global_step()
if not isinstance(step, ops.Tensor):
return ops.convert_to_tensor(step, dtypes.int64)
return step
def _check_create_file_writer_args(inside_function, **kwargs):
"""Helper to check the validity of arguments to a create_file_writer() call.
Args:
inside_function: whether the create_file_writer() call is in a tf.function
**kwargs: the arguments to check, as kwargs to give them names.
Raises:
ValueError: if the arguments are graph tensors.
"""
for arg_name, arg in kwargs.items():
if not isinstance(arg, ops.EagerTensor) and tensor_util.is_tensor(arg):
if inside_function:
raise ValueError(
"Invalid graph Tensor argument \"%s=%s\" to create_file_writer() "
"inside an @tf.function. The create call will be lifted into the "
"outer eager execution context, so it cannot consume graph tensors "
"defined inside the function body." % (arg_name, arg))
else:
raise ValueError(
"Invalid graph Tensor argument \"%s=%s\" to eagerly executed "
"create_file_writer()." % (arg_name, arg))
def run_metadata(name, data, step=None):
"""Writes entire RunMetadata summary.
A RunMetadata can contain DeviceStats, partition graphs, and function graphs.
Please refer to the proto for definition of each field.
Args:
name: A name for this summary. The summary tag used for TensorBoard will be
this name prefixed by any active name scopes.
data: A RunMetadata proto to write.
step: Explicit `int64`-castable monotonic step value for this summary. If
omitted, this defaults to `tf.summary.experimental.get_step()`, which must
not be None.
Returns:
True on success, or false if no summary was written because no default
summary writer was available.
Raises:
ValueError: if a default writer exists, but no step was provided and
`tf.summary.experimental.get_step()` is None.
"""
summary_metadata = summary_pb2.SummaryMetadata()
# Hard coding a plugin name. Please refer to go/tb-plugin-name-hardcode for
# the rationale.
summary_metadata.plugin_data.plugin_name = "graph_run_metadata"
# version number = 1
summary_metadata.plugin_data.content = b"1"
with summary_scope(name,
"graph_run_metadata_summary",
[data, step]) as (tag, _):
with ops.device("cpu:0"):
tensor = constant_op.constant(data.SerializeToString(),
dtype=dtypes.string)
return write(
tag=tag,
tensor=tensor,
step=step,
metadata=summary_metadata)
def run_metadata_graphs(name, data, step=None):
"""Writes graphs from a RunMetadata summary.
Args:
name: A name for this summary. The summary tag used for TensorBoard will be
this name prefixed by any active name scopes.
data: A RunMetadata proto to write.
step: Explicit `int64`-castable monotonic step value for this summary. If
omitted, this defaults to `tf.summary.experimental.get_step()`, which must
not be None.
Returns:
True on success, or false if no summary was written because no default
summary writer was available.
Raises:
ValueError: if a default writer exists, but no step was provided and
`tf.summary.experimental.get_step()` is None.
"""
summary_metadata = summary_pb2.SummaryMetadata()
# Hard coding a plugin name. Please refer to go/tb-plugin-name-hardcode for
# the rationale.
summary_metadata.plugin_data.plugin_name = "graph_run_metadata_graph"
# version number = 1
summary_metadata.plugin_data.content = b"1"
data = config_pb2.RunMetadata(
function_graphs=data.function_graphs,
partition_graphs=data.partition_graphs)
with summary_scope(name,
"graph_run_metadata_graph_summary",
[data, step]) as (tag, _):
with ops.device("cpu:0"):
tensor = constant_op.constant(data.SerializeToString(),
dtype=dtypes.string)
return write(
tag=tag,
tensor=tensor,
step=step,
metadata=summary_metadata)
def keras_model(name, data, step=None):
"""Writes a Keras model as JSON to as a Summary.
Writing the Keras model configuration allows the TensorBoard graph plugin to
render a conceptual graph, as opposed to graph of ops. In case the model fails
to serialze as JSON, it ignores and returns False.
Args:
name: A name for this summary. The summary tag used for TensorBoard will be
this name prefixed by any active name scopes.
data: A Keras Model to write.
step: Explicit `int64`-castable monotonic step value for this summary. If
omitted, this defaults to `tf.summary.experimental.get_step()`, which must
not be None.
Returns:
True on success, or False if no summary was written because no default
summary writer was available.
Raises:
ValueError: if a default writer exists, but no step was provided and
`tf.summary.experimental.get_step()` is None.
"""
summary_metadata = summary_pb2.SummaryMetadata()
# Hard coding a plugin name. Please refer to go/tb-plugin-name-hardcode for
# the rationale.
summary_metadata.plugin_data.plugin_name = "graph_keras_model"
# version number = 1
summary_metadata.plugin_data.content = b"1"
try:
json_string = data.to_json()
except Exception as exc: # pylint: disable=broad-except
# An exception should not break a model code.
logging.warn("Model failed to serialize as JSON. Ignoring... %s" % exc)
return False
with summary_scope(name, "graph_keras_model", [data, step]) as (tag, _):
with ops.device("cpu:0"):
tensor = constant_op.constant(json_string, dtype=dtypes.string)
return write(
tag=tag,
tensor=tensor,
step=step,
metadata=summary_metadata)
_TraceContext = collections.namedtuple("TraceContext", ("graph", "profiler"))
_current_trace_context_lock = threading.Lock()
_current_trace_context = None
@tf_export("summary.trace_on", v1=[])
def trace_on(graph=True, profiler=False): # pylint: disable=redefined-outer-name
"""Starts a trace to record computation graphs and profiling information.
Must be invoked in eager mode.
When enabled, TensorFlow runtime will collection information that can later be
exported and consumed by TensorBoard. The trace is activated across the entire
TensorFlow runtime and affects all threads of execution.
To stop the trace and export the collected information, use
`tf.summary.trace_export`. To stop the trace without exporting, use
`tf.summary.trace_off`.
Args:
graph: If True, enables collection of executed graphs. It includes ones from
tf.function invocation and ones from the legacy graph mode. The default
is True.
profiler: If True, enables the advanced profiler. Enabling profiler
implicitly enables the graph collection. The profiler may incur a high
memory overhead. The default is False.
"""
if ops.inside_function():
logging.warn("Cannot enable trace inside a tf.function.")
return
if not context.context().executing_eagerly():
logging.warn("Must enable trace in eager mode.")
return
global _current_trace_context
with _current_trace_context_lock:
if _current_trace_context:
logging.warn("Trace already enabled")
return
if graph and not profiler:
context.context().enable_graph_collection()
if profiler:
context.context().enable_run_metadata()
_profiler.start()
_current_trace_context = _TraceContext(graph=graph, profiler=profiler)
@tf_export("summary.trace_export", v1=[])
def trace_export(name, step=None, profiler_outdir=None):
"""Stops and exports the active trace as a Summary and/or profile file.
Stops the trace and exports all metadata collected during the trace to the
default SummaryWriter, if one has been set.
Args:
name: A name for the summary to be written.
step: Explicit `int64`-castable monotonic step value for this summary. If
omitted, this defaults to `tf.summary.experimental.get_step()`, which must
not be None.
profiler_outdir: Output directory for profiler. It is required when profiler
is enabled when trace was started. Otherwise, it is ignored.
Raises:
ValueError: if a default writer exists, but no step was provided and
`tf.summary.experimental.get_step()` is None.
"""
# TODO(stephanlee): See if we can remove profiler_outdir and infer it from
# the SummaryWriter's logdir.
global _current_trace_context
if ops.inside_function():
logging.warn("Cannot export trace inside a tf.function.")
return
if not context.context().executing_eagerly():
logging.warn("Can only export trace while executing eagerly.")
return
with _current_trace_context_lock:
if _current_trace_context is None:
raise ValueError("Must enable trace before export.")
graph, profiler = _current_trace_context # pylint: disable=redefined-outer-name
if profiler and profiler_outdir is None:
raise ValueError("Required profiler_outdir is not specified")
run_meta = context.context().export_run_metadata()
if graph and not profiler:
run_metadata_graphs(name, run_meta, step)
else:
run_metadata(name, run_meta, step)
if profiler:
_profiler.save(profiler_outdir, _profiler.stop())
trace_off()
@tf_export("summary.trace_off", v1=[])
def trace_off():
"""Stops the current trace and discards any collected information."""
global _current_trace_context
with _current_trace_context_lock:
_current_trace_context = None
# Disabling run_metadata disables graph collection as well.
context.context().disable_run_metadata()
# profiler only has start and stop. One needs to stop in order to export
# and stopping when it is not running will raise an error.
try:
_profiler.stop()
except _profiler.ProfilerNotRunningError:
pass
|
tensorflow-master
|
tensorflow/python/ops/summary_ops_v2.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""End-to-end benchmark for batch normalization."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import sys
import time
from tensorflow.python.client import session as session_lib
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import ops
from tensorflow.python.framework import test_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gen_nn_ops
from tensorflow.python.ops import gradients_impl
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import nn_impl
from tensorflow.python.ops import random_ops
from tensorflow.python.ops import variables
import tensorflow.python.ops.nn_grad # pylint: disable=unused-import
from tensorflow.python.platform import test
def batch_norm_op(tensor, mean, variance, beta, gamma, scale):
"""Fused kernel for batch normalization."""
# _batch_norm_with_global_normalization is deprecated in v9
test_util.set_producer_version(ops.get_default_graph(), 8)
# pylint: disable=protected-access
return gen_nn_ops._batch_norm_with_global_normalization(
tensor, mean, variance, beta, gamma, 0.001, scale)
# pylint: enable=protected-access
# Note that the naive implementation is much slower:
# batch_norm = (tensor - mean) * tf.math.rsqrt(variance + 0.001)
# if scale:
# batch_norm *= gamma
# return batch_norm + beta
def batch_norm_py(tensor, mean, variance, beta, gamma, scale):
"""Python implementation of batch normalization."""
return nn_impl.batch_normalization(tensor, mean, variance, beta, gamma if
scale else None, 0.001)
def batch_norm_slow(tensor, mean, variance, beta, gamma, scale):
batch_norm = (tensor - mean) * math_ops.rsqrt(variance + 0.001)
if scale:
batch_norm *= gamma
return batch_norm + beta
def build_graph(device, input_shape, axes, num_layers, mode, scale, train):
"""Build a graph containing a sequence of batch normalizations.
Args:
device: string, the device to run on.
input_shape: shape of the input tensor.
axes: axes that are to be normalized across.
num_layers: number of batch normalization layers in the graph.
mode: "op", "py" or "slow" depending on the implementation.
scale: scale after normalization.
train: if true, also run backprop.
Returns:
An array of tensors to run()
"""
moment_shape = []
keep_dims = mode == "py" or mode == "slow"
if keep_dims:
for axis in range(len(input_shape)):
if axis in axes:
moment_shape.append(1)
else:
moment_shape.append(input_shape[axis])
else:
for axis in range(len(input_shape)):
if axis not in axes:
moment_shape.append(input_shape[axis])
with ops.device("/%s:0" % device):
tensor = variables.Variable(random_ops.truncated_normal(input_shape))
for _ in range(num_layers):
if train:
mean, variance = nn_impl.moments(tensor, axes, keep_dims=keep_dims)
else:
mean = array_ops.zeros(moment_shape)
variance = array_ops.ones(moment_shape)
beta = variables.Variable(array_ops.zeros(moment_shape))
gamma = variables.Variable(constant_op.constant(1.0, shape=moment_shape))
if mode == "py":
tensor = batch_norm_py(tensor, mean, variance, beta, gamma, scale)
elif mode == "op":
tensor = batch_norm_op(tensor, mean, variance, beta, gamma, scale)
elif mode == "slow":
tensor = batch_norm_slow(tensor, mean, variance, beta, gamma, scale)
if train:
return gradients_impl.gradients([tensor], variables.trainable_variables())
else:
return [tensor]
def print_difference(mode, t1, t2):
"""Print the difference in timing between two runs."""
difference = (t2 - t1) / t1 * 100.0
print("=== %s: %.1f%% ===" % (mode, difference))
class BatchNormBenchmark(test.Benchmark):
"""Benchmark batch normalization."""
def _run_graph(self, device, input_shape, axes, num_layers, mode, scale,
train, num_iters):
"""Run the graph and print its execution time.
Args:
device: string, the device to run on.
input_shape: shape of the input tensor.
axes: axes that are to be normalized across.
num_layers: number of batch normalization layers in the graph.
mode: "op", "py" or "slow" depending on the implementation.
scale: scale after normalization.
train: if true, also run backprop.
num_iters: number of steps to run.
Returns:
The duration of the run in seconds.
"""
graph = ops.Graph()
with graph.as_default():
outputs = build_graph(device, input_shape, axes, num_layers, mode, scale,
train)
with session_lib.Session(graph=graph) as session:
variables.global_variables_initializer().run()
_ = session.run([out.op for out in outputs]) # warm up.
start_time = time.time()
for _ in range(num_iters):
_ = session.run([out.op for out in outputs])
duration = time.time() - start_time
print("%s shape:%d/%d #layers:%d mode:%s scale:%r train:%r - %f secs" %
(device, len(input_shape), len(axes), num_layers, mode, scale, train,
duration / num_iters))
name_template = (
"batch_norm_{device}_input_shape_{shape}_axes_{axes}_mode_{mode}_"
"layers_{num_layers}_scale_{scale}_"
"train_{train}")
self.report_benchmark(
name=name_template.format(
device=device,
mode=mode,
num_layers=num_layers,
scale=scale,
train=train,
shape=str(input_shape).replace(" ", ""),
axes=str(axes)).replace(" ", ""),
iters=num_iters,
wall_time=duration / num_iters)
return duration
def benchmark_batch_norm(self):
print("Forward convolution (lower layers).")
shape = [8, 128, 128, 32]
axes = [0, 1, 2]
t1 = self._run_graph("cpu", shape, axes, 10, "op", True, False, 5)
t2 = self._run_graph("cpu", shape, axes, 10, "py", True, False, 5)
t3 = self._run_graph("cpu", shape, axes, 10, "slow", True, False, 5)
print_difference("op vs py", t1, t2)
print_difference("py vs slow", t2, t3)
if FLAGS.use_gpu:
t1 = self._run_graph("gpu", shape, axes, 10, "op", True, False, 50)
t2 = self._run_graph("gpu", shape, axes, 10, "py", True, False, 50)
t3 = self._run_graph("gpu", shape, axes, 10, "slow", True, False, 50)
print_difference("op vs py", t1, t2)
print_difference("py vs slow", t2, t3)
print("Forward/backward convolution (lower layers).")
t1 = self._run_graph("cpu", shape, axes, 10, "op", True, True, 5)
t2 = self._run_graph("cpu", shape, axes, 10, "py", True, True, 5)
t3 = self._run_graph("cpu", shape, axes, 10, "slow", True, True, 5)
print_difference("op vs py", t1, t2)
print_difference("py vs slow", t2, t3)
if FLAGS.use_gpu:
t1 = self._run_graph("gpu", shape, axes, 10, "op", True, True, 50)
t2 = self._run_graph("gpu", shape, axes, 10, "py", True, True, 50)
t3 = self._run_graph("gpu", shape, axes, 10, "slow", True, True, 50)
print_difference("op vs py", t1, t2)
print_difference("py vs slow", t2, t3)
print("Forward convolution (higher layers).")
shape = [256, 17, 17, 32]
axes = [0, 1, 2]
t1 = self._run_graph("cpu", shape, axes, 10, "op", True, False, 5)
t2 = self._run_graph("cpu", shape, axes, 10, "py", True, False, 5)
t3 = self._run_graph("cpu", shape, axes, 10, "slow", True, False, 5)
print_difference("op vs py", t1, t2)
print_difference("py vs slow", t2, t3)
if FLAGS.use_gpu:
t1 = self._run_graph("gpu", shape, axes, 10, "op", True, False, 50)
t2 = self._run_graph("gpu", shape, axes, 10, "py", True, False, 50)
t3 = self._run_graph("gpu", shape, axes, 10, "slow", True, False, 50)
print_difference("op vs py", t1, t2)
print_difference("py vs slow", t2, t3)
print("Forward/backward convolution (higher layers).")
t1 = self._run_graph("cpu", shape, axes, 10, "op", True, True, 5)
t2 = self._run_graph("cpu", shape, axes, 10, "py", True, True, 5)
t3 = self._run_graph("cpu", shape, axes, 10, "slow", True, True, 5)
print_difference("op vs py", t1, t2)
print_difference("py vs slow", t2, t3)
if FLAGS.use_gpu:
t1 = self._run_graph("gpu", shape, axes, 10, "op", True, True, 50)
t2 = self._run_graph("gpu", shape, axes, 10, "py", True, True, 50)
t3 = self._run_graph("gpu", shape, axes, 10, "slow", True, True, 50)
print_difference("op vs py", t1, t2)
print_difference("py vs slow", t2, t3)
print("Forward fully-connected.")
shape = [1024, 32]
axes = [0]
t1 = self._run_graph("cpu", shape, axes, 10, "py", True, False, 5)
t2 = self._run_graph("cpu", shape, axes, 10, "slow", True, False, 5)
print_difference("py vs slow", t1, t2)
if FLAGS.use_gpu:
t1 = self._run_graph("gpu", shape, axes, 10, "py", True, False, 50)
t2 = self._run_graph("gpu", shape, axes, 10, "slow", True, False, 50)
print_difference("py vs slow", t1, t2)
print("Forward/backward fully-connected.")
t1 = self._run_graph("cpu", shape, axes, 10, "py", True, True, 50)
t2 = self._run_graph("cpu", shape, axes, 10, "slow", True, True, 50)
print_difference("py vs slow", t1, t2)
if FLAGS.use_gpu:
t1 = self._run_graph("gpu", shape, axes, 10, "py", True, True, 5)
t2 = self._run_graph("gpu", shape, axes, 10, "slow", True, True, 5)
print_difference("py vs slow", t1, t2)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.register("type", "bool", lambda v: v.lower() == "true")
parser.add_argument(
"--use_gpu",
type="bool",
nargs="?",
const=True,
default=True,
help="Run GPU benchmarks."
)
global FLAGS # pylint:disable=global-at-module-level
FLAGS, unparsed = parser.parse_known_args()
test.main(argv=[sys.argv[0]] + unparsed)
|
tensorflow-master
|
tensorflow/python/ops/batch_norm_benchmark.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Benchmark for Conv2D op."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import itertools
import time
from tensorflow.core.protobuf import config_pb2
from tensorflow.core.protobuf import rewriter_config_pb2
from tensorflow.python.client import session as session_lib
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import nn_ops
from tensorflow.python.ops import random_ops
from tensorflow.python.ops import variables
from tensorflow.python.platform import flags
from tensorflow.python.platform import test
FLAGS = flags.FLAGS
flags.DEFINE_boolean(
"enable_layout_optimizer", False,
"If true, enables layout optimizer to update input data format for faster "
"execution of convolution ops.")
def build_graph(device, dtype, data_format, input_shape, filter_shape, strides,
padding, num_iters, warmup_iters):
"""builds a graph containing a sequence of conv2d operations.
Args:
device: String, the device to run on.
dtype: Data type for the convolution.
data_format: A string from: "NHWC" or "NCHW". Data format for input and
output data.
input_shape: Shape of the input tensor.
filter_shape: Shape of the filter tensor.
strides: A list of ints. 1-D of length 4. The stride of sliding
window for each dimension of input.
padding: A string from: "SAME", "VALID". The type of padding
algorithm to use.
num_iters: number of iterations to run conv2d.
warmup_iters: number of iterations for warmup runs.
Returns:
An array of tensors to run()
"""
with ops.device("/%s:0" % device):
inp = variables.VariableV1(
random_ops.truncated_normal(input_shape, dtype=dtype))
filt = variables.VariableV1(
random_ops.truncated_normal(filter_shape, dtype=dtype))
outputs = []
conv2d_op = nn_ops.conv2d(
inp, filt, strides, padding, data_format=data_format)
outputs.append(conv2d_op)
for _ in range(1, num_iters):
with ops.control_dependencies([conv2d_op]):
conv2d_op = nn_ops.conv2d(
inp, filt, strides, padding, data_format=data_format)
outputs.append(conv2d_op)
warmup_groups = []
warmup_conv2d_op = nn_ops.conv2d(
inp, filt, strides, padding, data_format=data_format)
warmup_groups.append(warmup_conv2d_op)
for _ in range(1, warmup_iters):
with ops.control_dependencies([warmup_conv2d_op]):
warmup_conv2d_op = nn_ops.conv2d(
inp, filt, strides, padding, data_format=data_format)
warmup_groups.append(warmup_conv2d_op)
return control_flow_ops.group(*warmup_groups), control_flow_ops.group(
*outputs)
class Conv2DBenchmark(test.Benchmark):
"""Benchmark conv2d!"""
def _run_graph(self, device, dtype, data_format, input_shape, filter_shape,
strides, padding, num_iters, warmup_iters):
"""runs the graph and print its execution time.
Args:
device: String, the device to run on.
dtype: Data type for the convolution.
data_format: A string from: "NHWC" or "NCHW". Data format for input and
output data.
input_shape: Shape of the input tensor.
filter_shape: Shape of the filter tensor.
strides: A list of ints. 1-D of length 4. The stride of sliding
window for each dimension of input.
padding: A string from: "SAME", "VALID". The type of padding
algorithm to use. num_iters: Number of iterations to run the
benchmark.
num_iters: number of iterations to run conv2d.
warmup_iters: number of iterations for warmup runs.
Returns:
The duration of the run in seconds.
"""
graph = ops.Graph()
with graph.as_default():
warmup_outputs, outputs = build_graph(device, dtype, data_format,
input_shape, filter_shape, strides,
padding, num_iters, warmup_iters)
config = config_pb2.ConfigProto()
config.graph_options.optimizer_options.opt_level = -1
rewrite_options = config.graph_options.rewrite_options
# Disable layout optimizer to not change input data_format.
rewrite_options.layout_optimizer = (
rewriter_config_pb2.RewriterConfig.ON if FLAGS.enable_layout_optimizer
else rewriter_config_pb2.RewriterConfig.OFF)
# Convolution ops are effectively noop in the test graph as we are not
# fetching the convolution outputs. Disable dependency optimizer to not
# remove the conv ops.
rewrite_options.dependency_optimization = (
rewriter_config_pb2.RewriterConfig.OFF)
with session_lib.Session(graph=graph, config=config) as session:
# TODO(hinsu): Use run_op_benchmark method from test.Benchmark to run
# benchmark along with warmup.
variables.global_variables_initializer().run()
# warmup runs
session.run(warmup_outputs)
start_time = time.time()
session.run(outputs)
duration = (time.time() - start_time) / num_iters
print("%s %s %s inputshape:%s filtershape:%s strides:%s padding:%s "
"%d iters: %.8f sec" %
(device, str(dtype), data_format, str(input_shape).replace(
" ", ""), str(filter_shape).replace(" ", ""),
str(strides).replace(" ", ""), padding, num_iters, duration))
name_template = (
"conv2d_{device}_{datatype}_{data_format}_input_shape_{inputshape}_"
"filter_shape_{filtershape}_strides_{strides}_padding_{padding}")
self.report_benchmark(
name=name_template.format(
device=device,
datatype=str(dtype),
data_format=str(data_format),
inputshape=str(input_shape).replace(" ", ""),
filtershape=str(filter_shape).replace(" ", ""),
strides=str(strides).replace(" ", ""),
padding=padding).replace(" ", ""),
iters=num_iters,
wall_time=duration)
return duration
def benchmark_conv2d(self):
print("conv2d benchmark:")
data_types = [dtypes.float32, dtypes.float16]
data_formats = ["NHWC", "NCHW"]
in_channels = list(range(1, 10)) + list(range(10, 20, 2)) + list(
range(20, 33, 4))
out_channels = [4, 16, 32]
hw_strides = [[2, 2]]
paddings = ["VALID", "SAME"]
args_lists = [
data_types, data_formats, in_channels, out_channels, hw_strides,
paddings
]
for args in itertools.product(*args_lists):
dtype, data_format, in_channel, out_channel, hw_stride, padding = args
# Keep batch size same as out channels just to reduce the number of
# different configurations to benchmark.
batch_size = out_channel
h, w, fh, fw = 500, 500, 3, 3
if data_format == "NHWC":
ishape = [batch_size, h, w, in_channel]
stride = [1] + hw_stride + [1]
elif data_format == "NCHW":
ishape = [batch_size, in_channel, h, w]
stride = [1, 1] + hw_stride
else:
raise ValueError("Unknown data_format: " + str(data_format))
fshape = [fh, fw, in_channel, out_channel]
num_iters = 80
warmup_iters = 2
self._run_graph("gpu", dtype, data_format, ishape, fshape, stride,
padding, num_iters, warmup_iters)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/ops/conv2d_benchmark.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Implements the graph generation for computation of gradients."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_grad # pylint: disable=unused-import
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import check_ops # pylint: disable=unused-import
from tensorflow.python.ops import control_flow_grad # pylint: disable=unused-import
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import gradients_util
from tensorflow.python.ops import image_grad # pylint: disable=unused-import
from tensorflow.python.ops import linalg_grad # pylint: disable=unused-import
from tensorflow.python.ops import linalg_ops # pylint: disable=unused-import
from tensorflow.python.ops import logging_ops # pylint: disable=unused-import
from tensorflow.python.ops import manip_grad # pylint: disable=unused-import
from tensorflow.python.ops import math_grad # pylint: disable=unused-import
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import optional_grad # pylint: disable=unused-import
from tensorflow.python.ops import random_grad # pylint: disable=unused-import
from tensorflow.python.ops import tensor_array_ops
from tensorflow.python.ops.unconnected_gradients import UnconnectedGradients
from tensorflow.python.util.tf_export import tf_export
@tf_export(v1=["gradients"])
def gradients(ys,
xs,
grad_ys=None,
name="gradients",
colocate_gradients_with_ops=False,
gate_gradients=False,
aggregation_method=None,
stop_gradients=None,
unconnected_gradients=UnconnectedGradients.NONE):
"""Constructs symbolic derivatives of sum of `ys` w.r.t. x in `xs`.
`ys` and `xs` are each a `Tensor` or a list of tensors. `grad_ys`
is a list of `Tensor`, holding the gradients received by the
`ys`. The list must be the same length as `ys`.
`gradients()` adds ops to the graph to output the derivatives of `ys` with
respect to `xs`. It returns a list of `Tensor` of length `len(xs)` where
each tensor is the `sum(dy/dx)` for y in `ys`.
`grad_ys` is a list of tensors of the same length as `ys` that holds
the initial gradients for each y in `ys`. When `grad_ys` is None,
we fill in a tensor of '1's of the shape of y for each y in `ys`. A
user can provide their own initial `grad_ys` to compute the
derivatives using a different initial gradient for each y (e.g., if
one wanted to weight the gradient differently for each value in
each y).
`stop_gradients` is a `Tensor` or a list of tensors to be considered constant
with respect to all `xs`. These tensors will not be backpropagated through,
as though they had been explicitly disconnected using `stop_gradient`. Among
other things, this allows computation of partial derivatives as opposed to
total derivatives. For example:
```python
a = tf.constant(0.)
b = 2 * a
g = tf.gradients(a + b, [a, b], stop_gradients=[a, b])
```
Here the partial derivatives `g` evaluate to `[1.0, 1.0]`, compared to the
total derivatives `tf.gradients(a + b, [a, b])`, which take into account the
influence of `a` on `b` and evaluate to `[3.0, 1.0]`. Note that the above is
equivalent to:
```python
a = tf.stop_gradient(tf.constant(0.))
b = tf.stop_gradient(2 * a)
g = tf.gradients(a + b, [a, b])
```
`stop_gradients` provides a way of stopping gradient after the graph has
already been constructed, as compared to `tf.stop_gradient` which is used
during graph construction. When the two approaches are combined,
backpropagation stops at both `tf.stop_gradient` nodes and nodes in
`stop_gradients`, whichever is encountered first.
All integer tensors are considered constant with respect to all `xs`, as if
they were included in `stop_gradients`.
`unconnected_gradients` determines the value returned for each x in xs if it
is unconnected in the graph to ys. By default this is None to safeguard
against errors. MAthematically these gradients are zero which can be requested
using the `'zero'` option. `tf.UnconnectedGradients` provides the
following options and behaviors:
```python
a = tf.ones([1, 2])
b = tf.ones([3, 1])
g1 = tf.gradients([b], [a], unnconnected_gradients='none')
sess.run(g1) # [None]
g2 = tf.gradients([b], [a], unconnected_gradients='zero')
sess.run(g2) # [array([[0., 0.]], dtype=float32)]
```
Args:
ys: A `Tensor` or list of tensors to be differentiated.
xs: A `Tensor` or list of tensors to be used for differentiation.
grad_ys: Optional. A `Tensor` or list of tensors the same size as
`ys` and holding the gradients computed for each y in `ys`.
name: Optional name to use for grouping all the gradient ops together.
defaults to 'gradients'.
colocate_gradients_with_ops: If True, try colocating gradients with
the corresponding op.
gate_gradients: If True, add a tuple around the gradients returned
for an operations. This avoids some race conditions.
aggregation_method: Specifies the method used to combine gradient terms.
Accepted values are constants defined in the class `AggregationMethod`.
stop_gradients: Optional. A `Tensor` or list of tensors not to differentiate
through.
unconnected_gradients: Optional. Specifies the gradient value returned when
the given input tensors are unconnected. Accepted values are constants
defined in the class `tf.UnconnectedGradients` and the default value is
`none`.
Returns:
A list of `sum(dy/dx)` for each x in `xs`.
Raises:
LookupError: if one of the operations between `x` and `y` does not
have a registered gradient function.
ValueError: if the arguments are invalid.
RuntimeError: if called in Eager mode.
"""
# Creating the gradient graph for control flow mutates Operations.
# _mutation_lock ensures a Session.run call cannot occur between creating and
# mutating new ops.
# pylint: disable=protected-access
with ops.get_default_graph()._mutation_lock():
return gradients_util._GradientsHelper(
ys, xs, grad_ys, name, colocate_gradients_with_ops,
gate_gradients, aggregation_method, stop_gradients,
unconnected_gradients)
# pylint: enable=protected-access
@tf_export("gradients", v1=[])
def gradients_v2(ys, # pylint: disable=invalid-name
xs,
grad_ys=None,
name="gradients",
gate_gradients=False,
aggregation_method=None,
stop_gradients=None,
unconnected_gradients=UnconnectedGradients.NONE):
"""Constructs symbolic derivatives of sum of `ys` w.r.t. x in `xs`.
`ys` and `xs` are each a `Tensor` or a list of tensors. `grad_ys`
is a list of `Tensor`, holding the gradients received by the
`ys`. The list must be the same length as `ys`.
`gradients()` adds ops to the graph to output the derivatives of `ys` with
respect to `xs`. It returns a list of `Tensor` of length `len(xs)` where
each tensor is the `sum(dy/dx)` for y in `ys`.
`grad_ys` is a list of tensors of the same length as `ys` that holds
the initial gradients for each y in `ys`. When `grad_ys` is None,
we fill in a tensor of '1's of the shape of y for each y in `ys`. A
user can provide their own initial `grad_ys` to compute the
derivatives using a different initial gradient for each y (e.g., if
one wanted to weight the gradient differently for each value in
each y).
`stop_gradients` is a `Tensor` or a list of tensors to be considered constant
with respect to all `xs`. These tensors will not be backpropagated through,
as though they had been explicitly disconnected using `stop_gradient`. Among
other things, this allows computation of partial derivatives as opposed to
total derivatives. For example:
```python
a = tf.constant(0.)
b = 2 * a
g = tf.gradients(a + b, [a, b], stop_gradients=[a, b])
```
Here the partial derivatives `g` evaluate to `[1.0, 1.0]`, compared to the
total derivatives `tf.gradients(a + b, [a, b])`, which take into account the
influence of `a` on `b` and evaluate to `[3.0, 1.0]`. Note that the above is
equivalent to:
```python
a = tf.stop_gradient(tf.constant(0.))
b = tf.stop_gradient(2 * a)
g = tf.gradients(a + b, [a, b])
```
`stop_gradients` provides a way of stopping gradient after the graph has
already been constructed, as compared to `tf.stop_gradient` which is used
during graph construction. When the two approaches are combined,
backpropagation stops at both `tf.stop_gradient` nodes and nodes in
`stop_gradients`, whichever is encountered first.
All integer tensors are considered constant with respect to all `xs`, as if
they were included in `stop_gradients`.
`unconnected_gradients` determines the value returned for each x in xs if it
is unconnected in the graph to ys. By default this is None to safeguard
against errors. Mathematically these gradients are zero which can be requested
using the `'zero'` option. `tf.UnconnectedGradients` provides the
following options and behaviors:
```python
a = tf.ones([1, 2])
b = tf.ones([3, 1])
g1 = tf.gradients([b], [a], unnconnected_gradients='none')
sess.run(g1) # [None]
g2 = tf.gradients([b], [a], unconnected_gradients='zero')
sess.run(g2) # [array([[0., 0.]], dtype=float32)]
```
Args:
ys: A `Tensor` or list of tensors to be differentiated.
xs: A `Tensor` or list of tensors to be used for differentiation.
grad_ys: Optional. A `Tensor` or list of tensors the same size as
`ys` and holding the gradients computed for each y in `ys`.
name: Optional name to use for grouping all the gradient ops together.
defaults to 'gradients'.
gate_gradients: If True, add a tuple around the gradients returned
for an operations. This avoids some race conditions.
aggregation_method: Specifies the method used to combine gradient terms.
Accepted values are constants defined in the class `AggregationMethod`.
stop_gradients: Optional. A `Tensor` or list of tensors not to differentiate
through.
unconnected_gradients: Optional. Specifies the gradient value returned when
the given input tensors are unconnected. Accepted values are constants
defined in the class `tf.UnconnectedGradients` and the default value is
`none`.
Returns:
A list of `sum(dy/dx)` for each x in `xs`.
Raises:
LookupError: if one of the operations between `x` and `y` does not
have a registered gradient function.
ValueError: if the arguments are invalid.
RuntimeError: if called in Eager mode.
"""
# Creating the gradient graph for control flow mutates Operations.
# _mutation_lock ensures a Session.run call cannot occur between creating and
# mutating new ops.
# pylint: disable=protected-access
with ops.get_default_graph()._mutation_lock():
return gradients_util._GradientsHelper(
ys, xs, grad_ys, name, True, gate_gradients,
aggregation_method, stop_gradients,
unconnected_gradients)
# pylint: enable=protected-access
# TODO(vrv): Make this available when we want to make it public.
def _hessian_vector_product(ys, xs, v):
"""Multiply the Hessian of `ys` wrt `xs` by `v`.
This is an efficient construction that uses a backprop-like approach
to compute the product between the Hessian and another vector. The
Hessian is usually too large to be explicitly computed or even
represented, but this method allows us to at least multiply by it
for the same big-O cost as backprop.
Implicit Hessian-vector products are the main practical, scalable way
of using second derivatives with neural networks. They allow us to
do things like construct Krylov subspaces and approximate conjugate
gradient descent.
Example: if `y` = 1/2 `x`^T A `x`, then `hessian_vector_product(y,
x, v)` will return an expression that evaluates to the same values
as (A + A.T) `v`.
Args:
ys: A scalar value, or a tensor or list of tensors to be summed to
yield a scalar.
xs: A list of tensors that we should construct the Hessian over.
v: A list of tensors, with the same shapes as xs, that we want to
multiply by the Hessian.
Returns:
A list of tensors (or if the list would be length 1, a single tensor)
containing the product between the Hessian and `v`.
Raises:
ValueError: `xs` and `v` have different length.
"""
# Validate the input
length = len(xs)
if len(v) != length:
raise ValueError("xs and v must have the same length.")
# First backprop
grads = gradients(ys, xs)
assert len(grads) == length
elemwise_products = [
math_ops.multiply(grad_elem, array_ops.stop_gradient(v_elem))
for grad_elem, v_elem in zip(grads, v)
if grad_elem is not None
]
# Second backprop
return gradients(elemwise_products, xs)
@tf_export(v1=["hessians"])
def hessians(ys,
xs,
name="hessians",
colocate_gradients_with_ops=False,
gate_gradients=False,
aggregation_method=None):
"""Constructs the Hessian of sum of `ys` with respect to `x` in `xs`.
`hessians()` adds ops to the graph to output the Hessian matrix of `ys`
with respect to `xs`. It returns a list of `Tensor` of length `len(xs)`
where each tensor is the Hessian of `sum(ys)`.
The Hessian is a matrix of second-order partial derivatives of a scalar
tensor (see https://en.wikipedia.org/wiki/Hessian_matrix for more details).
Args:
ys: A `Tensor` or list of tensors to be differentiated.
xs: A `Tensor` or list of tensors to be used for differentiation.
name: Optional name to use for grouping all the gradient ops together.
defaults to 'hessians'.
colocate_gradients_with_ops: See `gradients()` documentation for details.
gate_gradients: See `gradients()` documentation for details.
aggregation_method: See `gradients()` documentation for details.
Returns:
A list of Hessian matrices of `sum(ys)` for each `x` in `xs`.
Raises:
LookupError: if one of the operations between `xs` and `ys` does not
have a registered gradient function.
"""
xs = gradients_util._AsList(xs) # pylint: disable=protected-access
kwargs = {
"colocate_gradients_with_ops": colocate_gradients_with_ops,
"gate_gradients": gate_gradients,
"aggregation_method": aggregation_method
}
# Compute first-order derivatives and iterate for each x in xs.
hessians = []
_gradients = gradients(ys, xs, **kwargs)
for gradient, x in zip(_gradients, xs):
# change shape to one-dimension without graph branching
gradient = array_ops.reshape(gradient, [-1])
# Declare an iterator and tensor array loop variables for the gradients.
n = array_ops.size(x)
loop_vars = [
array_ops.constant(0, dtypes.int32),
tensor_array_ops.TensorArray(x.dtype, n)
]
# Iterate over all elements of the gradient and compute second order
# derivatives.
_, hessian = control_flow_ops.while_loop(
lambda j, _: j < n,
lambda j, result: (j + 1,
result.write(j, gradients(gradient[j], x)[0])),
loop_vars
)
_shape = array_ops.shape(x)
_reshaped_hessian = array_ops.reshape(hessian.stack(),
array_ops.concat((_shape, _shape), 0))
hessians.append(_reshaped_hessian)
return hessians
@tf_export("hessians", v1=[])
def HessiansV2(ys,
xs,
gate_gradients=False,
aggregation_method=None,
name="hessians"):
return hessians(ys, xs, name=name, gate_gradients=gate_gradients,
aggregation_method=aggregation_method)
HessiansV2.__doc__ = hessians.__doc__
|
tensorflow-master
|
tensorflow/python/ops/gradients_impl.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Gradients for operators defined in linalg_ops.py.
Useful reference for derivative formulas is
An extended collection of matrix derivative results for forward and reverse
mode algorithmic differentiation by Mike Giles:
http://eprints.maths.ox.ac.uk/1079/1/NA-08-01.pdf
A detailed derivation of formulas for backpropagating through spectral layers
(SVD and Eig) by Ionescu, Vantzos & Sminchisescu:
https://arxiv.org/pdf/1509.07838v4.pdf
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import linalg_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops.linalg import linalg_impl as _linalg
@ops.RegisterGradient("MatrixInverse")
def _MatrixInverseGrad(op, grad):
"""Gradient for MatrixInverse."""
ainv = op.outputs[0]
return -math_ops.matmul(
ainv, math_ops.matmul(grad, ainv, adjoint_b=True), adjoint_a=True)
@ops.RegisterGradient("MatrixDeterminant")
def _MatrixDeterminantGrad(op, grad):
"""Gradient for MatrixDeterminant."""
a = op.inputs[0]
c = op.outputs[0]
a_adj_inv = linalg_ops.matrix_inverse(a, adjoint=True)
multipliers = array_ops.reshape(grad * c,
array_ops.concat([array_ops.shape(c), [1, 1]],
0))
return multipliers * a_adj_inv
@ops.RegisterGradient("MatrixSquareRoot")
def _MatrixSquareRootGrad(op, grad):
"""Gradient for MatrixSquareRoot."""
# Let A be an m x m square matrix (or batch of matrices)
# Let R = sqrtm(A)
# By definition, A = RR
# Take the differential: dA = d(RR) = RdR + dRR
# Solve the resulting Sylvester equation for dR
# Used to find Kronecker products within the Sylvester equation
def _KroneckerProduct(b1, b2):
"""Computes the Kronecker product of two batches of square matrices"""
b1_shape = array_ops.shape(b1)
b2_shape = array_ops.shape(b2)
b1_order = b1_shape[-1]
b2_order = b2_shape[-1]
shape_slice_size = [math_ops.subtract(array_ops.size(b1_shape), 2)]
shape_slice = array_ops.slice(b1_shape, [0],
shape_slice_size) # Same for both batches
b1_reshape_shape = array_ops.concat(
[shape_slice, [b1_order], [1], [b1_order], [1]], 0)
b2_reshape_shape = array_ops.concat(
[shape_slice, [1], [b2_order], [1], [b2_order]], 0)
b1_reshape = array_ops.reshape(b1, b1_reshape_shape)
b2_reshape = array_ops.reshape(b2, b2_reshape_shape)
order_prod = b1_order * b2_order
kprod_shape = array_ops.concat([shape_slice, [order_prod], [order_prod]], 0)
return array_ops.reshape(b1_reshape * b2_reshape, kprod_shape)
sqrtm = op.outputs[0] # R
shape = array_ops.shape(sqrtm)
order = shape[-1] # m
matrix_count = math_ops.reduce_prod(shape[0:-2])
# Get batch of m x m identity matrices
eye = linalg_ops.eye(order, dtype=sqrtm.dtype) # m x m identity matrix
eye_flat = array_ops.reshape(eye, [-1])
eye_tiled = array_ops.tile(eye_flat, [matrix_count])
eye_batch = array_ops.reshape(eye_tiled, shape)
# The transpose of R is taken in the k1 term instead of k2 in
# order to prevent redundant transposition of R (i.e. (R')' = R)
sqrtm_transpose = array_ops.matrix_transpose(sqrtm)
k1 = _KroneckerProduct(eye_batch, sqrtm_transpose)
k2 = _KroneckerProduct(sqrtm, eye_batch)
ksum = math_ops.add(k1, k2)
# Vectorize dA
shape_slice_size = [math_ops.subtract(array_ops.size(shape), 2)]
shape_slice = array_ops.slice(shape, [0], shape_slice_size)
shape_vec_da = array_ops.concat([shape_slice, [order * order], [1]], 0)
vec_da = array_ops.reshape(array_ops.matrix_transpose(grad), shape_vec_da)
# Solve for vec(dR)
vec_dsqrtm = linalg_ops.matrix_solve(ksum, vec_da)
# Solve for dR by inverse vectorizing vec(dR)
dsqrtm_transpose = array_ops.reshape(vec_dsqrtm, shape)
return array_ops.matrix_transpose(dsqrtm_transpose)
@ops.RegisterGradient("LogMatrixDeterminant")
def _LogMatrixDeterminantGrad(op, _, grad_b):
"""Gradient for LogMatrixDeterminant."""
a = op.inputs[0]
c = op.outputs[1]
a_adj_inv = linalg_ops.matrix_inverse(a, adjoint=True)
multipliers = array_ops.reshape(
grad_b, array_ops.concat([array_ops.shape(c), [1, 1]], 0))
return multipliers * a_adj_inv
@ops.RegisterGradient("Cholesky")
def _CholeskyGrad(op, grad):
"""Gradient for Cholesky."""
# Gradient is l^{-H} @ ((l^{H} @ grad) * (tril(ones)-1/2*eye)) @ l^{-1}
l = op.outputs[0]
num_rows = array_ops.shape(l)[-1]
batch_shape = array_ops.shape(l)[:-2]
l_inverse = linalg_ops.matrix_triangular_solve(l,
linalg_ops.eye(
num_rows,
batch_shape=batch_shape,
dtype=l.dtype))
middle = math_ops.matmul(l, grad, adjoint_a=True)
middle = array_ops.matrix_set_diag(middle,
0.5 * array_ops.matrix_diag_part(middle))
middle = array_ops.matrix_band_part(middle, -1, 0)
grad_a = math_ops.matmul(
math_ops.matmul(l_inverse, middle, adjoint_a=True), l_inverse)
grad_a += _linalg.adjoint(grad_a)
return grad_a * 0.5
@ops.RegisterGradient("Qr")
def _QrGrad(op, dq, dr):
"""Gradient for Qr."""
q, r = op.outputs
if q.dtype.is_complex:
raise NotImplementedError("QrGrad not implemented for dtype: %s" % q.dtype)
if (r.shape.ndims is None or r.shape.as_list()[-2] is None or
r.shape.as_list()[-1] is None):
raise NotImplementedError("QrGrad not implemented with dynamic shapes.")
if r.shape.dims[-2].value != r.shape.dims[-1].value:
raise NotImplementedError("QrGrad not implemented when ncols > nrows "
"or full_matrices is true and ncols != nrows.")
qdq = math_ops.matmul(q, dq, adjoint_a=True)
qdq_ = qdq - _linalg.adjoint(qdq)
rdr = math_ops.matmul(r, dr, adjoint_b=True)
rdr_ = rdr - _linalg.adjoint(rdr)
tril = array_ops.matrix_band_part(qdq_ + rdr_, -1, 0)
def _TriangularSolve(x, r):
"""Equiv to matmul(x, adjoint(matrix_inverse(r))) if r is upper-tri."""
return _linalg.adjoint(
linalg_ops.matrix_triangular_solve(
r, _linalg.adjoint(x), lower=False, adjoint=False))
grad_a = math_ops.matmul(q, dr + _TriangularSolve(tril, r))
grad_b = _TriangularSolve(dq - math_ops.matmul(q, qdq), r)
return grad_a + grad_b
@ops.RegisterGradient("MatrixSolve")
def _MatrixSolveGrad(op, grad):
"""Gradient for MatrixSolve."""
a = op.inputs[0]
adjoint_a = op.get_attr("adjoint")
c = op.outputs[0]
grad_b = linalg_ops.matrix_solve(a, grad, adjoint=not adjoint_a)
if adjoint_a:
grad_a = -math_ops.matmul(c, grad_b, adjoint_b=True)
else:
grad_a = -math_ops.matmul(grad_b, c, adjoint_b=True)
return (grad_a, grad_b)
@ops.RegisterGradient("MatrixSolveLs")
def _MatrixSolveLsGrad(op, grad):
"""Gradients for MatrixSolveLs."""
# TODO(rmlarsen): The implementation could be more efficient:
# a) Output the Cholesky factorization from forward op instead of
# recomputing it here.
# b) Implement a symmetric rank-k update op instead of computing
# x*z + transpose(x*z). This pattern occurs other places in TensorFlow.
def _Overdetermined(op, grad):
"""Gradients for the overdetermined case of MatrixSolveLs.
This is the backprop for the solution to the normal equations of the first
kind:
X = F(A, B) = (A^T * A + lambda * I)^{-1} * A^T * B
which solve the least squares problem
min ||A * X - B||_F^2 + lambda ||X||_F^2.
"""
a = op.inputs[0]
b = op.inputs[1]
x = op.outputs[0]
l2_regularizer = math_ops.cast(op.inputs[2], a.dtype.base_dtype)
# pylint: disable=protected-access
chol = linalg_ops._RegularizedGramianCholesky(
a, l2_regularizer=l2_regularizer, first_kind=True)
# pylint: enable=protected-access
# Temporary z = (A^T * A + lambda * I)^{-1} * grad.
z = linalg_ops.cholesky_solve(chol, grad)
xzt = math_ops.matmul(x, z, adjoint_b=True)
zx_sym = xzt + array_ops.matrix_transpose(xzt)
grad_a = -math_ops.matmul(a, zx_sym) + math_ops.matmul(b, z, adjoint_b=True)
grad_b = math_ops.matmul(a, z)
return (grad_a, grad_b, None)
def _Underdetermined(op, grad):
"""Gradients for the underdetermined case of MatrixSolveLs.
This is the backprop for the solution to the normal equations of the second
kind:
X = F(A, B) = A * (A*A^T + lambda*I)^{-1} * B
that (for lambda=0) solve the least squares problem
min ||X||_F subject to A*X = B.
"""
a = op.inputs[0]
b = op.inputs[1]
l2_regularizer = math_ops.cast(op.inputs[2], a.dtype.base_dtype)
# pylint: disable=protected-access
chol = linalg_ops._RegularizedGramianCholesky(
a, l2_regularizer=l2_regularizer, first_kind=False)
# pylint: enable=protected-access
grad_b = linalg_ops.cholesky_solve(chol, math_ops.matmul(a, grad))
# Temporary tmp = (A * A^T + lambda * I)^{-1} * B.
tmp = linalg_ops.cholesky_solve(chol, b)
a1 = math_ops.matmul(tmp, a, adjoint_a=True)
a1 = -math_ops.matmul(grad_b, a1)
a2 = grad - math_ops.matmul(a, grad_b, adjoint_a=True)
a2 = math_ops.matmul(tmp, a2, adjoint_b=True)
grad_a = a1 + a2
return (grad_a, grad_b, None)
fast = op.get_attr("fast")
if fast is False:
raise ValueError("Gradient not defined for fast=False")
matrix_shape = op.inputs[0].get_shape()[-2:]
if matrix_shape.is_fully_defined():
if matrix_shape[-2] >= matrix_shape[-1]:
return _Overdetermined(op, grad)
else:
return _Underdetermined(op, grad)
else:
# We have to defer determining the shape to runtime and use
# conditional execution of the appropriate graph.
matrix_shape = array_ops.shape(op.inputs[0])[-2:]
return control_flow_ops.cond(matrix_shape[-2] >= matrix_shape[-1],
lambda: _Overdetermined(op, grad),
lambda: _Underdetermined(op, grad))
@ops.RegisterGradient("MatrixTriangularSolve")
def _MatrixTriangularSolveGrad(op, grad):
"""Gradient for MatrixTriangularSolve."""
a = op.inputs[0]
adjoint_a = op.get_attr("adjoint")
lower_a = op.get_attr("lower")
c = op.outputs[0]
grad_b = linalg_ops.matrix_triangular_solve(
a, grad, lower=lower_a, adjoint=not adjoint_a)
if adjoint_a:
grad_a = -math_ops.matmul(c, grad_b, adjoint_b=True)
else:
grad_a = -math_ops.matmul(grad_b, c, adjoint_b=True)
if lower_a:
grad_a = array_ops.matrix_band_part(grad_a, -1, 0)
else:
grad_a = array_ops.matrix_band_part(grad_a, 0, -1)
return (grad_a, grad_b)
@ops.RegisterGradient("SelfAdjointEigV2")
def _SelfAdjointEigV2Grad(op, grad_e, grad_v):
"""Gradient for SelfAdjointEigV2."""
e = op.outputs[0]
compute_v = op.get_attr("compute_v")
# a = op.inputs[0], which satisfies
# a[...,:,:] * v[...,:,i] = e[...,i] * v[...,i]
with ops.control_dependencies([grad_e, grad_v]):
if compute_v:
v = op.outputs[1]
# Construct the matrix f(i,j) = (i != j ? 1 / (e_i - e_j) : 0).
# Notice that because of the term involving f, the gradient becomes
# infinite (or NaN in practice) when eigenvalues are not unique.
# Mathematically this should not be surprising, since for (k-fold)
# degenerate eigenvalues, the corresponding eigenvectors are only defined
# up to arbitrary rotation in a (k-dimensional) subspace.
f = array_ops.matrix_set_diag(
math_ops.reciprocal(
array_ops.expand_dims(e, -2) - array_ops.expand_dims(e, -1)),
array_ops.zeros_like(e))
grad_a = math_ops.matmul(
v,
math_ops.matmul(
array_ops.matrix_diag(grad_e) +
f * math_ops.matmul(v, grad_v, adjoint_a=True),
v,
adjoint_b=True))
else:
_, v = linalg_ops.self_adjoint_eig(op.inputs[0])
grad_a = math_ops.matmul(v,
math_ops.matmul(
array_ops.matrix_diag(grad_e),
v,
adjoint_b=True))
# The forward op only depends on the lower triangular part of a, so here we
# symmetrize and take the lower triangle
grad_a = array_ops.matrix_band_part(grad_a + _linalg.adjoint(grad_a), -1, 0)
grad_a = array_ops.matrix_set_diag(grad_a,
0.5 * array_ops.matrix_diag_part(grad_a))
return grad_a
@ops.RegisterGradient("Svd")
def _SvdGrad(op, grad_s, grad_u, grad_v):
"""Gradient for the singular value decomposition."""
# The derivation for the compute_uv=False case, and most of
# the derivation for the full_matrices=True case, are in
# Giles' paper (see reference at top of file). A derivation for
# the full_matrices=False case is available at
# https://j-towns.github.io/papers/svd-derivative.pdf
a = op.inputs[0]
a_shape = a.get_shape().with_rank_at_least(2)
grad_s_mat = array_ops.matrix_diag(grad_s)
if not op.get_attr("compute_uv"):
s, u, v = linalg_ops.svd(a, compute_uv=True)
grad_a = math_ops.matmul(u, math_ops.matmul(grad_s_mat, v, adjoint_b=True))
grad_a.set_shape(a_shape)
return grad_a
full_matrices = op.get_attr("full_matrices")
# TODO(rmlarsen): Make this work with complex types.
if a.dtype.is_complex:
raise NotImplementedError(
"SVD gradient is not implemented for complex types and "
"compute_uv=True.")
grad_u_shape = grad_u.get_shape().with_rank_at_least(2)
grad_v_shape = grad_v.get_shape().with_rank_at_least(2)
m = a_shape.dims[-2].merge_with(grad_u_shape[-2])
n = a_shape.dims[-1].merge_with(grad_v_shape[-2])
batch_shape = a_shape[:-2].merge_with(grad_u_shape[:-2]).merge_with(
grad_v_shape[:-2])
a_shape = batch_shape.concatenate([m, n])
m = a_shape.dims[-2].value
n = a_shape.dims[-1].value
# TODO(rmlarsen): Make this work with placeholders.
if m is None or n is None:
raise NotImplementedError(
"SVD gradient has not been implemented for input with unknown "
"inner matrix shape.")
s = op.outputs[0]
u = op.outputs[1]
v = op.outputs[2]
use_adjoint = False
if m > n:
# Compute the gradient for A^H = V * S^T * U^H, and (implicitly) take the
# Hermitian transpose of the gradient at the end.
use_adjoint = True
m, n = n, m
u, v = v, u
grad_u, grad_v = grad_v, grad_u
with ops.control_dependencies([grad_s, grad_u, grad_v]):
if full_matrices and abs(m - n) > 1:
raise NotImplementedError(
"svd gradient is not implemented for abs(m - n) > 1 "
"when full_matrices is True")
s_mat = array_ops.matrix_diag(s)
s2 = math_ops.square(s)
# NOTICE: Because of the term involving f, the gradient becomes
# infinite (or NaN in practice) when singular values are not unique.
# Mathematically this should not be surprising, since for (k-fold)
# degenerate singular values, the corresponding singular vectors are
# only defined up a (k-dimensional) subspace. In practice, this can
# lead to numerical instability when singular values are close but not
# exactly equal.
# Also, even with distinct singular values, the diagonal of f can have Inf
# values before setting to zero, which hurt when differentiating through
# this op. To avoid that, we add eye to the matrix before taking
# the reciprocal.
s_shape = array_ops.shape(s)
eye = _linalg.eye(s_shape[-1], batch_shape=s_shape[:-1], dtype=s.dtype)
f = array_ops.matrix_set_diag(
math_ops.reciprocal(
array_ops.expand_dims(s2, -2) - array_ops.expand_dims(s2, -1) +
eye), array_ops.zeros_like(s))
s_inv_mat = array_ops.matrix_diag(math_ops.reciprocal(s))
v1 = v[..., :, :m]
grad_v1 = grad_v[..., :, :m]
u_gu = math_ops.matmul(u, grad_u, adjoint_a=True)
v_gv = math_ops.matmul(v1, grad_v1, adjoint_a=True)
f_u = f * u_gu
f_v = f * v_gv
term1_nouv = (
grad_s_mat + math_ops.matmul(f_u + _linalg.adjoint(f_u), s_mat) +
math_ops.matmul(s_mat, f_v + _linalg.adjoint(f_v)))
term1 = math_ops.matmul(u, math_ops.matmul(term1_nouv, v1, adjoint_b=True))
if m == n:
grad_a_before_transpose = term1
else:
gv1t = array_ops.matrix_transpose(grad_v1)
gv1t_v1 = math_ops.matmul(gv1t, v1)
term2_nous = gv1t - math_ops.matmul(gv1t_v1, v1, adjoint_b=True)
if full_matrices:
v2 = v[..., :, m:n]
grad_v2 = grad_v[..., :, m:n]
v1t_gv2 = math_ops.matmul(v1, grad_v2, adjoint_a=True)
term2_nous -= math_ops.matmul(v1t_gv2, v2, adjoint_b=True)
u_s_inv = math_ops.matmul(u, s_inv_mat)
term2 = math_ops.matmul(u_s_inv, term2_nous)
grad_a_before_transpose = term1 + term2
if use_adjoint:
grad_a = array_ops.matrix_transpose(grad_a_before_transpose)
else:
grad_a = grad_a_before_transpose
grad_a.set_shape(a_shape)
return grad_a
def _LeftShift(x):
"""Shifts next-to-last dimension to the left, adding zero on the right."""
rank = array_ops.rank(x)
zeros = array_ops.zeros((rank - 2, 2), dtype=dtypes.int32)
pad = array_ops.concat([zeros, array_ops.constant([[0, 1], [0, 0]])], axis=0)
return array_ops.pad(x[..., 1:, :], pad)
def _RightShift(x):
"""Shifts next-to-last dimension to the right, adding zero on the left."""
rank = array_ops.rank(x)
zeros = array_ops.zeros((rank - 2, 2), dtype=dtypes.int32)
pad = array_ops.concat([zeros, array_ops.constant([[1, 0], [0, 0]])], axis=0)
return array_ops.pad(x[..., :-1, :], pad)
@ops.RegisterGradient("TridiagonalMatMul")
def _TridiagonalMatMulGrad(op, grad):
"""Gradient for TridiagonalMatMul."""
superdiag_conj = array_ops.matrix_transpose(op.inputs[0], conjugate=True)
maindiag_conj = array_ops.matrix_transpose(op.inputs[1], conjugate=True)
subdiag_conj = array_ops.matrix_transpose(op.inputs[2], conjugate=True)
rhs_conj = math_ops.conj(op.inputs[3])
superdiag_grad = math_ops.reduce_sum(_LeftShift(rhs_conj) * grad, axis=-1)
maindiag_grad = math_ops.reduce_sum(rhs_conj * grad, axis=-1)
subdiag_grad = math_ops.reduce_sum(_RightShift(rhs_conj) * grad, axis=-1)
rhs_grad = _RightShift(superdiag_conj * grad) + \
maindiag_conj * grad + _LeftShift(subdiag_conj * grad)
superdiag_grad = array_ops.expand_dims(superdiag_grad, -2)
maindiag_grad = array_ops.expand_dims(maindiag_grad, -2)
subdiag_grad = array_ops.expand_dims(subdiag_grad, -2)
return superdiag_grad, maindiag_grad, subdiag_grad, rhs_grad
@ops.RegisterGradient("TridiagonalSolve")
def _TridiagonalSolveGrad(op, grad):
"""Gradient for TridiagonalSolveGrad."""
diags = op.inputs[0]
x = op.outputs[0]
# Transposing the matrix within tridiagonal_solve kernel by interchanging
# superdiagonal and subdiagonal wouldn't work on GPU due to mismatch with
# paddings required by cusparse*gtsv routines.
# So constructing the transposed matrix in Python.
diags_transposed = _TransposeTridiagonalMatrix(diags)
grad_rhs = linalg_ops.tridiagonal_solve(diags_transposed, grad)
grad_diags = -_MatmulExtractingThreeDiagonals(grad_rhs, x)
return grad_diags, grad_rhs
def _TransposeTridiagonalMatrix(diags):
"""Transposes a tridiagonal matrix.
Args:
diags: the diagonals of the input matrix in the compact form (see
linalg_ops.tridiagonal_solve).
Returns:
Diagonals of the transposed matrix in the compact form.
"""
diag = diags[..., 1, :]
if diags.shape.is_fully_defined():
# For fully defined tensor we can concat with a tensor of zeros, which is
# faster than using array_ops.pad().
zeros = array_ops.zeros(list(diags.shape[:-2]) + [1], dtype=diags.dtype)
superdiag = array_ops.concat((diags[..., 2, 1:], zeros), axis=-1)
subdiag = array_ops.concat((zeros, diags[..., 0, :-1]), axis=-1)
else:
rank = array_ops.rank(diags)
zeros = array_ops.zeros((rank - 2, 2), dtype=dtypes.int32)
superdiag_pad = array_ops.concat((zeros, array_ops.constant([[0, 1]])),
axis=0)
superdiag = array_ops.pad(diags[..., 2, 1:], superdiag_pad)
subdiag_pad = array_ops.concat((zeros, array_ops.constant([[1, 0]])),
axis=0)
subdiag = array_ops.pad(diags[..., 0, :-1], subdiag_pad)
return array_ops.stack([superdiag, diag, subdiag], axis=-2)
def _MatmulExtractingThreeDiagonals(x, y_tr):
"""Multiplies matrices and extracts three diagonals from the product.
With sizes M x K and K x M, this function takes O(MK) time and O(M) space,
while using math_ops.matmul, and then extracting the diagonals would take
O(M^2 K) time and O(M^2) space.
Args:
x: first matrix
y_tr: second matrix transposed
Returns:
Diagonals of the product in compact format (see
linalg_ops.tridiagonal_solve)
"""
diag = math_ops.reduce_sum(x * y_tr, axis=-1)
if y_tr.shape.is_fully_defined():
zeros = array_ops.zeros(
list(x.shape[:-2]) + [1, x.shape[-1]], dtype=x.dtype)
superdiag = math_ops.reduce_sum(
x * array_ops.concat((y_tr[..., 1:, :], zeros), axis=-2), axis=-1)
subdiag = math_ops.reduce_sum(
x * array_ops.concat((zeros, y_tr[..., :-1, :]), axis=-2), axis=-1)
else:
rank = array_ops.rank(y_tr)
zeros = array_ops.zeros((rank - 2, 2), dtype=dtypes.int32)
superdiag_pad = array_ops.concat(
(zeros, array_ops.constant([[0, 1], [0, 0]])), axis=0)
superdiag = math_ops.reduce_sum(
x * array_ops.pad(y_tr[..., 1:, :], superdiag_pad), axis=-1)
subdiag_pad = array_ops.concat(
(zeros, array_ops.constant([[1, 0], [0, 0]])), axis=0)
subdiag = math_ops.reduce_sum(
x * array_ops.pad(y_tr[..., :-1, :], subdiag_pad), axis=-1)
return array_ops.stack([superdiag, diag, subdiag], axis=-2)
|
tensorflow-master
|
tensorflow/python/ops/linalg_grad.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Gradients for operators defined in nn_ops.py."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.eager import backprop
from tensorflow.python.eager import context
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import tensor_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gen_nn_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import nn_ops
@ops.RegisterGradient("Conv2DBackpropInput")
def _Conv2DBackpropInputGrad(op, grad):
"""The derivatives for deconvolution.
Args:
op: the Deconvolution op.
grad: the tensor representing the gradient w.r.t. the output
Returns:
the gradients w.r.t. the input and the filter
"""
# We call the gen_nn_ops backprop functions instead of nn_ops backprop
# functions for performance reasons in Eager mode. See _Conv2DGrad.
return [
None,
gen_nn_ops.conv2d_backprop_filter(
grad,
array_ops.shape(op.inputs[1]),
op.inputs[2],
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
explicit_paddings=op.get_attr("explicit_paddings"),
use_cudnn_on_gpu=op.get_attr("use_cudnn_on_gpu"),
data_format=op.get_attr("data_format").decode()),
gen_nn_ops.conv2d(
grad,
op.inputs[1],
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
explicit_paddings=op.get_attr("explicit_paddings"),
use_cudnn_on_gpu=op.get_attr("use_cudnn_on_gpu"),
data_format=op.get_attr("data_format").decode())
]
@ops.RegisterGradient("Conv2DBackpropFilter")
def _Conv2DBackpropFilterGrad(op, grad):
# We call the gen_nn_ops backprop functions instead of nn_ops backprop
# functions for performance reasons in Eager mode. See _Conv2DGrad.
return [
gen_nn_ops.conv2d_backprop_input(
array_ops.shape(op.inputs[0]),
grad,
op.inputs[2],
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
explicit_paddings=op.get_attr("explicit_paddings"),
use_cudnn_on_gpu=op.get_attr("use_cudnn_on_gpu"),
data_format=op.get_attr("data_format").decode()), None,
gen_nn_ops.conv2d(
op.inputs[0],
grad,
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
explicit_paddings=op.get_attr("explicit_paddings"),
use_cudnn_on_gpu=op.get_attr("use_cudnn_on_gpu"),
data_format=op.get_attr("data_format").decode())
]
@ops.RegisterGradient("DepthwiseConv2dNativeBackpropInput")
def _DepthwiseConv2dNativeBackpropInputGrad(op, grad):
"""The derivatives for deconvolution.
Args:
op: the Deconvolution op.
grad: the tensor representing the gradient w.r.t. the output
Returns:
the gradients w.r.t. the input and the filter
"""
return [
None,
nn_ops.depthwise_conv2d_native_backprop_filter(
grad,
array_ops.shape(op.inputs[1]),
op.inputs[2],
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format")),
nn_ops.depthwise_conv2d_native(
grad,
op.inputs[1],
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format"))
]
@ops.RegisterGradient("DepthwiseConv2dNativeBackpropFilter")
def _DepthwiseConv2dNativeBackpropFilterGrad(op, grad):
return [
nn_ops.depthwise_conv2d_native_backprop_input(
array_ops.shape(op.inputs[0]),
grad,
op.inputs[2],
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format")), None,
nn_ops.depthwise_conv2d_native(
op.inputs[0],
grad,
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format"))
]
@ops.RegisterGradient("Conv3D")
def _Conv3DGrad(op, grad):
data_format = op.get_attr("data_format").decode()
return [
nn_ops.conv3d_backprop_input_v2(
array_ops.shape(op.inputs[0]),
op.inputs[1],
grad,
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=data_format),
nn_ops.conv3d_backprop_filter_v2(
op.inputs[0],
array_ops.shape(op.inputs[1]),
grad,
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=data_format)
]
@ops.RegisterGradient("Conv3DBackpropInputV2")
def _Conv3DBackpropInputGrad(op, grad):
data_format = op.get_attr("data_format").decode()
return [
None,
nn_ops.conv3d_backprop_filter_v2(
grad,
array_ops.shape(op.inputs[1]),
op.inputs[2],
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=data_format),
nn_ops.conv3d(
grad,
op.inputs[1],
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=data_format)
]
@ops.RegisterGradient("Conv3DBackpropFilterV2")
def _Conv3DBackpropFilterGrad(op, grad):
data_format = op.get_attr("data_format").decode()
return [
nn_ops.conv3d_backprop_input_v2(
array_ops.shape(op.inputs[0]),
grad,
op.inputs[2],
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=data_format), None,
nn_ops.conv3d(
op.inputs[0],
grad,
dilations=op.get_attr("dilations"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=data_format)
]
@ops.RegisterGradient("AvgPool3D")
def _AvgPool3DGrad(op, grad):
return gen_nn_ops.avg_pool3d_grad(
array_ops.shape(op.inputs[0]),
grad,
ksize=op.get_attr("ksize"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format").decode())
@ops.RegisterGradient("AvgPool3DGrad")
def _AvgPool3DGradGrad(op, grad):
return (array_ops.stop_gradient(op.inputs[0]),
gen_nn_ops.avg_pool3d(
grad,
op.get_attr("ksize"),
op.get_attr("strides"),
op.get_attr("padding"),
data_format=op.get_attr("data_format").decode()))
@ops.RegisterGradient("MaxPool3D")
def _MaxPool3DGrad(op, grad):
return gen_nn_ops.max_pool3d_grad(
op.inputs[0],
op.outputs[0],
grad,
ksize=op.get_attr("ksize"),
strides=op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format").decode())
@ops.RegisterGradient("MaxPool3DGrad")
def _MaxPool3DGradGrad(op, grad):
return (array_ops.zeros(
shape=array_ops.shape(op.inputs[0]), dtype=op.inputs[0].dtype),
array_ops.zeros(
shape=array_ops.shape(op.inputs[1]), dtype=op.inputs[1].dtype),
gen_nn_ops.max_pool3d_grad_grad(
op.inputs[0],
op.inputs[1],
grad,
op.get_attr("ksize"),
op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format").decode()))
@ops.RegisterGradient("MaxPool3DGradGrad")
def _MaxPool3DGradGradGrad(op, grad):
return (array_ops.zeros(
shape=array_ops.shape(op.inputs[0]), dtype=op.inputs[0].dtype),
array_ops.zeros(
shape=array_ops.shape(op.inputs[1]), dtype=op.inputs[1].dtype),
gen_nn_ops.max_pool3d_grad(
op.inputs[0],
op.inputs[1],
grad,
op.get_attr("ksize"),
op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format").decode()))
@ops.RegisterGradient("Softmax")
def _SoftmaxGrad(op, grad_softmax):
"""The derivative of the softmax nonlinearity.
We assume that probs is of shape [batch_size * dim]
The formula for dsoftmax / dx = (diag(softmax) - softmax * softmax').
This matrix is diagonal minus a rank one matrix, so it is easy to implement
as follows:
grad_x = grad_softmax * softmax - sum(grad_softmax * softmax) * softmax
Args:
op: the Softmax op.
grad_softmax: the tensor representing the gradient w.r.t. the softmax
output.
Returns:
gradient w.r.t the input to the softmax
"""
softmax = op.outputs[0]
sum_channels = math_ops.reduce_sum(grad_softmax * softmax, -1, keepdims=True)
return (grad_softmax - sum_channels) * softmax
@ops.RegisterGradient("LogSoftmax")
def _LogSoftmaxGrad(op, grad):
"""The gradient for log_softmax.
log_softmax = input - log(sum(exp(input))
dlog_softmax/dinput = diag - softmax(input)
Args:
op: The log softmax op.
grad: The tensor representing the gradient w.r.t. the output.
Returns:
The gradients w.r.t. the input.
"""
softmax = math_ops.exp(op.outputs[0])
return grad - math_ops.reduce_sum(grad, -1, keepdims=True) * softmax
@ops.RegisterGradient("BiasAdd")
def _BiasAddGrad(op, received_grad):
"""Return the gradients for the 2 inputs of bias_op.
The first input of unused_bias_op is the tensor t, and its gradient is
just the gradient the unused_bias_op received.
The second input of unused_bias_op is the bias vector which has one fewer
dimension than "received_grad" (the batch dimension.) Its gradient is the
received gradient Summed on the batch dimension, which is the first dimension.
Args:
op: The BiasOp for which we need to generate gradients.
received_grad: Tensor. The gradients passed to the BiasOp.
Returns:
Two tensors, the first one for the "tensor" input of the BiasOp,
the second one for the "bias" input of the BiasOp.
"""
try:
data_format = op.get_attr("data_format")
except ValueError:
data_format = None
return (received_grad,
gen_nn_ops.bias_add_grad(
out_backprop=received_grad, data_format=data_format))
@ops.RegisterGradient("BiasAddGrad")
def _BiasAddGradGrad(op, received_grad):
"""Gradient for the BiasAddGrad op.
Args:
op: BiasAddGrad op for which we are calculating gradients.
received_grad: The gradients passed to the BiasAddGrad op.
Returns:
A single gradient Tensor for the input to BiasAddGrad (which
is the gradient of the bias term in BiasAdd)
"""
try:
data_format = op.get_attr("data_format")
except ValueError:
data_format = None
shape = array_ops.shape(op.inputs[0])
bias_shape = array_ops.shape(received_grad)
if data_format == b"NCHW":
expanded_shape = array_ops.concat([
array_ops.ones_like(shape[:1]), bias_shape,
array_ops.ones_like(shape[2:])
], 0)
tile_mults = array_ops.concat([shape[:1], [1], shape[2:]], 0)
else:
expanded_shape = array_ops.concat(
[array_ops.ones_like(shape[:-1]), bias_shape], 0)
tile_mults = array_ops.concat([shape[:-1], [1]], 0)
expanded_grad = array_ops.reshape(received_grad, expanded_shape)
return array_ops.tile(expanded_grad, tile_mults)
@ops.RegisterGradient("BiasAddV1")
def _BiasAddGradV1(unused_bias_op, received_grad):
"""Return the gradients for the 2 inputs of bias_op.
The first input of unused_bias_op is the tensor t, and its gradient is
just the gradient the unused_bias_op received.
The second input of unused_bias_op is the bias vector which has one fewer
dimension than "received_grad" (the batch dimension.) Its gradient is the
received gradient Summed on the batch dimension, which is the first dimension.
Args:
unused_bias_op: The BiasOp for which we need to generate gradients.
received_grad: Tensor. The gradients passed to the BiasOp.
Returns:
Two tensors, the first one for the "tensor" input of the BiasOp,
the second one for the "bias" input of the BiasOp.
"""
reduction_dim_tensor = math_ops.range(array_ops.rank(received_grad) - 1)
return (received_grad, math_ops.reduce_sum(received_grad,
reduction_dim_tensor))
@ops.RegisterGradient("Relu")
def _ReluGrad(op, grad):
return gen_nn_ops.relu_grad(grad, op.outputs[0])
@ops.RegisterGradient("EluGrad")
def _EluGradGrad(op, grad):
elu_x = op.inputs[1]
return (gen_nn_ops.elu_grad(grad, elu_x),
array_ops.where(
elu_x < 0, grad * op.inputs[0], array_ops.zeros_like(elu_x)))
@ops.RegisterGradient("SeluGrad")
def _SeluGradGrad(op, grad):
selu_x = op.inputs[1]
return (gen_nn_ops.selu_grad(grad, selu_x),
array_ops.where(
selu_x < 0., grad * op.inputs[0], array_ops.zeros_like(selu_x)))
@ops.RegisterGradient("Relu6")
def _Relu6Grad(op, grad):
return gen_nn_ops.relu6_grad(grad, op.outputs[0])
@ops.RegisterGradient("Relu6Grad")
def _Relu6GradGrad(op, grad):
x = op.inputs[1]
return (gen_nn_ops.relu6_grad(grad, x),
array_ops.zeros(shape=array_ops.shape(x), dtype=x.dtype))
@ops.RegisterGradient("LeakyRelu")
def _LeakyReluGrad(op, grad):
x = op.inputs[0]
alpha = op.get_attr("alpha")
return gen_nn_ops.leaky_relu_grad(grad, x, alpha=alpha)
@ops.RegisterGradient("LeakyReluGrad")
def _LeakyReluGradGrad(op, grad):
x = op.inputs[1]
alpha = op.get_attr("alpha")
return (gen_nn_ops.leaky_relu_grad(grad, x, alpha=alpha),
array_ops.zeros(shape=array_ops.shape(x), dtype=x.dtype))
@ops.RegisterGradient("Elu")
def _EluGrad(op, grad):
return gen_nn_ops.elu_grad(grad, op.outputs[0])
@ops.RegisterGradient("Selu")
def _SeluGrad(op, grad):
return gen_nn_ops.selu_grad(grad, op.outputs[0])
@ops.RegisterGradient("Softplus")
def _SoftplusGrad(op, grad):
return grad * math_ops.sigmoid(op.inputs[0])
@ops.RegisterGradient("SoftplusGrad")
def _SoftplusGradGrad(op, grad):
# Let:
# y = tf.nn.softplus(x)
# dx = gen_nn_ops.softplus_grad(dy, x) = dy / (1 + exp(-x))
# This op computes (ddy, d2x) from op.inputs == [dy, x] and grad == ddx.
dy, x = op.inputs
with ops.control_dependencies([grad]):
ddy = gen_nn_ops.softplus_grad(grad, x)
d2x = grad * dy / (math_ops.exp(-x) + 2.0 + math_ops.exp(x))
return (ddy, d2x)
@ops.RegisterGradient("Softsign")
def _SoftsignGrad(op, grad):
return gen_nn_ops.softsign_grad(grad, op.inputs[0])
@ops.RegisterGradient("ReluGrad")
def _ReluGradGrad(op, grad):
x = op.inputs[1]
return (gen_nn_ops.relu_grad(grad, x),
array_ops.zeros(shape=array_ops.shape(x), dtype=x.dtype))
def _BroadcastMul(vec, mat):
"""Multiply after broadcasting vec to match dimensions of mat.
Args:
vec: A 1-D tensor of dimension [D0]
mat: A 2-D tensor of dimension [D0, D1]
Returns:
A tensor of dimension [D0, D1], the result of vec * mat
"""
# Reshape vec to [D0, 1]
vec = array_ops.expand_dims(vec, -1)
return vec * mat
@ops.RegisterGradient("SoftmaxCrossEntropyWithLogits")
def _SoftmaxCrossEntropyWithLogitsGrad(op, grad_loss, grad_grad):
"""Gradient function for SoftmaxCrossEntropyWithLogits."""
# grad_loss is the backprop for cost, and we multiply it with the gradients
# (which is output[1])
# grad_grad is the backprop for softmax gradient.
#
# Second derivative is just softmax derivative w.r.t. logits.
softmax_grad = op.outputs[1]
grad = _BroadcastMul(grad_loss, softmax_grad)
def IsZero(g):
# Some introspection to check if the gradient is feeding zeros
if context.executing_eagerly():
# TODO(apassos) add an efficient way to detect eager zeros here.
return False
if g.op.type in ("ZerosLike", "Zeros"):
return True
const_fill_value = tensor_util.constant_value(g)
return const_fill_value is not None and (const_fill_value == 0).all()
logits = op.inputs[0]
if grad_grad is not None and not IsZero(grad_grad):
softmax = nn_ops.softmax(logits)
grad += ((grad_grad - array_ops.squeeze(
math_ops.matmul(
array_ops.expand_dims(grad_grad, 1),
array_ops.expand_dims(softmax, 2)),
axis=1)) * softmax)
return grad, _BroadcastMul(grad_loss, -nn_ops.log_softmax(logits))
@ops.RegisterGradient("SparseSoftmaxCrossEntropyWithLogits")
def _SparseSoftmaxCrossEntropyWithLogitsGrad(op, grad_0, _):
"""Gradient function for SparseSoftmaxCrossEntropyWithLogits."""
# grad_0 is the backprop for cost, and we multiply it with the gradients
# (which is output[1])
# There is no gradient for the labels
#
# Currently there is no way to take the second derivative of this op
# due to the fused implementation's interaction with tf.gradients(),
# so we make sure we prevent silently incorrect results by raising
# an error if the second derivative is requested via prevent_gradient.
sparse_softmax_grad_without_gradient = array_ops.prevent_gradient(
op.outputs[1],
message="Currently there is no way to take the second "
"derivative of sparse_softmax_cross_entropy_with_logits due to the fused "
"implementation's interaction with tf.gradients()")
return _BroadcastMul(grad_0, sparse_softmax_grad_without_gradient), None
@ops.RegisterGradient("Conv2D")
def _Conv2DGrad(op, grad):
"""Gradient function for Conv2D."""
dilations = op.get_attr("dilations")
strides = op.get_attr("strides")
padding = op.get_attr("padding")
explicit_paddings = op.get_attr("explicit_paddings")
use_cudnn_on_gpu = op.get_attr("use_cudnn_on_gpu")
data_format = op.get_attr("data_format")
shape_0, shape_1 = array_ops.shape_n([op.inputs[0], op.inputs[1]])
# We call the gen_nn_ops backprop functions instead of nn_ops backprop
# functions for performance reasons in Eager mode. gen_nn_ops functions take a
# `explicit_paddings` parameter, but nn_ops functions do not. So if were were
# to use the nn_ops functions, we would have to convert `padding` and
# `explicit_paddings` into a single `padding` parameter, increasing overhead
# in Eager mode.
return [
gen_nn_ops.conv2d_backprop_input(
shape_0,
op.inputs[1],
grad,
dilations=dilations,
strides=strides,
padding=padding,
explicit_paddings=explicit_paddings,
use_cudnn_on_gpu=use_cudnn_on_gpu,
data_format=data_format),
gen_nn_ops.conv2d_backprop_filter(
op.inputs[0],
shape_1,
grad,
dilations=dilations,
strides=strides,
padding=padding,
explicit_paddings=explicit_paddings,
use_cudnn_on_gpu=use_cudnn_on_gpu,
data_format=data_format)
]
@ops.RegisterGradient("DepthwiseConv2dNative")
def _DepthwiseConv2dNativeGrad(op, grad):
return [
nn_ops.depthwise_conv2d_native_backprop_input(
array_ops.shape(op.inputs[0]),
op.inputs[1],
grad,
op.get_attr("strides"),
op.get_attr("padding"),
data_format=op.get_attr("data_format")),
nn_ops.depthwise_conv2d_native_backprop_filter(
op.inputs[0],
array_ops.shape(op.inputs[1]),
grad,
op.get_attr("strides"),
op.get_attr("padding"),
data_format=op.get_attr("data_format"))
]
@ops.RegisterGradient("Dilation2D")
def _Dilation2DGrad(op, grad):
return [
nn_ops.dilation2d_backprop_input(op.inputs[0], op.inputs[1], grad,
op.get_attr("strides"),
op.get_attr("rates"),
op.get_attr("padding")),
nn_ops.dilation2d_backprop_filter(op.inputs[0], op.inputs[1], grad,
op.get_attr("strides"),
op.get_attr("rates"),
op.get_attr("padding"))
]
@ops.RegisterGradient("LRN")
def _LRNGrad(op, grad):
depth_radius = op.get_attr("depth_radius")
bias = op.get_attr("bias")
alpha = op.get_attr("alpha")
beta = op.get_attr("beta")
return [
gen_nn_ops.lrn_grad(grad, op.inputs[0], op.outputs[0], depth_radius, bias,
alpha, beta)
]
@ops.RegisterGradient("AvgPool")
def _AvgPoolGrad(op, grad):
return gen_nn_ops.avg_pool_grad(
array_ops.shape(op.inputs[0]),
grad,
op.get_attr("ksize"),
op.get_attr("strides"),
op.get_attr("padding"),
data_format=op.get_attr("data_format"))
@ops.RegisterGradient("AvgPoolGrad")
def _AvgPoolGradGrad(op, grad):
return (array_ops.stop_gradient(op.inputs[0]),
gen_nn_ops.avg_pool(
grad,
op.get_attr("ksize"),
op.get_attr("strides"),
op.get_attr("padding"),
data_format=op.get_attr("data_format")))
@ops.RegisterGradient("MaxPool")
def _MaxPoolGrad(op, grad):
return gen_nn_ops.max_pool_grad(
op.inputs[0],
op.outputs[0],
grad,
op.get_attr("ksize"),
op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format"))
@ops.RegisterGradient("MaxPoolV2")
def _MaxPoolGradV2(op, grad):
ksize = op.inputs[1]
strides = op.inputs[2]
return gen_nn_ops.max_pool_grad_v2(
op.inputs[0],
op.outputs[0],
grad,
ksize,
strides,
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format")), None, None
@ops.RegisterGradient("MaxPoolWithArgmax")
def _MaxPoolGradWithArgmax(op, grad, unused_argmax_grad):
del unused_argmax_grad
return gen_nn_ops.max_pool_grad_with_argmax(
op.inputs[0],
grad,
op.outputs[1],
op.get_attr("ksize"),
op.get_attr("strides"),
padding=op.get_attr("padding"),
include_batch_in_index=op.get_attr("include_batch_in_index"))
@ops.RegisterGradient("MaxPoolGrad")
def _MaxPoolGradGrad(op, grad):
return (array_ops.zeros(
shape=array_ops.shape(op.inputs[0]), dtype=op.inputs[0].dtype),
array_ops.zeros(
shape=array_ops.shape(op.inputs[1]), dtype=op.inputs[1].dtype),
gen_nn_ops.max_pool_grad_grad(
op.inputs[0],
op.inputs[1],
grad,
op.get_attr("ksize"),
op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format")))
@ops.RegisterGradient("MaxPoolGradV2")
def _MaxPoolGradGradV2(op, grad):
ksize = op.inputs[3]
strides = op.inputs[4]
return (array_ops.zeros(
shape=array_ops.shape(op.inputs[0]), dtype=op.inputs[0].dtype),
array_ops.zeros(
shape=array_ops.shape(op.inputs[1]), dtype=op.inputs[1].dtype),
gen_nn_ops.max_pool_grad_grad_v2(
op.inputs[0],
op.inputs[1],
grad,
ksize,
strides,
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format")), None, None)
@ops.RegisterGradient("MaxPoolGradGrad")
def _MaxPoolGradGradGrad(op, grad):
return (array_ops.zeros(
shape=array_ops.shape(op.inputs[0]), dtype=op.inputs[0].dtype),
array_ops.zeros(
shape=array_ops.shape(op.inputs[1]), dtype=op.inputs[1].dtype),
gen_nn_ops.max_pool_grad(
op.inputs[0],
op.inputs[1],
grad,
op.get_attr("ksize"),
op.get_attr("strides"),
padding=op.get_attr("padding"),
data_format=op.get_attr("data_format")))
@ops.RegisterGradient("FractionalMaxPool")
def _FractionalMaxPoolGrad(op, grad_0, unused_grad_1, unused_grad_2):
"""Returns gradient for FractionalMaxPool.
Since FractionalMaxPool has three outputs, there are three gradients passed in
for each of the outputs. Only the first one is useful, the other two gradients
are empty.
Args:
op: The FractionalMaxPoolOp.
grad_0: Gradient with respect to op.outputs[0]
unused_grad_1: Gradient with respect to op.outputs[1]/row_seq. It is empty.
unused_grad_2: Gradient with respect to op.outputs[2]/col_seq. It is empty.
Returns:
Input backprop for FractionalMaxPool op.
"""
return gen_nn_ops.fractional_max_pool_grad(
op.inputs[0], op.outputs[0], grad_0, op.outputs[1], op.outputs[2],
op.get_attr("overlapping"))
@ops.RegisterGradient("FractionalAvgPool")
def _FractionalAvgPoolGrad(op, grad_0, unused_grad_1, unused_grad_2):
"""Returns gradient for FractionalAvgPool.
Since FractionalAvgPool has three outputs, there are three gradients passed in
for each of the outputs. Only the first one is useful, the other two gradients
are empty.
Args:
op: The FractionalAvgPoolOp.
grad_0: Gradient with respect to op.outputs[0]
unused_grad_1: Gradient with respect to op.outputs[1]/row_seq. It is empty.
unused_grad_2: Gradient with respect to op.outputs[2]/col_seq. It is empty.
Returns:
Input backprop for FractionalAvgPool op.
"""
return gen_nn_ops.fractional_avg_pool_grad(op.inputs[0].get_shape(), grad_0,
op.outputs[1], op.outputs[2],
op.get_attr("overlapping"))
@ops.RegisterGradient("BatchNormWithGlobalNormalization")
def _BatchNormWithGlobalNormalizationGrad(op, grad):
"""Return the gradients for the 5 inputs of BatchNormWithGlobalNormalization.
We do not backprop anything for the mean and var intentionally as they are
not being trained with backprop in the operation.
Args:
op: The BatchNormOp for which we need to generate gradients.
grad: Tensor. The gradients passed to the BatchNormOp.
Returns:
dx: Backprop for input, which is (grad * (g * rsqrt(v + epsilon)))
dm: Backprop for mean, which is
sum_over_rest(grad * g) * (-1 / rsqrt(v + epsilon))
dv: Backprop for variance, which is
sum_over_rest(grad * g * (x - m)) * (-1/2) * (v + epsilon) ^ (-3/2)
db: Backprop for beta, which is grad reduced in all except the
last dimension.
dg: Backprop for gamma, which is (grad * ((x - m) * rsqrt(v + epsilon)))
"""
dx, dm, dv, db, dg = gen_nn_ops.batch_norm_with_global_normalization_grad(
op.inputs[0], op.inputs[1], op.inputs[2], op.inputs[4], grad,
op.get_attr("variance_epsilon"), op.get_attr("scale_after_normalization"))
return dx, dm, dv, db, dg
def _BaseFusedBatchNormGrad(op, version, *grad):
"""Return the gradients for the 3 inputs of BatchNorm.
Args:
op: The BatchNormOp for which we need to compute gradients.
version: Integer indicating which version to use of the fused batch
norm gradient.
*grad: An argument list for tensors of gradients wrt the outputs
with grad[0] as grad_y.
Returns:
grad_x: gradient for x, which is scale * rsqrt(variance + epsilon) *
[grad_y - mean(grad_y) - (x - mean(x)) *
mean(grad_y * (x - mean(x))) / (variance + epsilon)]
in training mode; grad_y * scale * rsqrt(pop_variance + epsilon)
in freeze mode.
grad_scale: gradient for scale, which is sum(grad_y * (x - mean(x)) *
rsqrt(variance + epsilon)) in training mode;
sum(grad_y * (x - pop_mean) * rsqrt(pop_variance + epsilon))
in freeze mode.
grad_offset: gradient for offset, which is sum(grad_y) in training mode;
sum(grad_y) in freeze mode.
"""
x = op.inputs[0]
grad_y = grad[0]
scale = op.inputs[1]
epsilon = op.get_attr("epsilon")
data_format = op.get_attr("data_format")
is_training = op.get_attr("is_training")
if version == 2:
grad_fun = gen_nn_ops.fused_batch_norm_grad_v3
elif version == 1:
grad_fun = gen_nn_ops.fused_batch_norm_grad_v2
else:
grad_fun = gen_nn_ops.fused_batch_norm_grad
if is_training:
args = {
"y_backprop": grad_y,
"x": x,
"scale": scale,
"reserve_space_1": op.outputs[3],
"reserve_space_2": op.outputs[4],
"epsilon": epsilon,
"data_format": data_format,
"is_training": is_training
}
if version == 2:
args["reserve_space_3"] = op.outputs[5]
return grad_fun(**args)
else:
pop_mean = op.inputs[3]
pop_var = op.inputs[4]
if data_format == b"NCHW":
x = array_ops.transpose(x, [0, 2, 3, 1])
grad_y = array_ops.transpose(grad_y, [0, 2, 3, 1])
args = {
"y_backprop": grad_y,
"x": x,
"scale": scale,
"reserve_space_1": pop_mean,
"reserve_space_2": pop_var,
"epsilon": epsilon,
"data_format": "NHWC",
"is_training": is_training
}
if version == 2:
args["reserve_space_3"] = op.outputs[5]
dx, dscale, doffset, _, _ = grad_fun(**args)
if data_format == b"NCHW":
dx = array_ops.transpose(dx, [0, 3, 1, 2])
return dx, dscale, doffset, None, None
@ops.RegisterGradient("FusedBatchNorm")
def _FusedBatchNormGrad(op, *grad):
return _BaseFusedBatchNormGrad(op, 0, *grad)
@ops.RegisterGradient("FusedBatchNormV2")
def _FusedBatchNormV2Grad(op, *grad):
return _BaseFusedBatchNormGrad(op, 1, *grad)
@ops.RegisterGradient("FusedBatchNormV3")
def _FusedBatchNormV3Grad(op, *grad):
return _BaseFusedBatchNormGrad(op, 2, *grad)
def _BatchNormGrad(grad_y,
x,
scale,
pop_mean,
pop_var,
epsilon,
data_format,
is_training=True):
"""Returns the gradients for the 3 inputs of BatchNorm.
Args:
grad_y: A `Tensor` of 4 dimensions for gradient for y.
x: A `Tensor` of 4 dimensions for x.
scale: A `Tensor` of 1 dimension for scaling.
pop_mean: A `Tensor` of 1 dimension for the population mean. Only used when
is_training=False.
pop_var: A `Tensor` of 1 dimension for the population variance. Only used
when is_training=False.
epsilon: A small float number added to the variance of x.
data_format: The data format for input. Either b"NHWC" or b"NCHW".
is_training: A bool value to indicate the operation is for training
(default) or inference.
Returns:
A tuple (grad_x, grad_scale, grad_offset), where grad_x is the gradient
for x, grad_scale the gradient for scale, and grad_offset the gradient
for offset.
"""
x_dtype = x.dtype.base_dtype
if x_dtype == dtypes.float16:
# float16 math is too imprecise, so we do the batch norm gradient
# computations in float32.
x = math_ops.cast(x, dtypes.float32)
grad_y = math_ops.cast(grad_y, dtypes.float32)
if is_training:
if data_format == b"NHWC":
keepdims = False
reduce_axis = [0, 1, 2]
else:
keepdims = True
reduce_axis = [0, 2, 3]
shape = [1, array_ops.size(scale), 1, 1]
scale = array_ops.reshape(scale, shape)
mean_grad_y = math_ops.reduce_mean(grad_y, reduce_axis, keepdims=keepdims)
mean_x = math_ops.reduce_mean(x, reduce_axis, keepdims=keepdims)
var_x = math_ops.reduce_mean(
math_ops.squared_difference(x, array_ops.stop_gradient(mean_x)),
reduce_axis,
keepdims=keepdims)
grad_y_offset = grad_y - mean_grad_y
x_offset = x - mean_x
mean = math_ops.reduce_mean(
grad_y * x_offset, axis=reduce_axis, keepdims=keepdims)
grad_x = scale * math_ops.rsqrt(var_x + epsilon) * (
grad_y_offset - math_ops.reciprocal(var_x + epsilon) * mean * x_offset)
grad_scale = math_ops.rsqrt(var_x + epsilon) * math_ops.reduce_sum(
grad_y * x_offset, axis=reduce_axis, keepdims=keepdims)
if data_format == b"NCHW":
grad_scale = array_ops.squeeze(grad_scale)
grad_offset = math_ops.reduce_sum(grad_y, axis=reduce_axis)
return math_ops.cast(grad_x, x_dtype), grad_scale, grad_offset
else:
if data_format == b"NHWC":
reduce_axis = [0, 1, 2]
else:
reduce_axis = [0, 2, 3]
shape = [1, array_ops.size(pop_mean), 1, 1]
pop_mean = array_ops.reshape(pop_mean, shape)
pop_var = array_ops.reshape(pop_var, shape)
scale = array_ops.reshape(scale, shape)
grad_offset = math_ops.reduce_sum(grad_y, axis=reduce_axis)
var_rsqrt = math_ops.rsqrt(pop_var + epsilon)
grad_scale = math_ops.reduce_sum(
grad_y * (x - pop_mean) * var_rsqrt, axis=reduce_axis)
grad_x = grad_y * scale * var_rsqrt
return math_ops.cast(grad_x, x_dtype), grad_scale, grad_offset
@ops.RegisterGradient("FusedBatchNormGrad")
def _FusedBatchNormGradGrad(op, *grad):
"""Returns the gradients for the 3 inputs of FusedBatchNormGrad.
Args:
op: The FusedBatchNormGradOp for which we need to compute gradients.
*grad: An argument list for tensors of gradients wrt the outputs with
grad[0] as grad_grad_x, grad[1] as grad_grad_scale, grad[2] as
grad_grad_offset.
Returns:
A tuple (grad_grad_y, grad_x, grad_scale, None, None), where grad_grad_y
is the gradient for grad_y, grad_x the gradient for x, grad_scale the
gradient for scale.
"""
data_format = op.get_attr("data_format")
epsilon = op.get_attr("epsilon")
is_training = op.get_attr("is_training")
grad_y = op.inputs[0]
x = op.inputs[1]
scale = op.inputs[2]
pop_mean = op.inputs[3]
pop_var = op.inputs[4]
grad_grad_x = grad[0]
grad_grad_scale = grad[1]
grad_grad_offset = grad[2]
with backprop.GradientTape() as tape:
tape.watch(grad_y)
tape.watch(x)
tape.watch(scale)
grad_x, grad_scale, grad_offset = _BatchNormGrad(
grad_y, x, scale, pop_mean, pop_var, epsilon, data_format, is_training)
grad_initial = [grad_grad_x, grad_grad_scale, grad_grad_offset]
grad_grad_y, grad_x, grad_scale = tape.gradient(
[grad_x, grad_scale, grad_offset], [grad_y, x, scale], grad_initial)
return grad_grad_y, grad_x, grad_scale, None, None
@ops.RegisterGradient("FusedBatchNormGradV2")
def _FusedBatchNormGradGradV2(op, *grad):
return _FusedBatchNormGradGrad(op, *grad)
@ops.RegisterGradient("FusedBatchNormGradV3")
def _FusedBatchNormGradGradV3(op, *grad):
grad_grad_y, grad_x, grad_scale, _, _ = _FusedBatchNormGradGrad(op, *grad)
return grad_grad_y, grad_x, grad_scale, None, None, None
@ops.RegisterGradient("L2Loss")
def _L2LossGrad(op, grad):
"""Return the gradients for L2Loss.
Args:
op: The L2LossOp for which we need to generate gradients.
grad: Tensor containing a single number.
Returns:
The gradient, which is (x * grad).
"""
return op.inputs[0] * grad
@ops.RegisterGradient("TopK")
@ops.RegisterGradient("TopKV2")
def _TopKGrad(op, grad, _):
"""Return the gradients for TopK.
Args:
op: The TopKOp for which we need to generate gradients.
grad: Tensor. The gradients passed to the TopKOp.
Returns:
A list of two tensors, the first being the gradient w.r.t to the input and
TopK, and the second being the gradient w.r.t. to the indices (all zero).
"""
in_shape = array_ops.shape(op.inputs[0])
ind_shape = array_ops.shape(op.outputs[1])
# int32 is not supported on GPU hence up-casting
ind_lastdim = array_ops.gather(
math_ops.cast(ind_shape, dtypes.int64),
array_ops.size(ind_shape) - 1)
# Flatten indices to 2D.
ind_2d = array_ops.reshape(op.outputs[1], array_ops.stack([-1, ind_lastdim]))
in_lastdim = array_ops.gather(
math_ops.cast(in_shape, dtypes.int64),
array_ops.size(in_shape) - 1)
outerdim = array_ops.shape(ind_2d)[0]
# Compute linear indices (flattened to 1D).
ind = array_ops.reshape(
ind_2d + math_ops.cast(
array_ops.expand_dims(
math_ops.range(0,
math_ops.cast(outerdim, dtypes.int64) * in_lastdim,
in_lastdim), -1), dtypes.int32), [-1])
# Substitute grad to appropriate locations and fill the rest with zeros,
# finally reshaping it to the original input shape.
return [
array_ops.reshape(
array_ops.scatter_nd(
array_ops.expand_dims(ind, -1), array_ops.reshape(grad, [-1]),
[math_ops.reduce_prod(in_shape)]), in_shape),
array_ops.zeros([], dtype=dtypes.int32)
]
@ops.RegisterGradient("NthElement")
def _NthElementGrad(op, grad):
"""Return the gradients for NthElement.
Args:
op: The NthElementOp for which we need to generate gradients.
grad: Tensor. The gradients passed to the NthElementOp
Returns:
A list of two tensors, the first being the gradient w.r.t. the input,
the second being the gradient w.r.t. the N (None).
"""
input = op.inputs[0] # pylint: disable=redefined-builtin
output = op.outputs[0]
# Compute the number of elements which equal to output in each reduction
# dimension. If there are multiple elements then the gradient will be
# divided between them.
indicators = math_ops.cast(
math_ops.equal(array_ops.expand_dims(output, -1), input), grad.dtype)
grad = array_ops.expand_dims(grad, -1)
num_selected = array_ops.expand_dims(math_ops.reduce_sum(indicators, -1), -1)
return [math_ops.div(indicators, num_selected) * grad, None]
|
tensorflow-master
|
tensorflow/python/ops/nn_grad.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for initializers in init_ops_v2."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import tensor_shape
from tensorflow.python.framework import test_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import init_ops_v2
from tensorflow.python.ops import random_ops
from tensorflow.python.ops import variables
from tensorflow.python.platform import test
class InitializersTest(test.TestCase):
def _identical_test(self,
init1,
init2,
assertion,
shape=None,
dtype=dtypes.float32):
if shape is None:
shape = [100]
t1 = self.evaluate(init1(shape, dtype))
t2 = self.evaluate(init2(shape, dtype))
self.assertEqual(tensor_shape.as_shape(shape), t1.shape)
self.assertEqual(tensor_shape.as_shape(shape), t2.shape)
self.assertEqual(assertion, np.allclose(t1, t2, rtol=1e-15, atol=1e-15))
def _duplicated_test(self,
init,
shape=None,
dtype=dtypes.float32):
if shape is None:
shape = [100]
t1 = self.evaluate(init(shape, dtype))
t2 = self.evaluate(init(shape, dtype))
self.assertEqual(tensor_shape.as_shape(shape), t1.shape)
self.assertEqual(tensor_shape.as_shape(shape), t2.shape)
self.assertFalse(np.allclose(t1, t2, rtol=1e-15, atol=1e-15))
def _range_test(self,
init,
shape,
target_mean=None,
target_std=None,
target_max=None,
target_min=None):
output = self.evaluate(init(shape))
self.assertEqual(output.shape, shape)
lim = 3e-2
if target_std is not None:
self.assertGreater(lim, abs(output.std() - target_std))
if target_mean is not None:
self.assertGreater(lim, abs(output.mean() - target_mean))
if target_max is not None:
self.assertGreater(lim, abs(output.max() - target_max))
if target_min is not None:
self.assertGreater(lim, abs(output.min() - target_min))
class ConstantInitializersTest(InitializersTest):
@test_util.run_in_graph_and_eager_modes
def testZeros(self):
self._range_test(init_ops_v2.Zeros(), shape=(4, 5),
target_mean=0., target_max=0.)
@test_util.run_in_graph_and_eager_modes
def testOnes(self):
self._range_test(init_ops_v2.Ones(), shape=(4, 5),
target_mean=1., target_max=1.)
@test_util.run_in_graph_and_eager_modes
def testConstantInt(self):
self._range_test(
init_ops_v2.Constant(2),
shape=(5, 6, 4),
target_mean=2,
target_max=2,
target_min=2)
@test_util.run_in_graph_and_eager_modes
def testConstantTuple(self):
init = init_ops_v2.constant_initializer((10, 20, 30))
tensor = init(shape=[3])
self.assertAllEqual(self.evaluate(tensor), [10, 20, 30])
self.assertEqual(tensor.shape, [3])
@test_util.run_in_graph_and_eager_modes
def testConstantInvalidValue(self):
c = constant_op.constant([1.0, 2.0, 3.0])
with self.assertRaisesRegexp(
TypeError, r"Invalid type for initial value: .*Tensor.*"):
init_ops_v2.constant_initializer(c)
v = variables.Variable([3.0, 2.0, 1.0])
with self.assertRaisesRegexp(
TypeError, r"Invalid type for initial value: .*Variable.*"):
init_ops_v2.constant_initializer(v)
def _testNDimConstantInitializer(self, value, shape, expected):
with test_util.use_gpu():
init = init_ops_v2.constant_initializer(value)
x = init(shape)
actual = self.evaluate(array_ops.reshape(x, [-1]))
self.assertEqual(len(actual), len(expected))
for a, e in zip(actual, expected):
self.assertEqual(a, e)
@test_util.run_in_graph_and_eager_modes
def testNDimConstantInitializer(self):
value = [0, 1, 2, 3, 4, 5]
shape = [2, 3]
expected = list(value)
self._testNDimConstantInitializer(value, shape, expected)
self._testNDimConstantInitializer(np.asarray(value), shape, expected)
self._testNDimConstantInitializer(np.asarray(value).reshape(tuple(shape)),
shape, expected)
def _testNDimConstantInitializerIncorrectNumberValues(self, value, shape):
with test_util.use_gpu():
init = init_ops_v2.constant_initializer(value)
self.assertRaises(TypeError,
init,
shape=shape)
@test_util.run_in_graph_and_eager_modes
def testNDimConstantInitializerIncorrectNumberValues(self):
value = [0, 1, 2, 3, 4, 5]
for shape in [[2, 4], [2, 2]]:
self._testNDimConstantInitializerIncorrectNumberValues(value, shape)
self._testNDimConstantInitializerIncorrectNumberValues(np.asarray(value),
shape)
self._testNDimConstantInitializerIncorrectNumberValues(
np.asarray(value).reshape(tuple([2, 3])), shape)
class RandomUniformInitializerTest(InitializersTest):
@test_util.run_in_graph_and_eager_modes
def testRangeInitializer(self):
shape = (9, 6, 7)
self._range_test(
init_ops_v2.RandomUniform(minval=-1, maxval=1, seed=124),
shape,
target_mean=0.,
target_max=1,
target_min=-1)
@test_util.run_in_graph_and_eager_modes
def testInitializerIdentical(self):
self.skipTest("Doesn't work without the graphs")
init1 = init_ops_v2.RandomUniform(0, 7, seed=1)
init2 = init_ops_v2.RandomUniform(0, 7, seed=1)
self._identical_test(init1, init2, True)
@test_util.run_in_graph_and_eager_modes
def testInitializerDifferent(self):
init1 = init_ops_v2.RandomUniform(0, 7, seed=1)
init2 = init_ops_v2.RandomUniform(0, 7, seed=2)
self._identical_test(init1, init2, False)
@test_util.run_in_graph_and_eager_modes
def testDuplicatedInitializer(self):
init = init_ops_v2.RandomUniform(0.0, 1.0)
self._duplicated_test(init)
class RandomNormalInitializerTest(InitializersTest):
@test_util.run_in_graph_and_eager_modes
def testRangeInitializer(self):
self._range_test(
init_ops_v2.RandomNormal(mean=0, stddev=1, seed=153),
shape=(8, 12, 99),
target_mean=0.,
target_std=1)
@test_util.run_in_graph_and_eager_modes
def testInitializerIdentical(self):
self.skipTest("Doesn't work without the graphs")
init1 = init_ops_v2.RandomNormal(0, 7, seed=1)
init2 = init_ops_v2.RandomNormal(0, 7, seed=1)
self._identical_test(init1, init2, True)
@test_util.run_in_graph_and_eager_modes
def testInitializerDifferent(self):
init1 = init_ops_v2.RandomNormal(0, 7, seed=1)
init2 = init_ops_v2.RandomNormal(0, 7, seed=2)
self._identical_test(init1, init2, False)
@test_util.run_in_graph_and_eager_modes
def testDuplicatedInitializer(self):
init = init_ops_v2.RandomNormal(0.0, 1.0)
self._duplicated_test(init)
class TruncatedNormalInitializerTest(InitializersTest):
@test_util.run_in_graph_and_eager_modes
def testRangeInitializer(self):
self._range_test(
init_ops_v2.TruncatedNormal(mean=0, stddev=1, seed=126),
shape=(12, 99, 7),
target_mean=0.,
target_max=2,
target_min=-2)
@test_util.run_in_graph_and_eager_modes
def testInitializerIdentical(self):
self.skipTest("Not seeming to work in Eager mode")
init1 = init_ops_v2.TruncatedNormal(0.0, 1.0, seed=1)
init2 = init_ops_v2.TruncatedNormal(0.0, 1.0, seed=1)
self._identical_test(init1, init2, True)
@test_util.run_in_graph_and_eager_modes
def testInitializerDifferent(self):
init1 = init_ops_v2.TruncatedNormal(0.0, 1.0, seed=1)
init2 = init_ops_v2.TruncatedNormal(0.0, 1.0, seed=2)
self._identical_test(init1, init2, False)
@test_util.run_in_graph_and_eager_modes
def testDuplicatedInitializer(self):
init = init_ops_v2.TruncatedNormal(0.0, 1.0)
self._duplicated_test(init)
def testInvalidDataType(self):
init = init_ops_v2.TruncatedNormal(0.0, 1.0)
with self.assertRaises(ValueError):
init([1], dtype=dtypes.int32)
class VarianceScalingInitializerTest(InitializersTest):
@test_util.run_in_graph_and_eager_modes
def testTruncatedNormalDistribution(self):
shape = [100, 100]
expect_mean = 0.
expect_var = 1. / shape[0]
init = init_ops_v2.VarianceScaling(distribution="truncated_normal")
with test_util.use_gpu(), test.mock.patch.object(
random_ops, "truncated_normal",
wraps=random_ops.truncated_normal) as mock_truncated_normal:
x = self.evaluate(init(shape))
self.assertTrue(mock_truncated_normal.called)
self.assertNear(np.mean(x), expect_mean, err=1e-2)
self.assertNear(np.var(x), expect_var, err=1e-2)
@test_util.run_in_graph_and_eager_modes
def testNormalDistribution(self):
shape = [100, 100]
expect_mean = 0.
expect_var = 1. / shape[0]
init = init_ops_v2.VarianceScaling(distribution="truncated_normal")
with test_util.use_gpu(), test.mock.patch.object(
random_ops, "truncated_normal",
wraps=random_ops.truncated_normal) as mock_truncated_normal:
x = self.evaluate(init(shape))
self.assertTrue(mock_truncated_normal.called)
self.assertNear(np.mean(x), expect_mean, err=1e-2)
self.assertNear(np.var(x), expect_var, err=1e-2)
@test_util.run_in_graph_and_eager_modes
def testUntruncatedNormalDistribution(self):
shape = [100, 100]
expect_mean = 0.
expect_var = 1. / shape[0]
init = init_ops_v2.VarianceScaling(
distribution="untruncated_normal")
with test_util.use_gpu(), test.mock.patch.object(
random_ops, "random_normal",
wraps=random_ops.random_normal) as mock_random_normal:
x = self.evaluate(init(shape))
self.assertTrue(mock_random_normal.called)
self.assertNear(np.mean(x), expect_mean, err=1e-2)
self.assertNear(np.var(x), expect_var, err=1e-2)
@test_util.run_in_graph_and_eager_modes
def testUniformDistribution(self):
shape = [100, 100]
expect_mean = 0.
expect_var = 1. / shape[0]
init = init_ops_v2.VarianceScaling(distribution="uniform")
with test_util.use_gpu():
x = self.evaluate(init(shape))
self.assertNear(np.mean(x), expect_mean, err=1e-2)
self.assertNear(np.var(x), expect_var, err=1e-2)
class OrthogonalInitializerTest(InitializersTest):
@test_util.run_in_graph_and_eager_modes
def testRangeInitializer(self):
self._range_test(init_ops_v2.Orthogonal(seed=123), shape=(20, 20),
target_mean=0.)
@test_util.run_in_graph_and_eager_modes
def testInitializerIdentical(self):
self.skipTest("Doesn't work without the graphs")
init1 = init_ops_v2.Orthogonal(seed=1)
init2 = init_ops_v2.Orthogonal(seed=1)
self._identical_test(init1, init2, True, (10, 10))
@test_util.run_in_graph_and_eager_modes
def testInitializerDifferent(self):
init1 = init_ops_v2.Orthogonal(seed=1)
init2 = init_ops_v2.Orthogonal(seed=2)
self._identical_test(init1, init2, False, (10, 10))
@test_util.run_in_graph_and_eager_modes
def testDuplicatedInitializer(self):
init = init_ops_v2.Orthogonal()
self._duplicated_test(init, (10, 10))
@test_util.run_in_graph_and_eager_modes
def testInvalidDataType(self):
init = init_ops_v2.Orthogonal()
self.assertRaises(ValueError, init, shape=(10, 10), dtype=dtypes.string)
@test_util.run_in_graph_and_eager_modes
def testInvalidShape(self):
init = init_ops_v2.Orthogonal()
with test_util.use_gpu():
self.assertRaises(ValueError, init, shape=[5])
@test_util.run_in_graph_and_eager_modes
def testGain(self):
self.skipTest("Doesn't work without the graphs")
init1 = init_ops_v2.Orthogonal(seed=1)
init2 = init_ops_v2.Orthogonal(gain=3.14, seed=1)
with test_util.use_gpu():
t1 = self.evaluate(init1(shape=(10, 10)))
t2 = self.evaluate(init2(shape=(10, 10)))
self.assertAllClose(t1, t2 / 3.14)
@test_util.run_in_graph_and_eager_modes
def testShapesValues(self):
for shape in [(10, 10), (10, 9, 8), (100, 5, 5), (50, 40), (40, 50)]:
init = init_ops_v2.Orthogonal()
tol = 1e-5
with test_util.use_gpu():
# Check the shape
t = self.evaluate(init(shape))
self.assertAllEqual(shape, t.shape)
# Check orthogonality by computing the inner product
t = t.reshape((np.prod(t.shape[:-1]), t.shape[-1]))
if t.shape[0] > t.shape[1]:
self.assertAllClose(
np.dot(t.T, t), np.eye(t.shape[1]), rtol=tol, atol=tol)
else:
self.assertAllClose(
np.dot(t, t.T), np.eye(t.shape[0]), rtol=tol, atol=tol)
class IdentityInitializerTest(InitializersTest):
@test_util.run_in_graph_and_eager_modes
def testRange(self):
with self.assertRaises(ValueError):
shape = (3, 4, 5)
self._range_test(
init_ops_v2.Identity(),
shape=shape,
target_mean=1. / shape[0],
target_max=1.)
shape = (3, 3)
self._range_test(
init_ops_v2.Identity(),
shape=shape,
target_mean=1. / shape[0],
target_max=1.)
@test_util.run_in_graph_and_eager_modes
def testInvalidDataType(self):
init = init_ops_v2.Identity()
self.assertRaises(ValueError, init, shape=[10, 5], dtype=dtypes.int32)
@test_util.run_in_graph_and_eager_modes
def testInvalidShape(self):
init = init_ops_v2.Identity()
with test_util.use_gpu():
self.assertRaises(ValueError, init, shape=[5, 7, 7])
self.assertRaises(ValueError, init, shape=[5])
self.assertRaises(ValueError, init, shape=[])
@test_util.run_in_graph_and_eager_modes
def testNonSquare(self):
init = init_ops_v2.Identity()
shape = (10, 5)
with test_util.use_gpu():
self.assertAllClose(self.evaluate(init(shape)), np.eye(*shape))
@test_util.run_in_graph_and_eager_modes
def testGain(self):
shape = (10, 10)
for dtype in [dtypes.float32, dtypes.float64]:
init_default = init_ops_v2.Identity()
init_custom = init_ops_v2.Identity(gain=0.9)
with test_util.use_gpu():
self.assertAllClose(self.evaluate(init_default(shape, dtype=dtype)),
np.eye(*shape))
with test_util.use_gpu():
self.assertAllClose(self.evaluate(init_custom(shape, dtype=dtype)),
np.eye(*shape) * 0.9)
class GlorotInitializersTest(InitializersTest):
@test_util.run_in_graph_and_eager_modes
def testGlorotUniform(self):
shape = (5, 6, 4, 2)
fan_in, fan_out = init_ops_v2._compute_fans(shape)
std = np.sqrt(2. / (fan_in + fan_out))
self._range_test(
init_ops_v2.GlorotUniform(seed=123),
shape,
target_mean=0.,
target_std=std)
@test_util.run_in_graph_and_eager_modes
def test_GlorotNormal(self):
shape = (5, 6, 4, 2)
fan_in, fan_out = init_ops_v2._compute_fans(shape)
std = np.sqrt(2. / (fan_in + fan_out))
self._range_test(
init_ops_v2.GlorotNormal(seed=123),
shape,
target_mean=0.,
target_std=std)
class MethodInitializers(InitializersTest):
@test_util.run_in_graph_and_eager_modes
def testLecunUniform(self):
shape = (5, 6, 4, 2)
fan_in, _ = init_ops_v2._compute_fans(shape)
std = np.sqrt(1. / fan_in)
self._range_test(
init_ops_v2.lecun_uniform(seed=123),
shape,
target_mean=0.,
target_std=std)
@test_util.run_in_graph_and_eager_modes
def testHeUniform(self):
shape = (5, 6, 4, 2)
fan_in, _ = init_ops_v2._compute_fans(shape)
std = np.sqrt(2. / fan_in)
self._range_test(
init_ops_v2.he_uniform(seed=123),
shape,
target_mean=0.,
target_std=std)
@test_util.run_in_graph_and_eager_modes
def testLecunNormal(self):
shape = (5, 6, 4, 2)
fan_in, _ = init_ops_v2._compute_fans(shape)
std = np.sqrt(1. / fan_in)
self._range_test(
init_ops_v2.lecun_normal(seed=123),
shape,
target_mean=0.,
target_std=std)
@test_util.run_in_graph_and_eager_modes
def testHeNormal(self):
shape = (5, 6, 4, 2)
fan_in, _ = init_ops_v2._compute_fans(shape)
std = np.sqrt(2. / fan_in)
self._range_test(
init_ops_v2.he_normal(seed=123),
shape,
target_mean=0.,
target_std=std)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/ops/init_ops_v2_test.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Gradients for operators defined in tensor_array_ops.py."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import ops
from tensorflow.python.ops import tensor_array_ops
# TODO(b/31222613): These ops may be differentiable, and there may be
# latent bugs here.
ops.NotDifferentiable("TensorArray")
ops.NotDifferentiable("TensorArrayGrad")
ops.NotDifferentiable("TensorArraySize")
ops.NotDifferentiable("TensorArrayClose")
ops.NotDifferentiable("TensorArrayV2")
ops.NotDifferentiable("TensorArrayGradV2")
ops.NotDifferentiable("TensorArraySizeV2")
ops.NotDifferentiable("TensorArrayCloseV2")
ops.NotDifferentiable("TensorArrayV3")
ops.NotDifferentiable("TensorArrayGradV3")
ops.NotDifferentiable("TensorArrayGradWithShape")
ops.NotDifferentiable("TensorArraySizeV3")
ops.NotDifferentiable("TensorArrayCloseV3")
def _GetGradSource(op_or_tensor):
"""Identify which call to tf.gradients created this gradient op or tensor.
TensorArray gradient calls use an accumulator TensorArray object. If
multiple gradients are calculated and run in the same session, the multiple
gradient nodes may accidentally flow throuth the same accumulator TensorArray.
This double counting breaks the TensorArray gradient flow.
The solution is to identify which gradient call this particular
TensorArray*Grad is being called in, by looking at the input gradient
tensor's name, and create or lookup an accumulator gradient TensorArray
associated with this specific call. This solves any confusion and ensures
different gradients from the same forward graph get their own accumulators.
This function creates the unique label associated with the tf.gradients call
that is used to create the gradient TensorArray.
Args:
op_or_tensor: `Tensor` or `Operation` which is an input to a
TensorArray*Grad call.
Returns:
A python string, the unique label associated with this particular
gradients calculation.
Raises:
ValueError: If not called within a gradients calculation.
"""
name_tokens = op_or_tensor.name.split("/")
grad_pos = [i for i, x in enumerate(name_tokens) if x.startswith("gradients")]
if not grad_pos:
raise ValueError(
"Expected op/tensor name to start with gradients (excluding scope)"
", got: {}. This means that a tf.gradients op with this op in its "
"dependency path has a custom name that does not start with "
"'gradients'. Please make sure all calls to tf.gradients that have "
"non-empty 'name' arguments use names that start with "
"'gradients'.".format(op_or_tensor.name))
return "/".join(name_tokens[:grad_pos[-1] + 1])
@ops.RegisterGradient("TensorArrayRead")
@ops.RegisterGradient("TensorArrayReadV2")
@ops.RegisterGradient("TensorArrayReadV3")
def _TensorArrayReadGrad(op, grad):
"""Gradient for TensorArrayRead.
Args:
op: Forward TensorArrayRead op.
grad: Gradient `Tensor` to TensorArrayRead.
Returns:
A flow `Tensor`, which can be used in control dependencies to
force the write of `grad` to the gradient `TensorArray`.
"""
# Note: the forward flow dependency in the call to grad() is necessary for
# the case of dynamic sized TensorArrays. When creating the gradient
# TensorArray, the final size of the forward array must be known.
# For this we need to wait until it has been created by depending on
# the input flow of the original op.
handle = op.inputs[0]
index = op.inputs[1]
flow = op.inputs[2]
dtype = op.get_attr("dtype")
grad_source = _GetGradSource(grad)
g = (tensor_array_ops.TensorArray(dtype=dtype, handle=handle, flow=flow,
colocate_with_first_write_call=False)
.grad(source=grad_source, flow=flow))
w_g = g.write(index, grad)
return [None, None, w_g.flow]
@ops.RegisterGradient("TensorArrayWrite")
@ops.RegisterGradient("TensorArrayWriteV2")
@ops.RegisterGradient("TensorArrayWriteV3")
def _TensorArrayWriteGrad(op, flow):
"""Gradient for TensorArrayWrite.
Args:
op: Forward TensorArrayWrite op.
flow: Gradient `Tensor` flow to TensorArrayWrite.
Returns:
A grad `Tensor`, the gradient created in an upstream ReadGrad or PackGrad.
"""
# handle is the output store_handle of TensorArrayReadGrad or
# the handle output of TensorArrayWriteGrad. we must use this one.
handle = op.inputs[0]
index = op.inputs[1]
dtype = op.get_attr("T")
grad_source = _GetGradSource(flow)
g = (tensor_array_ops.TensorArray(dtype=dtype, handle=handle, flow=flow,
colocate_with_first_write_call=False)
.grad(source=grad_source, flow=flow))
grad = g.read(index)
return [None, None, grad, flow]
@ops.RegisterGradient("TensorArrayGather")
@ops.RegisterGradient("TensorArrayGatherV2")
@ops.RegisterGradient("TensorArrayGatherV3")
def _TensorArrayGatherGrad(op, grad):
"""Gradient for TensorArrayGather.
Args:
op: Forward TensorArrayGather op.
grad: Gradient `Tensor` to TensorArrayGather.
Returns:
A flow `Tensor`, which can be used in control dependencies to
force the write of `grad` to the gradient `TensorArray`.
"""
# Note: the forward flow dependency in the call to grad() is necessary for
# the case of dynamic sized TensorArrays. When creating the gradient
# TensorArray, the final size of the forward array must be known.
# For this we need to wait until it has been created by depending on
# the input flow of the original op.
handle = op.inputs[0]
indices = op.inputs[1]
flow = op.inputs[2]
dtype = op.get_attr("dtype")
grad_source = _GetGradSource(grad)
g = (tensor_array_ops.TensorArray(dtype=dtype, handle=handle, flow=flow,
colocate_with_first_write_call=False)
.grad(source=grad_source, flow=flow))
u_g = g.scatter(indices, grad)
return [None, None, u_g.flow]
@ops.RegisterGradient("TensorArrayScatter")
@ops.RegisterGradient("TensorArrayScatterV2")
@ops.RegisterGradient("TensorArrayScatterV3")
def _TensorArrayScatterGrad(op, flow):
"""Gradient for TensorArrayScatter.
Args:
op: Forward TensorArrayScatter op.
flow: Gradient `Tensor` flow to TensorArrayScatter.
Returns:
A grad `Tensor`, the gradient created in upstream ReadGrads or PackGrad.
"""
handle = op.inputs[0]
indices = op.inputs[1]
dtype = op.get_attr("T")
grad_source = _GetGradSource(flow)
g = (tensor_array_ops.TensorArray(dtype=dtype, handle=handle, flow=flow,
colocate_with_first_write_call=False)
.grad(source=grad_source, flow=flow))
grad = g.gather(indices)
return [None, None, grad, flow]
@ops.RegisterGradient("TensorArrayConcat")
@ops.RegisterGradient("TensorArrayConcatV2")
@ops.RegisterGradient("TensorArrayConcatV3")
def _TensorArrayConcatGrad(op, grad, unused_lengths_grad):
"""Gradient for TensorArrayConcat.
Args:
op: Forward TensorArrayConcat op.
grad: Gradient `Tensor` to TensorArrayConcat.
Returns:
A flow `Tensor`, which can be used in control dependencies to
force the write of `grad` to the gradient `TensorArray`.
"""
# Note: the forward flow dependency in the call to grad() is necessary for
# the case of dynamic sized TensorArrays. When creating the gradient
# TensorArray, the final size of the forward array must be known.
# For this we need to wait until it has been created by depending on
# the input flow of the original op.
handle = op.inputs[0]
flow = op.inputs[1]
lengths = op.outputs[1]
dtype = op.get_attr("dtype")
grad_source = _GetGradSource(grad)
g = (tensor_array_ops.TensorArray(dtype=dtype, handle=handle, flow=flow,
colocate_with_first_write_call=False)
.grad(source=grad_source, flow=flow))
u_g = g.split(grad, lengths=lengths)
# handle, flow_in
return [None, u_g.flow]
@ops.RegisterGradient("TensorArraySplit")
@ops.RegisterGradient("TensorArraySplitV2")
@ops.RegisterGradient("TensorArraySplitV3")
def _TensorArraySplitGrad(op, flow):
"""Gradient for TensorArraySplit.
Args:
op: Forward TensorArraySplit op.
flow: Gradient `Tensor` flow to TensorArraySplit.
Returns:
A grad `Tensor`, the gradient created in upstream ReadGrads or PackGrad.
"""
handle = op.inputs[0]
dtype = op.get_attr("T")
grad_source = _GetGradSource(flow)
g = (tensor_array_ops.TensorArray(dtype=dtype, handle=handle, flow=flow,
colocate_with_first_write_call=False)
.grad(source=grad_source, flow=flow))
grad = g.concat()
# handle, value, lengths, flow_in
return [None, grad, None, flow]
|
tensorflow-master
|
tensorflow/python/ops/tensor_array_grad.py
|
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Arithmetic Operations that don't fit into math_ops due to dependencies.
To avoid circular dependencies, some math_ops should go here.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import re
import string
from six.moves import xrange # pylint: disable=redefined-builtin
from tensorflow.compiler.tf2xla.ops import gen_xla_ops
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.platform import tf_logging as logging
from tensorflow.python.util import deprecation
from tensorflow.python.util.tf_export import tf_export
# TODO(b/27419586) Change docstring for required dtype of x once int allowed
@tf_export('math.lbeta', v1=['math.lbeta', 'lbeta'])
@deprecation.deprecated_endpoints('lbeta')
def lbeta(x, name=None):
r"""Computes \\(ln(|Beta(x)|)\\), reducing along the last dimension.
Given one-dimensional `z = [z_0,...,z_{K-1}]`, we define
$$Beta(z) = \prod_j Gamma(z_j) / Gamma(\sum_j z_j)$$
And for `n + 1` dimensional `x` with shape `[N1, ..., Nn, K]`, we define
$$lbeta(x)[i1, ..., in] = Log(|Beta(x[i1, ..., in, :])|)$$.
In other words, the last dimension is treated as the `z` vector.
Note that if `z = [u, v]`, then
\\(Beta(z) = int_0^1 t^{u-1} (1 - t)^{v-1} dt\\), which defines the
traditional bivariate beta function.
If the last dimension is empty, we follow the convention that the sum over
the empty set is zero, and the product is one.
Args:
x: A rank `n + 1` `Tensor`, `n >= 0` with type `float`, or `double`.
name: A name for the operation (optional).
Returns:
The logarithm of \\(|Beta(x)|\\) reducing along the last dimension.
"""
# In the event that the last dimension has zero entries, we return -inf.
# This is consistent with a convention that the sum over the empty set 0, and
# the product is 1.
# This is standard. See https://en.wikipedia.org/wiki/Empty_set.
with ops.name_scope(name, 'lbeta', [x]):
x = ops.convert_to_tensor(x, name='x')
# Note reduce_sum([]) = 0.
log_prod_gamma_x = math_ops.reduce_sum(math_ops.lgamma(x), axis=[-1])
# Note lgamma(0) = infinity, so if x = []
# log_gamma_sum_x = lgamma(0) = infinity, and
# log_prod_gamma_x = lgamma(1) = 0,
# so result = -infinity
sum_x = math_ops.reduce_sum(x, axis=[-1])
log_gamma_sum_x = math_ops.lgamma(sum_x)
result = log_prod_gamma_x - log_gamma_sum_x
return result
@tf_export('math.bessel_i0')
def bessel_i0(x, name=None):
"""Computes the Bessel i0 function of `x` element-wise.
Modified Bessel function of order 0.
It is preferable to use the numerically stabler function `i0e(x)` instead.
Args:
x: A `Tensor` or `SparseTensor`. Must be one of the following types: `half`,
`float32`, `float64`.
name: A name for the operation (optional).
Returns:
A `Tensor` or `SparseTensor`, respectively. Has the same type as `x`.
@compatibility(scipy)
Equivalent to scipy.special.i0
@end_compatibility
"""
with ops.name_scope(name, 'bessel_i0', [x]):
return math_ops.exp(math_ops.abs(x)) * math_ops.bessel_i0e(x)
@tf_export('math.bessel_i1')
def bessel_i1(x, name=None):
"""Computes the Bessel i1 function of `x` element-wise.
Modified Bessel function of order 1.
It is preferable to use the numerically stabler function `i1e(x)` instead.
Args:
x: A `Tensor` or `SparseTensor`. Must be one of the following types: `half`,
`float32`, `float64`.
name: A name for the operation (optional).
Returns:
A `Tensor` or `SparseTensor`, respectively. Has the same type as `x`.
@compatibility(scipy)
Equivalent to scipy.special.i1
@end_compatibility
"""
with ops.name_scope(name, 'bessel_i1', [x]):
return math_ops.exp(math_ops.abs(x)) * math_ops.bessel_i1e(x)
@ops.RegisterGradient('XlaEinsum')
def _einsum_grad(op, grad):
equation = op.get_attr('equation')
if isinstance(equation, bytes):
equation = equation.decode()
inputs, output = equation.split('->')
left, right = inputs.split(',')
return [
gen_xla_ops.xla_einsum(
grad,
op.inputs[1],
equation='{},{}->{}'.format(output, right, left),
name=None),
gen_xla_ops.xla_einsum(
grad,
op.inputs[0],
equation='{},{}->{}'.format(output, left, right),
name=None)
]
def _enclosing_tpu_context():
# pylint: disable=protected-access
context = ops.get_default_graph()._get_control_flow_context()
# pylint: enable=protected-access
while context is not None and not isinstance(
context, control_flow_ops.XLAControlFlowContext):
context = context.outer_context
return context
@tf_export('einsum', 'linalg.einsum')
def einsum(equation, *inputs, **kwargs):
"""A generalized contraction between tensors of arbitrary dimension.
This function returns a tensor whose elements are defined by `equation`,
which is written in a shorthand form inspired by the Einstein summation
convention. As an example, consider multiplying two matrices
A and B to form a matrix C. The elements of C are given by:
```
C[i,k] = sum_j A[i,j] * B[j,k]
```
The corresponding `equation` is:
```
ij,jk->ik
```
In general, the `equation` is obtained from the more familiar element-wise
equation by
1. removing variable names, brackets, and commas,
2. replacing "*" with ",",
3. dropping summation signs, and
4. moving the output to the right, and replacing "=" with "->".
Many common operations can be expressed in this way. For example:
```python
# Matrix multiplication
>>> einsum('ij,jk->ik', m0, m1) # output[i,k] = sum_j m0[i,j] * m1[j, k]
# Dot product
>>> einsum('i,i->', u, v) # output = sum_i u[i]*v[i]
# Outer product
>>> einsum('i,j->ij', u, v) # output[i,j] = u[i]*v[j]
# Transpose
>>> einsum('ij->ji', m) # output[j,i] = m[i,j]
# Trace
>>> einsum('ii', m) # output[j,i] = trace(m) = sum_i m[i, i]
# Batch matrix multiplication
>>> einsum('aij,ajk->aik', s, t) # out[a,i,k] = sum_j s[a,i,j] * t[a, j, k]
```
To enable and control broadcasting, use an ellipsis. For example, to do
batch matrix multiplication, you could use:
```python
>>> einsum('...ij,...jk->...ik', u, v)
```
This function behaves like `numpy.einsum`, but does not support:
* Subscripts where an axis appears more than once for a single input
(e.g. `ijj,k->ik`) unless it is a trace (e.g. `ijji`).
Args:
equation: a `str` describing the contraction, in the same format as
`numpy.einsum`.
*inputs: the inputs to contract (each one a `Tensor`), whose shapes should
be consistent with `equation`.
name: A name for the operation (optional).
Returns:
The contracted `Tensor`, with shape determined by `equation`.
Raises:
ValueError: If
- the format of `equation` is incorrect,
- the number of inputs implied by `equation` does not match `len(inputs)`,
- an axis appears in the output subscripts but not in any of the inputs,
- the number of dimensions of an input differs from the number of
indices in its subscript, or
- the input shapes are inconsistent along a particular axis.
"""
name = kwargs.pop('name', None)
if kwargs:
raise TypeError('invalid keyword arguments for this function: ' + ', '.join(
[format(key) for key in sorted(list(kwargs.keys()))]))
with ops.name_scope(name, 'einsum', [equation, inputs]) as name:
inputs = list(inputs)
input_shapes = [x.get_shape() for x in inputs]
input_axis_labels, output_axis_labels = _einsum_parse_and_resolve_equation(
equation, input_shapes)
axis_labels = set(''.join(input_axis_labels) + output_axis_labels)
for a in axis_labels:
for input_labels in input_axis_labels:
if (len(input_axis_labels) == 1 and input_labels.count(a) == 2 and
input_labels == input_labels[::-1] and '->' not in equation):
return math_ops.trace(inputs[0])
if input_labels.count(a) > 1:
raise ValueError(
'Subscript not supported: an axis appears more than once: %s' %
input_labels)
for a in axis_labels:
input_count = sum(1 for s in input_axis_labels if a in s)
if input_count > 2 and a not in output_axis_labels:
logging.warn(
'Falling back to exponential-space implementation of einsum()'
' because index "%s" is summed over more than two inputs.', a)
return _exponential_space_einsum(equation, *inputs)
# Use xla_einsum if executing on TPU and if the operation is a 2 input
# einsum supported by XlaEinsumOp.
if _enclosing_tpu_context() is not None and len(inputs) == 2:
return gen_xla_ops.xla_einsum(
inputs[0], inputs[1], input_axis_labels[0] + ',' +
input_axis_labels[1] + '->' + output_axis_labels)
temp = inputs[0]
temp_axis_labels = input_axis_labels[0]
for i in xrange(len(inputs) - 1):
axes_to_sum = (
set(temp_axis_labels) &
set(input_axis_labels[i + 1]) - set(output_axis_labels))
temp, temp_axis_labels = _einsum_reduction(
temp, temp_axis_labels, inputs[i + 1], input_axis_labels[i + 1],
axes_to_sum)
missing_indices = set(temp_axis_labels) - set(output_axis_labels)
if missing_indices:
axis = [
i for i, a in enumerate(temp_axis_labels)
if a not in output_axis_labels
]
temp = math_ops.reduce_sum(temp, axis=axis)
temp_axis_labels = ''.join(
a for a in temp_axis_labels if a in output_axis_labels)
if sorted(temp_axis_labels) != sorted(output_axis_labels):
raise ValueError('Invalid equation: %s' % equation)
perm = [temp_axis_labels.index(a) for a in output_axis_labels]
return _transpose_if_necessary(temp, perm)
def _einsum_parse_and_resolve_equation(equation, input_shapes):
"""Helper for einsum() that splits/resolves inputs & outputs.
Args:
equation: Equation string given as argument to einsum().
input_shapes: List of the shapes of all inputs given to einsum()
Returns:
input_axis_labels, output_axis_labels where:
input_axis_labels: List of length len(input_shapes) of strings
representing the character label for each dimension of each given input,
resolving any broadcast (...) axes,
output_axis_labels: A string of character labels for each axes of output
tensor, filling in missing output subscripts and broadcast axes.
Raises:
ValueError: If equation is in the uncorrect format, incorrect number of
inputs given or broadcast axes "..." or output axes could not be resolved.
"""
equation = equation.replace(' ', '')
match = re.match('^([a-zA-Z,.]+)(->[a-zA-Z.]*)?$', equation)
if not match:
raise ValueError('Indices have incorrect format: %s' % equation)
input_axis_labels = match.group(1).split(',')
output_axis_labels = match.group(2)[2:] if match.group(2) else None
if len(input_shapes) != len(input_axis_labels):
raise ValueError('Got %d arguments for equation "%s", expecting %d' %
(len(input_shapes), equation, len(input_axis_labels)))
# Resolve Ellipsis
# Assign axes labels for unspecified dimensions in inputs. Labels taken
# from unused labels. Follow numpy einsum broadcasting conventions for
# tensors of different length and unlabeled output.
ellipsis_axes = ''
if '...' in equation:
unused = ''.join([c for c in string.ascii_letters
if c not in ''.join(input_axis_labels)])
for i, ax in enumerate(input_axis_labels):
if '...' in ax:
parts = ax.split('...')
if len(parts) != 2:
raise ValueError('Unable to resolve ellipsis. Excess number found.')
if input_shapes[i].ndims is None:
raise ValueError('Unable to statically infer ellipsis axes.')
n = input_shapes[i].ndims - len(''.join(parts))
if n < 0:
raise ValueError('Ellipses lengths do not match.')
if len(unused) < n:
raise ValueError(
'Unable to resolve ellipsis, too many distinct labels.')
replace_axes = unused[-n:] if n > 0 else ''
input_axis_labels[i] = input_axis_labels[i].replace('...',
replace_axes)
if len(replace_axes) > len(ellipsis_axes):
ellipsis_axes = replace_axes
if any(['.' in ax for ax in input_axis_labels]):
raise ValueError('period "." found outside of ellipsis')
if output_axis_labels is not None:
output_axis_labels = output_axis_labels.replace('...', ellipsis_axes)
if '.' in output_axis_labels:
raise ValueError('period "." found outside of ellipsis')
if output_axis_labels is None:
# infer the output subscripts if not given, assume alphabetical order,
# but always place ellipsis axes before given.
axis_labels = set(''.join(input_axis_labels)) - set(ellipsis_axes)
indices = ''.join(sorted(axis_labels))
counts = {ax: 0 for ax in indices}
for axes_ in input_axis_labels:
for ax in axes_:
if ax not in ellipsis_axes:
counts[ax] += 1
output_axis_labels = ellipsis_axes + ''.join(
sorted(ax for ax in axis_labels if counts[ax] == 1))
return input_axis_labels, output_axis_labels
def _einsum_reduction(t0, t0_axis_labels, t1, t1_axis_labels, axes_to_sum):
"""Helper for einsum() that computes the result of a two-argument einsum().
Args:
t0: a `Tensor`
t0_axis_labels: a string of axis labels. This string's length must equal
the rank of t0.
t1: a `Tensor`
t1_axis_labels: a string to axis labels. This string's length must equal
the rank of t1.
axes_to_sum: set of labels of axes to be summed over
Returns:
A `Tensor` whose elements are obtained by summing, over all axes in
`axes_to_sum`, the corresponding elements of `t0` and `t1`.
For example, if t0_axis_labels == 'abijk', t1_axis_labels == 'acjkl', and
axes_to_sum == {j,k}, this will return a tensor x where
out[a,b,c,i,l] = sum_j sum_k t0[a,b,i,j,k] * t1[a,c,j,k,l]
Raises:
ValueError: if the rank of `t0` does not match the length of
`t0_axis_labels`, or that of `t1` does not match the length of
`t1_axis_labels`.
"""
if len(t0_axis_labels) != len(t0.get_shape()):
raise ValueError(
'Tensor t0 of rank %d does not match einsum reduction of length %d' %
(len(t0.get_shape()), len(t0_axis_labels)))
if len(t1_axis_labels) != len(t1.get_shape()):
raise ValueError(
'Tensor t1 of rank %d does not match einsum reduction of length %d' %
(len(t1.get_shape()), len(t1_axis_labels)))
# This function computes the result of a two-argument einsum() using batch
# matrix multiplication. This involves
# 1. transposing t0 and t1 so that axes are in the correct order for
# batch matrix multiplication, and
# 2. reshaping t0 and t1 so that they are both of rank 3.
# First, we divide axes into three groups:
# * "preserved" axes are present in both inputs and the output
# * "summed" axes are present in both inputs but not the output
# * "broadcast" axes are present in exactly one input and the output
#
# As an example, if the einsum is abijk,acjkl->abcil, then "a" is a
# preserved axis, "b" and "c" are broadcast axes, and "j" and "k" are
# summed axes.
assert all(a in t0_axis_labels and a in t1_axis_labels for a in axes_to_sum)
preserved_axes = (set(t0_axis_labels) & set(t1_axis_labels)) - axes_to_sum
broadcast_axes = {}
for i, sym_list in enumerate([t0_axis_labels, t1_axis_labels]):
broadcast_axes[i] = set(sym_list) - preserved_axes - axes_to_sum
# Reorder the axes so that:
# 1. preserved axes come first in both inputs
# 2. in input 0, broadcast axes come next, followed by summed axes
# 3. in input 1, summed axes come next, followed by broadcast axes
def sort_key(input_index, a):
if a in preserved_axes:
return (-1, a)
elif ((input_index == 0 and a in broadcast_axes[0]) or
(input_index == 1 and a in axes_to_sum)):
return (0, a)
else:
return (1, a)
axis_labels = [t0_axis_labels, t1_axis_labels]
sorted_axes = [
sorted(sym_list, key=lambda a: sort_key(i, a))
for i, sym_list in enumerate(axis_labels)
]
inputs = [t0, t1]
for i, axes_str in enumerate(axis_labels):
perm = [axes_str.find(a) for a in sorted_axes[i]]
inputs[i] = _transpose_if_necessary(inputs[i], perm)
t0, t1 = inputs
if not axes_to_sum:
# In the special case where there are no axes to sum over, reduce to mul()
# rather than to batch matrix multiplication.
for _ in broadcast_axes[1]:
t0 = array_ops.expand_dims(t0, -1)
for _ in broadcast_axes[0]:
t1 = array_ops.expand_dims(t1, len(preserved_axes))
product = math_ops.multiply(t0, t1)
product_axes = sorted_axes[0] + sorted_axes[1][len(preserved_axes):]
return product, ''.join(product_axes)
else:
# Reduce to matmul().
# Reshape both inputs so as to combine multiple broadcast axes
# into a single axis, and combine multiple summed axes into a
# single axis.
t0_shape = _get_shape(t0)
num_broadcast_elements_t0 = _total_size(
t0_shape[len(preserved_axes):-len(axes_to_sum)])
num_summed_elements = _total_size(t0_shape[-len(axes_to_sum):])
new_shape = (
t0_shape[:len(preserved_axes)] +
[num_broadcast_elements_t0, num_summed_elements])
t0 = _reshape_if_necessary(t0, new_shape)
t1_shape = _get_shape(t1)
num_broadcast_elements_t1 = _total_size(
t1_shape[len(preserved_axes) + len(axes_to_sum):])
new_shape = (
t1_shape[:len(preserved_axes)] +
[num_summed_elements, num_broadcast_elements_t1])
t1 = _reshape_if_necessary(t1, new_shape)
product = math_ops.matmul(t0, t1)
# Undo compaction of broadcast axes
uncompacted_shape = (
t0_shape[:len(preserved_axes) + len(broadcast_axes[0])] +
t1_shape[len(t1_shape) - len(broadcast_axes[1]):])
product = _reshape_if_necessary(product, uncompacted_shape)
product_axes = (
sorted_axes[0][:len(preserved_axes) + len(broadcast_axes[0])] +
sorted_axes[1][len(sorted_axes[1]) - len(broadcast_axes[1]):])
return product, ''.join(product_axes)
def _transpose_if_necessary(tensor, perm):
"""Like transpose(), but avoids creating a new tensor if possible."""
if perm != range(len(perm)):
return array_ops.transpose(tensor, perm=perm)
else:
return tensor
def _reshape_if_necessary(tensor, new_shape):
"""Like reshape(), but avoids creating a new tensor if possible."""
# Accept None as an alias for -1 in new_shape.
new_shape = tuple(-1 if x is None else x for x in new_shape)
cur_shape = tuple(x.value for x in tensor.get_shape().dims)
if (len(new_shape) == len(cur_shape) and
all(d0 == d1 or d1 == -1 for d0, d1 in zip(cur_shape, new_shape))):
return tensor
else:
return array_ops.reshape(tensor, new_shape)
def _get_shape(tensor):
"""Like get_shape().as_list(), but explicitly queries the shape of a tensor
if necessary to ensure that the returned value contains no unknown value."""
shape = tensor.get_shape().as_list()
none_indices = [i for i, d in enumerate(shape) if d is None]
if none_indices:
# Query the shape if shape contains None values
shape_tensor = array_ops.shape(tensor)
for i in none_indices:
shape[i] = shape_tensor[i]
return shape
def _total_size(shape_values):
"""Given list of tensor shape values, returns total size.
If shape_values contains tensor values (which are results of
array_ops.shape), then it returns a scalar tensor.
If not, it returns an integer."""
result = 1
for val in shape_values:
result *= val
return result
def _exponential_space_einsum(equation, *inputs):
"""Fallback implementation that supports summing an index over > 2 inputs."""
inputs = list(inputs)
input_shapes = [x.get_shape() for x in inputs]
idx_in, idx_out = _einsum_parse_and_resolve_equation(equation, input_shapes)
idx_all = set(''.join(idx_in) + idx_out)
indices = ''.join(sorted(idx_all))
missing_idx = set(idx_out).difference(idx_all)
if missing_idx:
raise ValueError('Unknown output axes: %s' % missing_idx)
axis_order = {}
for ax in indices:
if ax not in idx_out:
axis_order[ax] = len(axis_order)
for ax in idx_out:
axis_order[ax] = len(axis_order)
# transpose inputs so axes are in order
for i, (input_, axes_) in enumerate(zip(inputs, idx_in)):
if input_.get_shape().ndims != len(axes_):
raise ValueError(
'Input %d with axes %s has incorrect' \
' number of dimensions (expected %d, got %d)' % (
i, axes_, len(axes_), input_.get_shape().ndims
)
)
sorted_idx = sorted(axes_, key=axis_order.get)
if len(set(axes_)) != len(axes_):
raise ValueError(
'Subscript not supported: an axis appears more than once: %s' % axes_)
if list(axes_) != sorted_idx:
permuted = [axes_.find(ax) for ax in sorted_idx]
inputs[i] = array_ops.transpose(input_, permuted)
idx_in[i] = sorted_idx
reduction_idx = []
shapes = [[dim if dim else -1
for dim in tensor.get_shape().as_list()]
for tensor in inputs]
# validate shapes for broadcasting
for j, ax in enumerate(sorted(idx_all, key=axis_order.get)):
dims = []
for i, idx in enumerate(idx_in):
if ax not in idx:
shapes[i].insert(j, 1)
else:
dim = shapes[i][j]
if isinstance(dim, int) and dim > 1:
dims.append(dim)
if len(set(dims)) > 1:
raise ValueError('Dimension mismatch on axis: %s' % ax)
if ax not in idx_out:
reduction_idx.append(j)
# reshape, multiply
expanded_inputs = [
array_ops.reshape(input_, shape) for input_, shape in zip(inputs, shapes)
]
expanded_output = 1
for input_ in expanded_inputs:
expanded_output *= input_
# contract
return math_ops.reduce_sum(expanded_output, reduction_idx)
|
tensorflow-master
|
tensorflow/python/ops/special_math_ops.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Gradient functions for optional ops."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import ops
from tensorflow.python.ops import gen_dataset_ops
@ops.RegisterGradient("OptionalFromValue")
def _OptionalFromValueGrad(op, grad):
return gen_dataset_ops.optional_get_value(
grad, [t.dtype for t in op.inputs], [t.shape for t in op.inputs])
@ops.RegisterGradient("OptionalGetValue")
def _OptionalGetValueGrad(unused_op, *grads):
return gen_dataset_ops.optional_from_value(grads)
|
tensorflow-master
|
tensorflow/python/ops/optional_grad.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Contains Gradient functions for image ops."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gen_image_ops
@ops.RegisterGradient("ResizeNearestNeighbor")
def _ResizeNearestNeighborGrad(op, grad):
"""The derivatives for nearest neighbor resizing.
Args:
op: The ResizeNearestNeighbor op.
grad: The tensor representing the gradient w.r.t. the output.
Returns:
The gradients w.r.t. the input and the output.
"""
image = op.inputs[0]
if image.get_shape()[1:3].is_fully_defined():
image_shape = image.get_shape()[1:3]
else:
image_shape = array_ops.shape(image)[1:3]
grads = gen_image_ops.resize_nearest_neighbor_grad(
grad,
image_shape,
align_corners=op.get_attr("align_corners"),
half_pixel_centers=op.get_attr("half_pixel_centers"))
return [grads, None]
@ops.RegisterGradient("ResizeBilinear")
def _ResizeBilinearGrad(op, grad):
"""The derivatives for bilinear resizing.
Args:
op: The ResizeBilinear op.
grad: The tensor representing the gradient w.r.t. the output.
Returns:
The gradients w.r.t. the input.
"""
grad0 = gen_image_ops.resize_bilinear_grad(
grad,
op.inputs[0],
align_corners=op.get_attr("align_corners"),
half_pixel_centers=op.get_attr("half_pixel_centers"))
return [grad0, None]
@ops.RegisterGradient("ScaleAndTranslate")
def _ScaleAndTranslateGrad(op, grad):
"""The derivatives for ScaleAndTranslate transformation op.
Args:
op: The ScaleAndTranslate op.
grad: The tensor representing the gradient w.r.t. the output.
Returns:
The gradients w.r.t. the input.
"""
grad0 = gen_image_ops.scale_and_translate_grad(
grad,
op.inputs[0],
op.inputs[2],
op.inputs[3],
kernel_type=op.get_attr("kernel_type"),
antialias=op.get_attr("antialias"))
return [grad0, None, None, None]
@ops.RegisterGradient("ResizeBicubic")
def _ResizeBicubicGrad(op, grad):
"""The derivatives for bicubic resizing.
Args:
op: The ResizeBicubic op.
grad: The tensor representing the gradient w.r.t. the output.
Returns:
The gradients w.r.t. the input.
"""
allowed_types = [dtypes.float32, dtypes.float64]
grad0 = None
if op.inputs[0].dtype in allowed_types:
grad0 = gen_image_ops.resize_bicubic_grad(
grad,
op.inputs[0],
align_corners=op.get_attr("align_corners"),
half_pixel_centers=op.get_attr("half_pixel_centers"))
return [grad0, None]
@ops.RegisterGradient("CropAndResize")
def _CropAndResizeGrad(op, grad):
"""The derivatives for crop_and_resize.
We back-propagate to the image only when the input image tensor has floating
point dtype but we always back-propagate to the input boxes tensor.
Args:
op: The CropAndResize op.
grad: The tensor representing the gradient w.r.t. the output.
Returns:
The gradients w.r.t. the input image, boxes, as well as the always-None
gradients w.r.t. box_ind and crop_size.
"""
image = op.inputs[0]
if image.get_shape().is_fully_defined():
image_shape = image.get_shape().as_list()
else:
image_shape = array_ops.shape(image)
allowed_types = [dtypes.float16, dtypes.float32, dtypes.float64]
if op.inputs[0].dtype in allowed_types:
# pylint: disable=protected-access
grad0 = gen_image_ops.crop_and_resize_grad_image(
grad, op.inputs[1], op.inputs[2], image_shape, T=op.get_attr("T"),
method=op.get_attr("method"))
# pylint: enable=protected-access
else:
grad0 = None
# `grad0` is the gradient to the input image pixels and it
# has been implemented for nearest neighbor and bilinear sampling
# respectively. `grad1` is the gradient to the input crop boxes' coordinates.
# When using nearest neighbor sampling, the gradient to crop boxes'
# coordinates are not well defined. In practice, we still approximate
# grad1 using the gradient derived from bilinear sampling.
grad1 = gen_image_ops.crop_and_resize_grad_boxes(
grad, op.inputs[0], op.inputs[1], op.inputs[2])
return [grad0, grad1, None, None]
|
tensorflow-master
|
tensorflow/python/ops/image_grad.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Script Language Operators."""
# pylint: disable=g-bad-name
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import threading
# Used by py_util.cc to get tracebacks.
import traceback # pylint: disable=unused-import
import weakref
import numpy as np
import six
from tensorflow.python import pywrap_tensorflow
from tensorflow.python.eager import backprop
from tensorflow.python.eager import context
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import function
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gen_script_ops
from tensorflow.python.ops import resource_variable_ops
from tensorflow.python.util import compat
from tensorflow.python.util import deprecation
from tensorflow.python.util import nest
from tensorflow.python.util.tf_export import tf_export
# Map from EagerPyFunc token to tuple (tape, eager args, eager outputs);
# used for differentiation.
tape_cache = {}
class EagerFunc(object):
"""A wrapper for a function owned by an EagerPyFunc."""
def __init__(self, func, Tout, is_grad_func):
"""Constructs an EagerFunc.
Args:
func: The function to wrap.
Tout: A list of datatypes for the output; an empty list if the output is
None.
is_grad_func: Whether this EagerFunc is the gradient of another
EagerPyFunc.
"""
self._func = func
self._out_dtypes = Tout
self._is_grad_func = is_grad_func
context.ensure_initialized()
def _convert(self, value, dtype):
"""Converts `value` to a tensor of type `dtype`, with error checking.
Args:
value: The tensor to convert.
dtype: The desired dtype.
Returns:
A tensor of type `dtype`, or a zeros tensor if value is None and
this function is in fact a grdient function.
Raises:
RuntimeError: if `value` is a variable.
"""
if isinstance(value, resource_variable_ops.ResourceVariable):
raise RuntimeError(
"Attempting to return a variable from an eagerly executed py_func. "
"Only numeric data structures like Tensors or NumPy arrays should "
"be returned; to return the value of a variable, make sure to obtain "
"the Tensor backing it by calling `.read_value()` on the variable in "
"question: %s" % value)
if value is None and self._is_grad_func:
# Gradient functions may legitimately return a list that contains
# both Tensors and Python Nones. Unfortuantely this breaks the
# OpKernel, so for now we replace None objects with zeros, which is
# mathematically correct but will prevent short-circuiting gradient
# computations.
#
# TODO(akshayka): Make it possible to return a list of both Tensors and
# Nones from an EagerPyFunc.
return constant_op.constant(0.0, dtype=dtype)
return ops.convert_to_tensor(value, dtype=dtype)
def __call__(self, device, token, args):
"""Passes `args` to `self._func`, which is executed eagerly."""
with context.eager_mode(), backprop.GradientTape() as tape:
for tensor in args:
tape.watch(tensor)
ret = self._func(*args)
# Use tf.identity to copy the returned tensors to device if neccesary.
with ops.device(device):
if isinstance(ret, (tuple, list)):
outputs = [
array_ops.identity(self._convert(x, dtype=dtype))
for (x, dtype) in zip(ret, self._out_dtypes)
]
elif ret is None:
outputs = None
else:
outputs = array_ops.identity(
self._convert(ret, dtype=self._out_dtypes[0]))
tape_cache[compat.as_bytes(token)] = (tape, args, outputs)
return outputs
class FuncRegistry(object):
"""A helper class to keep track of registered py functions.
FuncRegistry keeps a map from unique tokens (string) to python
functions, which takes numpy arrays and outputs numpy arrays.
"""
def __init__(self):
self._lock = threading.Lock()
self._unique_id = 0 # GUARDED_BY(self._lock)
# Only store weakrefs to the functions. The strong reference is stored in
# the graph.
self._funcs = weakref.WeakValueDictionary()
def insert(self, func):
"""Registers `func` and returns a unique token for this entry."""
token = self._next_unique_token()
# Store a weakref to the function
self._funcs[token] = func
return token
def remove(self, token):
"""Removes the registered function corresponding to `token`."""
self._funcs.pop(token, None)
@staticmethod
def _convert(value, dtype=None):
"""Converts an arg to numpy, avoiding dangerous string and unicode dtypes.
Numpy pads with zeros when using string and unicode dtypes if different
components of a tensor have different lengths. This is bad: ignoring the
padding is wrong for text data, and removing the padding is wrong for binary
data. To avoid this bug, we redo the conversion using an object dtype.
Additionally, we convert unicode strings to (byte-)strings for
compatibility.
Args:
value: Value to convert to a numpy array.
dtype: (Optional.) Desired NumPy type for the returned value.
Returns:
A numpy array.
"""
result = np.asarray(value, dtype=dtype, order="C")
if result.dtype.char == "S" and result is not value:
return np.asarray(value, order="C", dtype=object)
elif result.dtype.char == "U" and result is not value:
value = np.vectorize(lambda x: x.encode("utf8"))(value)
return np.asarray(value, order="C", dtype=object)
elif result.dtype.char == "U":
return result.astype(np.bytes_)
else:
return result
def __call__(self, token, device, args):
"""Calls the registered function for `token` with args.
Args:
token: A key into this `FuncRegistry` identifying which function to call.
device: Name of the device on which outputs of `token`'s corresponding
operation should be placed. Used iff the function registered for `token`
is an EagerPyFunc.
args: The arguments to pass to the function registered for `token`.
Returns:
The output of the function registered for `token`.
Raises:
ValueError: if no function is registered for `token`.
"""
func = self._funcs.get(token, None)
if func is None:
raise ValueError("callback %s is not found" % token)
if isinstance(func, EagerFunc):
# NB: Different invocations of the same py_func will share the same
# token, and the entries they stash in the tape_cache will collide.
# In practice, when executing a graph, this should only happen if
# the py_func is in a while_loop whose iterations are run in parallel
# or if the graph is being driven by concurrent session.run() calls.
#
# TODO(akshayka): Key the tape cache in a thread-safe way.
return func(device, token, args)
else:
ret = func(*args)
# Strings seem to lead to a memory leak here if they're not wrapped in a
# list.
if isinstance(ret, six.binary_type):
ret = [ret]
# Ensures that we return either a single numpy array or a list of numpy
# arrays.
if isinstance(ret, (tuple, list)):
return [self._convert(x) for x in ret]
else:
return self._convert(ret)
def size(self):
"""Returns how many functions are currently registered."""
return len(self._funcs)
def _next_unique_token(self):
"""Returns a unique token."""
with self._lock:
uid = self._unique_id
self._unique_id += 1
return "pyfunc_%d" % uid
# Global registry for py functions.
_py_funcs = FuncRegistry()
pywrap_tensorflow.InitializePyTrampoline(_py_funcs)
def _internal_py_func(func,
inp,
Tout,
stateful=None,
eager=False,
is_grad_func=False,
name=None):
"""See documentation for py_func and eager_py_func."""
is_list_or_tuple = False
if isinstance(Tout, (list, tuple)):
is_list_or_tuple = True
else:
Tout = [Tout]
if eager:
func = EagerFunc(func, Tout, is_grad_func)
token = _py_funcs.insert(func)
# We tie the registered function's lifetime with the current default graph,
# i.e., when the current graph is destroyed, we remove its py funcs.
graph = ops.get_default_graph()
# pylint: disable=protected-access
while isinstance(graph, function._FuncGraph):
# If the py_func was declared inside a _FuncGraph, its lifetime should be
# bound to that of the outer graph instead.
graph = graph._outer_graph
# TODO(zhifengc): Consider adding a Graph method to collect
# `cleanup` objects in one of its member.
if not hasattr(graph, "_py_funcs_used_in_graph"):
graph._py_funcs_used_in_graph = []
# Store a reference to the function in the graph to ensure it stays alive
# as long as the graph lives. When the graph is destroyed, the function
# is left to the garbage collector for destruction as well.
graph._py_funcs_used_in_graph.append(func)
# pylint: enable=protected-access
if eager:
result = gen_script_ops.eager_py_func(
input=inp, token=token, Tout=Tout, name=name)
else:
if stateful:
result = gen_script_ops.py_func(
input=inp, token=token, Tout=Tout, name=name)
else:
result = gen_script_ops.py_func_stateless(
input=inp, token=token, Tout=Tout, name=name)
return result if is_list_or_tuple else result[0]
# TODO(akshayka): Implement higher-order derivatives.
@ops.RegisterGradient("EagerPyFunc")
def _EagerPyFuncGrad(op, *dy):
"""Computes the gradient of an EagerPyFunc."""
token = op.get_attr("token")
def eagerly_executed_grad(*dy):
tape, eager_inputs, eager_outputs = tape_cache.pop(compat.as_bytes(token))
return tape.gradient(eager_outputs, eager_inputs, output_gradients=dy)
with ops.control_dependencies(op.outputs):
return _internal_py_func(
func=eagerly_executed_grad,
inp=dy,
Tout=[tensor.dtype for tensor in op.inputs],
eager=True,
is_grad_func=True)
@tf_export("py_function")
def eager_py_func(func, inp, Tout, name=None):
"""Wraps a python function into a TensorFlow op that executes it eagerly.
This function allows expressing computations in a TensorFlow graph as
Python functions. In particular, it wraps a Python function `func`
in a once-differentiable TensorFlow operation that executes it with eager
execution enabled. As a consequence, `tf.py_function` makes it
possible to express control flow using Python constructs (`if`, `while`,
`for`, etc.), instead of TensorFlow control flow constructs (`tf.cond`,
`tf.while_loop`). For example, you might use `tf.py_function` to
implement the log huber function:
```python
def log_huber(x, m):
if tf.abs(x) <= m:
return x**2
else:
return m**2 * (1 - 2 * tf.math.log(m) + tf.math.log(x**2))
x = tf.compat.v1.placeholder(tf.float32)
m = tf.compat.v1.placeholder(tf.float32)
y = tf.py_function(func=log_huber, inp=[x, m], Tout=tf.float32)
dy_dx = tf.gradients(y, x)[0]
with tf.compat.v1.Session() as sess:
# The session executes `log_huber` eagerly. Given the feed values below,
# it will take the first branch, so `y` evaluates to 1.0 and
# `dy_dx` evaluates to 2.0.
y, dy_dx = sess.run([y, dy_dx], feed_dict={x: 1.0, m: 2.0})
```
You can also use `tf.py_function` to debug your models at runtime
using Python tools, i.e., you can isolate portions of your code that
you want to debug, wrap them in Python functions and insert `pdb` tracepoints
or print statements as desired, and wrap those functions in
`tf.py_function`.
For more information on eager execution, see the
[Eager guide](https://tensorflow.org/guide/eager).
`tf.py_function` is similar in spirit to `tf.compat.v1.py_func`, but unlike
the latter, the former lets you use TensorFlow operations in the wrapped
Python function. In particular, while `tf.compat.v1.py_func` only runs on CPUs
and
wraps functions that take NumPy arrays as inputs and return NumPy arrays as
outputs, `tf.py_function` can be placed on GPUs and wraps functions
that take Tensors as inputs, execute TensorFlow operations in their bodies,
and return Tensors as outputs.
Like `tf.compat.v1.py_func`, `tf.py_function` has the following limitations
with respect to serialization and distribution:
* The body of the function (i.e. `func`) will not be serialized in a
`GraphDef`. Therefore, you should not use this function if you need to
serialize your model and restore it in a different environment.
* The operation must run in the same address space as the Python program
that calls `tf.py_function()`. If you are using distributed
TensorFlow, you must run a `tf.distribute.Server` in the same process as the
program that calls `tf.py_function()` and you must pin the created
operation to a device in that server (e.g. using `with tf.device():`).
Args:
func: A Python function which accepts a list of `Tensor` objects having
element types that match the corresponding `tf.Tensor` objects in `inp`
and returns a list of `Tensor` objects (or a single `Tensor`, or `None`)
having element types that match the corresponding values in `Tout`.
inp: A list of `Tensor` objects.
Tout: A list or tuple of tensorflow data types or a single tensorflow data
type if there is only one, indicating what `func` returns; an empty list
if no value is returned (i.e., if the return value is `None`).
name: A name for the operation (optional).
Returns:
A list of `Tensor` or a single `Tensor` which `func` computes; an empty list
if `func` returns None.
"""
return _internal_py_func(func=func, inp=inp, Tout=Tout, eager=True, name=name)
def py_func_common(func, inp, Tout, stateful=True, name=None):
"""Wraps a python function and uses it as a TensorFlow op.
Given a python function `func`, which takes numpy arrays as its
arguments and returns numpy arrays as its outputs, wrap this function as an
operation in a TensorFlow graph. The following snippet constructs a simple
TensorFlow graph that invokes the `np.sinh()` NumPy function as a operation
in the graph:
```python
def my_func(x):
# x will be a numpy array with the contents of the placeholder below
return np.sinh(x)
input = tf.compat.v1.placeholder(tf.float32)
y = tf.compat.v1.py_func(my_func, [input], tf.float32)
```
**N.B.** The `tf.compat.v1.py_func()` operation has the following known
limitations:
* The body of the function (i.e. `func`) will not be serialized in a
`GraphDef`. Therefore, you should not use this function if you need to
serialize your model and restore it in a different environment.
* The operation must run in the same address space as the Python program
that calls `tf.compat.v1.py_func()`. If you are using distributed
TensorFlow, you
must run a `tf.distribute.Server` in the same process as the program that
calls
`tf.compat.v1.py_func()` and you must pin the created operation to a device
in that
server (e.g. using `with tf.device():`).
Args:
func: A Python function, which accepts `ndarray` objects as arguments and
returns a list of `ndarray` objects (or a single `ndarray`). This function
must accept as many arguments as there are tensors in `inp`, and these
argument types will match the corresponding `tf.Tensor` objects in `inp`.
The returns `ndarray`s must match the number and types defined `Tout`.
Important Note: Input and output numpy `ndarray`s of `func` are not
guaranteed to be copies. In some cases their underlying memory will be
shared with the corresponding TensorFlow tensors. In-place modification
or storing `func` input or return values in python datastructures
without explicit (np.)copy can have non-deterministic consequences.
inp: A list of `Tensor` objects.
Tout: A list or tuple of tensorflow data types or a single tensorflow data
type if there is only one, indicating what `func` returns.
stateful: (Boolean.) If True, the function should be considered stateful. If
a function is stateless, when given the same input it will return the same
output and have no observable side effects. Optimizations such as common
subexpression elimination are only performed on stateless operations.
name: A name for the operation (optional).
Returns:
A list of `Tensor` or a single `Tensor` which `func` computes.
"""
if context.executing_eagerly():
result = func(*[x.numpy() for x in inp])
result = nest.flatten(result)
result = [x if x is None else ops.convert_to_tensor(x) for x in result]
if len(result) == 1:
# Mimic the automatic unwrapping in graph-mode py_func
result, = result
return result
return _internal_py_func(
func=func, inp=inp, Tout=Tout, stateful=stateful, eager=False, name=name)
@deprecation.deprecated(
date=None,
instructions="""tf.py_func is deprecated in TF V2. Instead, there are two
options available in V2.
- tf.py_function takes a python function which manipulates tf eager
tensors instead of numpy arrays. It's easy to convert a tf eager tensor to
an ndarray (just call tensor.numpy()) but having access to eager tensors
means `tf.py_function`s can use accelerators such as GPUs as well as
being differentiable using a gradient tape.
- tf.numpy_function maintains the semantics of the deprecated tf.py_func
(it is not differentiable, and manipulates numpy arrays). It drops the
stateful argument making all functions stateful.
""")
@tf_export(v1=["py_func"])
def py_func(func, inp, Tout, stateful=True, name=None):
return py_func_common(func, inp, Tout, stateful, name=name)
py_func.__doc__ = "%s" % py_func_common.__doc__
@tf_export("numpy_function")
def numpy_function(func, inp, Tout, name=None):
return py_func_common(func, inp, Tout, stateful=True, name=name)
numpy_function.__doc__ = py_func_common.__doc__.replace("py_func",
"numpy_function")
ops.NotDifferentiable("PyFunc")
ops.NotDifferentiable("PyFuncStateless")
|
tensorflow-master
|
tensorflow/python/ops/script_ops.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Gradients for operators defined in state_ops.py."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import ops
# TODO(b/31222613): These ops may be differentiable, and there may be
# latent bugs here.
ops.NotDifferentiable("Assign")
ops.NotDifferentiable("AssignAdd")
ops.NotDifferentiable("AssignSub")
ops.NotDifferentiable("ScatterAdd")
ops.NotDifferentiable("ScatterSub")
ops.NotDifferentiable("ScatterMul")
ops.NotDifferentiable("ScatterDiv")
ops.NotDifferentiable("ScatterNdUpdate")
ops.NotDifferentiable("ScatterNdAdd")
ops.NotDifferentiable("ScatterNdSub")
ops.NotDifferentiable("ScatterNdMul")
ops.NotDifferentiable("ScatterNdDiv")
|
tensorflow-master
|
tensorflow/python/ops/state_grad.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# =============================================================================
"""Implementation of Neural Net (NN) functions."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import math
from tensorflow.python.compat import compat
from tensorflow.python.distribute import distribution_strategy_context as ds
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import function
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import candidate_sampling_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import embedding_ops
from tensorflow.python.ops import gen_array_ops # pylint: disable=unused-import
from tensorflow.python.ops import gen_nn_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import nn_ops
from tensorflow.python.ops import gen_sparse_ops
from tensorflow.python.ops import variables
from tensorflow.python.ops.losses import util as losses_util
from tensorflow.python.util.deprecation import deprecated_args
from tensorflow.python.util.deprecation import deprecated_argument_lookup
from tensorflow.python.util.tf_export import tf_export
@tf_export("nn.log_poisson_loss")
def log_poisson_loss(targets, log_input, compute_full_loss=False, name=None):
"""Computes log Poisson loss given `log_input`.
Gives the log-likelihood loss between the prediction and the target under the
assumption that the target has a Poisson distribution.
Caveat: By default, this is not the exact loss, but the loss minus a
constant term [log(z!)]. That has no effect for optimization, but
does not play well with relative loss comparisons. To compute an
approximation of the log factorial term, specify
compute_full_loss=True to enable Stirling's Approximation.
For brevity, let `c = log(x) = log_input`, `z = targets`. The log Poisson
loss is
-log(exp(-x) * (x^z) / z!)
= -log(exp(-x) * (x^z)) + log(z!)
~ -log(exp(-x)) - log(x^z) [+ z * log(z) - z + 0.5 * log(2 * pi * z)]
[ Note the second term is the Stirling's Approximation for log(z!).
It is invariant to x and does not affect optimization, though
important for correct relative loss comparisons. It is only
computed when compute_full_loss == True. ]
= x - z * log(x) [+ z * log(z) - z + 0.5 * log(2 * pi * z)]
= exp(c) - z * c [+ z * log(z) - z + 0.5 * log(2 * pi * z)]
Args:
targets: A `Tensor` of the same type and shape as `log_input`.
log_input: A `Tensor` of type `float32` or `float64`.
compute_full_loss: whether to compute the full loss. If false, a constant
term is dropped in favor of more efficient optimization.
name: A name for the operation (optional).
Returns:
A `Tensor` of the same shape as `log_input` with the componentwise
logistic losses.
Raises:
ValueError: If `log_input` and `targets` do not have the same shape.
"""
with ops.name_scope(name, "log_poisson_loss", [log_input, targets]) as name:
log_input = ops.convert_to_tensor(log_input, name="log_input")
targets = ops.convert_to_tensor(targets, name="targets")
try:
targets.get_shape().merge_with(log_input.get_shape())
except ValueError:
raise ValueError(
"log_input and targets must have the same shape (%s vs %s)" %
(log_input.get_shape(), targets.get_shape()))
result = math_ops.exp(log_input) - log_input * targets
if compute_full_loss:
# need to create constant tensors here so that their dtypes can be matched
# to that of the targets.
point_five = constant_op.constant(0.5, dtype=targets.dtype)
two_pi = constant_op.constant(2 * math.pi, dtype=targets.dtype)
stirling_approx = (targets * math_ops.log(targets)) - targets + (
point_five * math_ops.log(two_pi * targets))
zeros = array_ops.zeros_like(targets, dtype=targets.dtype)
ones = array_ops.ones_like(targets, dtype=targets.dtype)
cond = math_ops.logical_and(targets >= zeros, targets <= ones)
result += array_ops.where(cond, zeros, stirling_approx)
return result
@tf_export(v1=["nn.sigmoid_cross_entropy_with_logits"])
def sigmoid_cross_entropy_with_logits( # pylint: disable=invalid-name
_sentinel=None,
labels=None,
logits=None,
name=None):
"""Computes sigmoid cross entropy given `logits`.
Measures the probability error in discrete classification tasks in which each
class is independent and not mutually exclusive. For instance, one could
perform multilabel classification where a picture can contain both an elephant
and a dog at the same time.
For brevity, let `x = logits`, `z = labels`. The logistic loss is
z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))
= z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))
= z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))
= z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))
= (1 - z) * x + log(1 + exp(-x))
= x - x * z + log(1 + exp(-x))
For x < 0, to avoid overflow in exp(-x), we reformulate the above
x - x * z + log(1 + exp(-x))
= log(exp(x)) - x * z + log(1 + exp(-x))
= - x * z + log(1 + exp(x))
Hence, to ensure stability and avoid overflow, the implementation uses this
equivalent formulation
max(x, 0) - x * z + log(1 + exp(-abs(x)))
`logits` and `labels` must have the same type and shape.
Args:
_sentinel: Used to prevent positional parameters. Internal, do not use.
labels: A `Tensor` of the same type and shape as `logits`.
logits: A `Tensor` of type `float32` or `float64`.
name: A name for the operation (optional).
Returns:
A `Tensor` of the same shape as `logits` with the componentwise
logistic losses.
Raises:
ValueError: If `logits` and `labels` do not have the same shape.
"""
# pylint: disable=protected-access
nn_ops._ensure_xent_args("sigmoid_cross_entropy_with_logits", _sentinel,
labels, logits)
# pylint: enable=protected-access
with ops.name_scope(name, "logistic_loss", [logits, labels]) as name:
logits = ops.convert_to_tensor(logits, name="logits")
labels = ops.convert_to_tensor(labels, name="labels")
try:
labels.get_shape().merge_with(logits.get_shape())
except ValueError:
raise ValueError("logits and labels must have the same shape (%s vs %s)" %
(logits.get_shape(), labels.get_shape()))
# The logistic loss formula from above is
# x - x * z + log(1 + exp(-x))
# For x < 0, a more numerically stable formula is
# -x * z + log(1 + exp(x))
# Note that these two expressions can be combined into the following:
# max(x, 0) - x * z + log(1 + exp(-abs(x)))
# To allow computing gradients at zero, we define custom versions of max and
# abs functions.
zeros = array_ops.zeros_like(logits, dtype=logits.dtype)
cond = (logits >= zeros)
relu_logits = array_ops.where(cond, logits, zeros)
neg_abs_logits = array_ops.where(cond, -logits, logits)
return math_ops.add(
relu_logits - logits * labels,
math_ops.log1p(math_ops.exp(neg_abs_logits)),
name=name)
# Note: intentionally calling this v2 to not allow existing code with indirect
# imports to ignore the sentinel behavior.
@tf_export("nn.sigmoid_cross_entropy_with_logits", v1=[])
def sigmoid_cross_entropy_with_logits_v2( # pylint: disable=invalid-name
labels=None,
logits=None,
name=None):
"""Computes sigmoid cross entropy given `logits`.
Measures the probability error in discrete classification tasks in which each
class is independent and not mutually exclusive. For instance, one could
perform multilabel classification where a picture can contain both an elephant
and a dog at the same time.
For brevity, let `x = logits`, `z = labels`. The logistic loss is
z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))
= z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))
= z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))
= z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))
= (1 - z) * x + log(1 + exp(-x))
= x - x * z + log(1 + exp(-x))
For x < 0, to avoid overflow in exp(-x), we reformulate the above
x - x * z + log(1 + exp(-x))
= log(exp(x)) - x * z + log(1 + exp(-x))
= - x * z + log(1 + exp(x))
Hence, to ensure stability and avoid overflow, the implementation uses this
equivalent formulation
max(x, 0) - x * z + log(1 + exp(-abs(x)))
`logits` and `labels` must have the same type and shape.
Args:
labels: A `Tensor` of the same type and shape as `logits`.
logits: A `Tensor` of type `float32` or `float64`.
name: A name for the operation (optional).
Returns:
A `Tensor` of the same shape as `logits` with the componentwise
logistic losses.
Raises:
ValueError: If `logits` and `labels` do not have the same shape.
"""
return sigmoid_cross_entropy_with_logits(
logits=logits, labels=labels, name=name)
@tf_export("nn.weighted_cross_entropy_with_logits", v1=[])
def weighted_cross_entropy_with_logits_v2(labels, logits, pos_weight,
name=None):
"""Computes a weighted cross entropy.
This is like `sigmoid_cross_entropy_with_logits()` except that `pos_weight`,
allows one to trade off recall and precision by up- or down-weighting the
cost of a positive error relative to a negative error.
The usual cross-entropy cost is defined as:
labels * -log(sigmoid(logits)) +
(1 - labels) * -log(1 - sigmoid(logits))
A value `pos_weights > 1` decreases the false negative count, hence increasing
the recall.
Conversely setting `pos_weights < 1` decreases the false positive count and
increases the precision.
This can be seen from the fact that `pos_weight` is introduced as a
multiplicative coefficient for the positive labels term
in the loss expression:
labels * -log(sigmoid(logits)) * pos_weight +
(1 - labels) * -log(1 - sigmoid(logits))
For brevity, let `x = logits`, `z = labels`, `q = pos_weight`.
The loss is:
qz * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))
= qz * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))
= qz * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))
= qz * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))
= (1 - z) * x + (qz + 1 - z) * log(1 + exp(-x))
= (1 - z) * x + (1 + (q - 1) * z) * log(1 + exp(-x))
Setting `l = (1 + (q - 1) * z)`, to ensure stability and avoid overflow,
the implementation uses
(1 - z) * x + l * (log(1 + exp(-abs(x))) + max(-x, 0))
`logits` and `labels` must have the same type and shape.
Args:
labels: A `Tensor` of the same type and shape as `logits`.
logits: A `Tensor` of type `float32` or `float64`.
pos_weight: A coefficient to use on the positive examples.
name: A name for the operation (optional).
Returns:
A `Tensor` of the same shape as `logits` with the componentwise
weighted logistic losses.
Raises:
ValueError: If `logits` and `labels` do not have the same shape.
"""
with ops.name_scope(name, "logistic_loss", [logits, labels]) as name:
logits = ops.convert_to_tensor(logits, name="logits")
labels = ops.convert_to_tensor(labels, name="labels")
try:
labels.get_shape().merge_with(logits.get_shape())
except ValueError:
raise ValueError("logits and labels must have the same shape (%s vs %s)" %
(logits.get_shape(), labels.get_shape()))
# The logistic loss formula from above is
# (1 - z) * x + (1 + (q - 1) * z) * log(1 + exp(-x))
# For x < 0, a more numerically stable formula is
# (1 - z) * x + (1 + (q - 1) * z) * log(1 + exp(x)) - l * x
# To avoid branching, we use the combined version
# (1 - z) * x + l * (log(1 + exp(-abs(x))) + max(-x, 0))
log_weight = 1 + (pos_weight - 1) * labels
return math_ops.add(
(1 - labels) * logits,
log_weight * (math_ops.log1p(math_ops.exp(-math_ops.abs(logits))) +
nn_ops.relu(-logits)),
name=name)
@tf_export(v1=["nn.weighted_cross_entropy_with_logits"])
@deprecated_args(None, "targets is deprecated, use labels instead", "targets")
def weighted_cross_entropy_with_logits(labels=None,
logits=None,
pos_weight=None,
name=None,
targets=None):
"""Computes a weighted cross entropy.
This is like `sigmoid_cross_entropy_with_logits()` except that `pos_weight`,
allows one to trade off recall and precision by up- or down-weighting the
cost of a positive error relative to a negative error.
The usual cross-entropy cost is defined as:
labels * -log(sigmoid(logits)) +
(1 - labels) * -log(1 - sigmoid(logits))
A value `pos_weights > 1` decreases the false negative count, hence increasing
the recall.
Conversely setting `pos_weights < 1` decreases the false positive count and
increases the precision.
This can be seen from the fact that `pos_weight` is introduced as a
multiplicative coefficient for the positive labels term
in the loss expression:
labels * -log(sigmoid(logits)) * pos_weight +
(1 - labels) * -log(1 - sigmoid(logits))
For brevity, let `x = logits`, `z = labels`, `q = pos_weight`.
The loss is:
qz * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))
= qz * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))
= qz * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))
= qz * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))
= (1 - z) * x + (qz + 1 - z) * log(1 + exp(-x))
= (1 - z) * x + (1 + (q - 1) * z) * log(1 + exp(-x))
Setting `l = (1 + (q - 1) * z)`, to ensure stability and avoid overflow,
the implementation uses
(1 - z) * x + l * (log(1 + exp(-abs(x))) + max(-x, 0))
`logits` and `labels` must have the same type and shape.
Args:
labels: A `Tensor` of the same type and shape as `logits`.
logits: A `Tensor` of type `float32` or `float64`.
pos_weight: A coefficient to use on the positive examples.
name: A name for the operation (optional).
targets: Deprecated alias for labels.
Returns:
A `Tensor` of the same shape as `logits` with the componentwise
weighted logistic losses.
Raises:
ValueError: If `logits` and `labels` do not have the same shape.
"""
labels = deprecated_argument_lookup("labels", labels, "targets", targets)
return weighted_cross_entropy_with_logits_v2(labels, logits, pos_weight, name)
@tf_export("nn.compute_average_loss")
def compute_average_loss(per_example_loss,
sample_weight=None,
global_batch_size=None):
"""Scales per-example losses with sample_weights and computes their average.
Usage with distribution strategy and custom training loop:
```python
with strategy.scope():
def compute_loss(labels, predictions, sample_weight=None):
# If you are using a `Loss` class instead, set reduction to `NONE` so that
# we can do the reduction afterwards and divide by global batch size.
per_example_loss = tf.keras.losses.sparse_categorical_crossentropy(
labels, predictions)
# Compute loss that is scaled by sample_weight and by global batch size.
return tf.compute_average_loss(
per_example_loss,
sample_weight=sample_weight,
global_batch_size=GLOBAL_BATCH_SIZE)
```
Args:
per_example_loss: Per-example loss.
sample_weight: Optional weighting for each example.
global_batch_size: Optional global batch size value. Defaults to (size of
first dimension of `losses`) * (number of replicas).
Returns:
Scalar loss value.
""" # pylint: disable=g-doc-exception
per_example_loss = ops.convert_to_tensor(per_example_loss)
input_dtype = per_example_loss.dtype
with losses_util.check_per_example_loss_rank(per_example_loss):
if sample_weight is not None:
per_example_loss = losses_util.scale_losses_by_sample_weight(
per_example_loss, sample_weight)
per_example_loss = math_ops.cast(per_example_loss, input_dtype)
if global_batch_size is None:
if ds.has_strategy() and ds.in_cross_replica_context():
raise RuntimeError(
"You are calling `compute_average_loss` in cross replica context, "
"while it was expected to be called in replica context.")
num_replicas = ds.get_strategy().num_replicas_in_sync
per_replica_batch_size = array_ops.shape_v2(per_example_loss)[0]
global_batch_size = per_replica_batch_size * num_replicas
global_batch_size = math_ops.cast(global_batch_size, input_dtype)
return math_ops.reduce_sum(per_example_loss) / global_batch_size
@tf_export("nn.scale_regularization_loss")
def scale_regularization_loss(regularization_loss):
"""Scales the sum of the given regularization losses by number of replicas.
Usage with distribution strategy and custom training loop:
```python
with strategy.scope():
def compute_loss(self, label, predictions):
per_example_loss = tf.keras.losses.sparse_categorical_crossentropy(
labels, predictions)
# Compute loss that is scaled by sample_weight and by global batch size.
loss = tf.compute_average_loss(
per_example_loss,
sample_weight=sample_weight,
global_batch_size=GLOBAL_BATCH_SIZE)
# Add scaled regularization losses.
loss += tf.scale_regularization_loss(tf.nn.l2_loss(weights))
return loss
```
Args:
regularization_loss: Regularization loss.
Returns:
Scalar loss value.
""" # pylint: disable=g-doc-exception
if ds.has_strategy() and ds.in_cross_replica_context():
raise RuntimeError(
"You are calling `scale_regularization_loss` in cross replica context, "
"while it was expected to be called in replica context.")
num_replicas = ds.get_strategy().num_replicas_in_sync
return math_ops.reduce_sum(regularization_loss) / num_replicas
@tf_export(v1=["nn.relu_layer"])
def relu_layer(x, weights, biases, name=None):
"""Computes Relu(x * weight + biases).
Args:
x: a 2D tensor. Dimensions typically: batch, in_units
weights: a 2D tensor. Dimensions typically: in_units, out_units
biases: a 1D tensor. Dimensions: out_units
name: A name for the operation (optional). If not specified
"nn_relu_layer" is used.
Returns:
A 2-D Tensor computing relu(matmul(x, weights) + biases).
Dimensions typically: batch, out_units.
"""
with ops.name_scope(name, "relu_layer", [x, weights, biases]) as name:
x = ops.convert_to_tensor(x, name="x")
weights = ops.convert_to_tensor(weights, name="weights")
biases = ops.convert_to_tensor(biases, name="biases")
xw_plus_b = nn_ops.bias_add(math_ops.matmul(x, weights), biases)
return nn_ops.relu(xw_plus_b, name=name)
def _swish_shape(op):
"""Shape helper function for swish and _swish_grad function below."""
return [op.inputs[0].shape]
@function.Defun(shape_func=_swish_shape, func_name="swish_grad", noinline=True)
def _swish_grad(features, grad):
"""Gradient of Swish function defined below."""
sigmoid_features = math_ops.sigmoid(features)
activation_grad = (
sigmoid_features * (1.0 + features * (1.0 - sigmoid_features)))
return grad * activation_grad
# Naively, x * tf.nn.sigmoid(x) requires keeping both x and sigmoid(x) around
# for backprop, effectively doubling the tensor's memory consumption. We use a
# @Defun decorator with noinline=True so that sigmoid(features) is re-computed
# during backprop, and we can free the sigmoid(features) expression immediately
# after use during the forward pass.
@tf_export("nn.swish")
@function.Defun(
grad_func=_swish_grad,
shape_func=_swish_shape,
func_name="swish",
noinline=True)
def swish(features):
# pylint: disable=g-doc-args
"""Computes the Swish activation function: `x * sigmoid(x)`.
Source: "Searching for Activation Functions" (Ramachandran et al. 2017)
https://arxiv.org/abs/1710.05941
Args:
features: A `Tensor` representing preactivation values.
name: A name for the operation (optional).
Returns:
The activation value.
"""
# pylint: enable=g-doc-args
features = ops.convert_to_tensor(features, name="features")
return features * math_ops.sigmoid(features)
@tf_export(v1=["math.l2_normalize", "linalg.l2_normalize", "nn.l2_normalize"])
@deprecated_args(None, "dim is deprecated, use axis instead", "dim")
def l2_normalize(x, axis=None, epsilon=1e-12, name=None, dim=None):
"""Normalizes along dimension `axis` using an L2 norm.
For a 1-D tensor with `axis = 0`, computes
output = x / sqrt(max(sum(x**2), epsilon))
For `x` with more dimensions, independently normalizes each 1-D slice along
dimension `axis`.
Args:
x: A `Tensor`.
axis: Dimension along which to normalize. A scalar or a vector of
integers.
epsilon: A lower bound value for the norm. Will use `sqrt(epsilon)` as the
divisor if `norm < sqrt(epsilon)`.
name: A name for this operation (optional).
dim: Deprecated alias for axis.
Returns:
A `Tensor` with the same shape as `x`.
"""
axis = deprecated_argument_lookup("axis", axis, "dim", dim)
return l2_normalize_v2(x, axis, epsilon, name)
@tf_export("math.l2_normalize", "linalg.l2_normalize", "nn.l2_normalize", v1=[])
def l2_normalize_v2(x, axis=None, epsilon=1e-12, name=None):
"""Normalizes along dimension `axis` using an L2 norm.
For a 1-D tensor with `axis = 0`, computes
output = x / sqrt(max(sum(x**2), epsilon))
For `x` with more dimensions, independently normalizes each 1-D slice along
dimension `axis`.
Args:
x: A `Tensor`.
axis: Dimension along which to normalize. A scalar or a vector of
integers.
epsilon: A lower bound value for the norm. Will use `sqrt(epsilon)` as the
divisor if `norm < sqrt(epsilon)`.
name: A name for this operation (optional).
Returns:
A `Tensor` with the same shape as `x`.
"""
with ops.name_scope(name, "l2_normalize", [x]) as name:
x = ops.convert_to_tensor(x, name="x")
square_sum = math_ops.reduce_sum(math_ops.square(x), axis, keepdims=True)
x_inv_norm = math_ops.rsqrt(math_ops.maximum(square_sum, epsilon))
return math_ops.multiply(x, x_inv_norm, name=name)
def _count_nonzero(input_tensor, dtype=dtypes.int64):
"""Same as math_ops.count_nonzero.
The reduction is done in dtype, which can be faster for 32-bit dtypes.
Args:
input_tensor: numeric tensor
dtype: reduction dtype
Returns:
number of nonzero values with type dtype
"""
with ops.name_scope("count_nonzero", values=[input_tensor]):
zero = array_ops.zeros([], dtype=input_tensor.dtype)
nonzero_count = math_ops.reduce_sum(
math_ops.cast(
math_ops.not_equal(input_tensor, zero),
dtype=dtype), name="nonzero_count")
return nonzero_count
@tf_export("math.zero_fraction", "nn.zero_fraction")
def zero_fraction(value, name=None):
"""Returns the fraction of zeros in `value`.
If `value` is empty, the result is `nan`.
This is useful in summaries to measure and report sparsity. For example,
```python
z = tf.nn.relu(...)
summ = tf.compat.v1.summary.scalar('sparsity', tf.nn.zero_fraction(z))
```
Args:
value: A tensor of numeric type.
name: A name for the operation (optional).
Returns:
The fraction of zeros in `value`, with type `float32`.
"""
with ops.name_scope(name, "zero_fraction", [value]):
value = ops.convert_to_tensor(value, name="value")
size = array_ops.size(value, out_type=dtypes.int64)
# If the count is small, we can save memory/CPU with an int32 reduction.
num_nonzero = control_flow_ops.cond(
size <= dtypes.int32.max,
# pylint: disable=g-long-lambda
true_fn=lambda: math_ops.cast(
_count_nonzero(value, dtype=dtypes.int32),
dtype=dtypes.int64),
false_fn=lambda: _count_nonzero(value, dtype=dtypes.int64))
with ops.name_scope("counts_to_fraction"):
num_zero = size - num_nonzero
num_zero_float32 = math_ops.cast(num_zero, dtype=dtypes.float32)
size_float32 = math_ops.cast(size, dtype=dtypes.float32)
zero_fraction_float32 = num_zero_float32 / size_float32
return array_ops.identity(zero_fraction_float32, "fraction")
# pylint: disable=redefined-builtin
@tf_export(v1=["nn.depthwise_conv2d"])
def depthwise_conv2d(input,
filter,
strides,
padding,
rate=None,
name=None,
data_format=None,
dilations=None):
"""Depthwise 2-D convolution.
Given a 4D input tensor ('NHWC' or 'NCHW' data formats)
and a filter tensor of shape
`[filter_height, filter_width, in_channels, channel_multiplier]`
containing `in_channels` convolutional filters of depth 1, `depthwise_conv2d`
applies a different filter to each input channel (expanding from 1 channel
to `channel_multiplier` channels for each), then concatenates the results
together. The output has `in_channels * channel_multiplier` channels.
In detail, with the default NHWC format,
output[b, i, j, k * channel_multiplier + q] = sum_{di, dj}
filter[di, dj, k, q] * input[b, strides[1] * i + rate[0] * di,
strides[2] * j + rate[1] * dj, k]
Must have `strides[0] = strides[3] = 1`. For the most common case of the
same horizontal and vertical strides, `strides = [1, stride, stride, 1]`.
If any value in `rate` is greater than 1, we perform atrous depthwise
convolution, in which case all values in the `strides` tensor must be equal
to 1.
Args:
input: 4-D with shape according to `data_format`.
filter: 4-D with shape
`[filter_height, filter_width, in_channels, channel_multiplier]`.
strides: 1-D of size 4. The stride of the sliding window for each
dimension of `input`.
padding: A string, either `'VALID'` or `'SAME'`. The padding algorithm.
See the "returns" section of `tf.nn.convolution` for details.
rate: 1-D of size 2. The dilation rate in which we sample input values
across the `height` and `width` dimensions in atrous convolution. If it is
greater than 1, then all values of strides must be 1.
name: A name for this operation (optional).
data_format: The data format for input. Either "NHWC" (default) or "NCHW".
dilations: Alias of rate.
Returns:
A 4-D `Tensor` with shape according to `data_format`. E.g., for
"NHWC" format, shape is
`[batch, out_height, out_width, in_channels * channel_multiplier].`
"""
rate = deprecated_argument_lookup("dilations", dilations, "rate", rate)
with ops.name_scope(name, "depthwise", [input, filter]) as name:
input = ops.convert_to_tensor(input, name="tensor_in")
filter = ops.convert_to_tensor(filter, name="filter_in")
if rate is None:
rate = [1, 1]
def op(input_converted, _, padding):
return nn_ops.depthwise_conv2d_native(
input=input_converted,
filter=filter,
strides=strides,
padding=padding,
data_format=data_format,
name=name)
return nn_ops.with_space_to_batch(
input=input,
filter_shape=array_ops.shape(filter),
dilation_rate=rate,
padding=padding,
data_format=data_format,
op=op)
@tf_export("nn.depthwise_conv2d", v1=[])
def depthwise_conv2d_v2(input,
filter,
strides,
padding,
data_format=None,
dilations=None,
name=None):
"""Depthwise 2-D convolution.
Given a 4D input tensor ('NHWC' or 'NCHW' data formats)
and a filter tensor of shape
`[filter_height, filter_width, in_channels, channel_multiplier]`
containing `in_channels` convolutional filters of depth 1, `depthwise_conv2d`
applies a different filter to each input channel (expanding from 1 channel
to `channel_multiplier` channels for each), then concatenates the results
together. The output has `in_channels * channel_multiplier` channels.
In detail, with the default NHWC format,
output[b, i, j, k * channel_multiplier + q] = sum_{di, dj}
filter[di, dj, k, q] * input[b, strides[1] * i + rate[0] * di,
strides[2] * j + rate[1] * dj, k]
Must have `strides[0] = strides[3] = 1`. For the most common case of the
same horizontal and vertical strides, `strides = [1, stride, stride, 1]`.
If any value in `rate` is greater than 1, we perform atrous depthwise
convolution, in which case all values in the `strides` tensor must be equal
to 1.
Args:
input: 4-D with shape according to `data_format`.
filter: 4-D with shape
`[filter_height, filter_width, in_channels, channel_multiplier]`.
strides: 1-D of size 4. The stride of the sliding window for each
dimension of `input`.
padding: A string, either `'VALID'` or `'SAME'`. The padding algorithm.
See the "returns" section of `tf.nn.convolution` for details.
data_format: The data format for input. Either "NHWC" (default) or "NCHW".
dilations: 1-D of size 2. The dilation rate in which we sample input values
across the `height` and `width` dimensions in atrous convolution. If it is
greater than 1, then all values of strides must be 1.
name: A name for this operation (optional).
Returns:
A 4-D `Tensor` with shape according to `data_format`. E.g., for
"NHWC" format, shape is
`[batch, out_height, out_width, in_channels * channel_multiplier].`
"""
return depthwise_conv2d(input=input,
filter=filter,
strides=strides,
padding=padding,
rate=dilations,
name=name,
data_format=data_format)
# pylint: enable=redefined-builtin
# pylint: disable=redefined-builtin,line-too-long
@tf_export(v1=["nn.separable_conv2d"])
def separable_conv2d(input,
depthwise_filter,
pointwise_filter,
strides,
padding,
rate=None,
name=None,
data_format=None,
dilations=None):
"""2-D convolution with separable filters.
Performs a depthwise convolution that acts separately on channels followed by
a pointwise convolution that mixes channels. Note that this is separability
between dimensions `[1, 2]` and `3`, not spatial separability between
dimensions `1` and `2`.
In detail, with the default NHWC format,
output[b, i, j, k] = sum_{di, dj, q, r}
input[b, strides[1] * i + di, strides[2] * j + dj, q] *
depthwise_filter[di, dj, q, r] *
pointwise_filter[0, 0, q * channel_multiplier + r, k]
`strides` controls the strides for the depthwise convolution only, since
the pointwise convolution has implicit strides of `[1, 1, 1, 1]`. Must have
`strides[0] = strides[3] = 1`. For the most common case of the same
horizontal and vertical strides, `strides = [1, stride, stride, 1]`.
If any value in `rate` is greater than 1, we perform atrous depthwise
convolution, in which case all values in the `strides` tensor must be equal
to 1.
Args:
input: 4-D `Tensor` with shape according to `data_format`.
depthwise_filter: 4-D `Tensor` with shape
`[filter_height, filter_width, in_channels, channel_multiplier]`.
Contains `in_channels` convolutional filters of depth 1.
pointwise_filter: 4-D `Tensor` with shape
`[1, 1, channel_multiplier * in_channels, out_channels]`. Pointwise
filter to mix channels after `depthwise_filter` has convolved spatially.
strides: 1-D of size 4. The strides for the depthwise convolution for
each dimension of `input`.
padding: A string, either `'VALID'` or `'SAME'`. The padding algorithm.
See the "returns" section of `tf.nn.convolution` for details.
rate: 1-D of size 2. The dilation rate in which we sample input values
across the `height` and `width` dimensions in atrous convolution. If it is
greater than 1, then all values of strides must be 1.
name: A name for this operation (optional).
data_format: The data format for input. Either "NHWC" (default) or "NCHW".
dilations: Alias of rate.
Returns:
A 4-D `Tensor` with shape according to 'data_format'. For
example, with data_format="NHWC", shape is [batch, out_height,
out_width, out_channels].
"""
rate = deprecated_argument_lookup("dilations", dilations, "rate", rate)
with ops.name_scope(name, "separable_conv2d",
[input, depthwise_filter, pointwise_filter]) as name:
input = ops.convert_to_tensor(input, name="tensor_in")
depthwise_filter = ops.convert_to_tensor(
depthwise_filter, name="depthwise_filter")
pointwise_filter = ops.convert_to_tensor(
pointwise_filter, name="pointwise_filter")
pointwise_filter_shape = pointwise_filter.get_shape().with_rank(4)
pointwise_filter_shape.dims[0].assert_is_compatible_with(1)
pointwise_filter_shape.dims[1].assert_is_compatible_with(1)
if rate is None:
rate = [1, 1]
# The layout of the ops in the graph are expected to be as follows:
# depthwise_conv2d // Conv2D op corresponding to native deptwise conv.
# separable_conv2d // Conv2D op corresponding to the pointwise conv.
def op(input_converted, _, padding):
return nn_ops.depthwise_conv2d_native(
input=input_converted,
filter=depthwise_filter,
strides=strides,
padding=padding,
data_format=data_format,
name="depthwise")
depthwise = nn_ops.with_space_to_batch(
input=input,
filter_shape=array_ops.shape(depthwise_filter),
dilation_rate=rate,
padding=padding,
data_format=data_format,
op=op)
return nn_ops.conv2d(
depthwise,
pointwise_filter, [1, 1, 1, 1],
padding="VALID",
data_format=data_format,
name=name)
@tf_export("nn.separable_conv2d", v1=[])
def separable_conv2d_v2(
input,
depthwise_filter,
pointwise_filter,
strides,
padding,
data_format=None,
dilations=None,
name=None,
):
"""2-D convolution with separable filters.
Performs a depthwise convolution that acts separately on channels followed by
a pointwise convolution that mixes channels. Note that this is separability
between dimensions `[1, 2]` and `3`, not spatial separability between
dimensions `1` and `2`.
In detail, with the default NHWC format,
output[b, i, j, k] = sum_{di, dj, q, r}
input[b, strides[1] * i + di, strides[2] * j + dj, q] *
depthwise_filter[di, dj, q, r] *
pointwise_filter[0, 0, q * channel_multiplier + r, k]
`strides` controls the strides for the depthwise convolution only, since
the pointwise convolution has implicit strides of `[1, 1, 1, 1]`. Must have
`strides[0] = strides[3] = 1`. For the most common case of the same
horizontal and vertical strides, `strides = [1, stride, stride, 1]`.
If any value in `rate` is greater than 1, we perform atrous depthwise
convolution, in which case all values in the `strides` tensor must be equal
to 1.
Args:
input: 4-D `Tensor` with shape according to `data_format`.
depthwise_filter: 4-D `Tensor` with shape `[filter_height, filter_width,
in_channels, channel_multiplier]`. Contains `in_channels` convolutional
filters of depth 1.
pointwise_filter: 4-D `Tensor` with shape `[1, 1, channel_multiplier *
in_channels, out_channels]`. Pointwise filter to mix channels after
`depthwise_filter` has convolved spatially.
strides: 1-D of size 4. The strides for the depthwise convolution for each
dimension of `input`.
padding: A string, either `'VALID'` or `'SAME'`. The padding algorithm. See
the "returns" section of `tf.nn.convolution` for details.
data_format: The data format for input. Either "NHWC" (default) or "NCHW".
dilations: 1-D of size 2. The dilation rate in which we sample input values
across the `height` and `width` dimensions in atrous convolution. If it is
greater than 1, then all values of strides must be 1.
name: A name for this operation (optional).
Returns:
A 4-D `Tensor` with shape according to 'data_format'. For
example, with data_format="NHWC", shape is [batch, out_height,
out_width, out_channels].
"""
return separable_conv2d(
input,
depthwise_filter,
pointwise_filter,
strides,
padding,
rate=dilations,
name=name,
data_format=data_format)
# pylint: enable=redefined-builtin,line-too-long
@tf_export(v1=["nn.sufficient_statistics"])
def sufficient_statistics(x, axes, shift=None, keep_dims=None, name=None,
keepdims=None):
"""Calculate the sufficient statistics for the mean and variance of `x`.
These sufficient statistics are computed using the one pass algorithm on
an input that's optionally shifted. See:
https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance#Computing_shifted_data
Args:
x: A `Tensor`.
axes: Array of ints. Axes along which to compute mean and variance.
shift: A `Tensor` containing the value by which to shift the data for
numerical stability, or `None` if no shift is to be performed. A shift
close to the true mean provides the most numerically stable results.
keep_dims: produce statistics with the same dimensionality as the input.
name: Name used to scope the operations that compute the sufficient stats.
keepdims: Alias for keep_dims.
Returns:
Four `Tensor` objects of the same type as `x`:
* the count (number of elements to average over).
* the (possibly shifted) sum of the elements in the array.
* the (possibly shifted) sum of squares of the elements in the array.
* the shift by which the mean must be corrected or None if `shift` is None.
"""
axes = list(set(axes))
keep_dims = deprecated_argument_lookup(
"keepdims", keepdims, "keep_dims", keep_dims)
if keep_dims is None:
keep_dims = False
with ops.name_scope(name, "sufficient_statistics", [x, shift]):
x = ops.convert_to_tensor(x, name="x")
x_shape = x.get_shape()
if x_shape.rank is not None and all(
x_shape.dims[d].value is not None for d in axes):
counts = 1
for d in axes:
counts *= x_shape.dims[d].value
counts = constant_op.constant(counts, dtype=x.dtype)
else: # shape needs to be inferred at runtime.
x_dims = array_ops.gather(
math_ops.cast(array_ops.shape(x), x.dtype), axes)
counts = math_ops.reduce_prod(x_dims, name="count")
if shift is not None:
shift = ops.convert_to_tensor(shift, name="shift")
m_ss = math_ops.subtract(x, shift)
v_ss = math_ops.squared_difference(x, shift)
else: # no shift.
m_ss = x
v_ss = math_ops.square(x)
m_ss = math_ops.reduce_sum(m_ss, axes, keepdims=keep_dims, name="mean_ss")
v_ss = math_ops.reduce_sum(v_ss, axes, keepdims=keep_dims, name="var_ss")
return counts, m_ss, v_ss, shift
@tf_export("nn.sufficient_statistics", v1=[])
def sufficient_statistics_v2(x, axes, shift=None, keepdims=False, name=None):
"""Calculate the sufficient statistics for the mean and variance of `x`.
These sufficient statistics are computed using the one pass algorithm on
an input that's optionally shifted. See:
https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance#Computing_shifted_data
Args:
x: A `Tensor`.
axes: Array of ints. Axes along which to compute mean and variance.
shift: A `Tensor` containing the value by which to shift the data for
numerical stability, or `None` if no shift is to be performed. A shift
close to the true mean provides the most numerically stable results.
keepdims: produce statistics with the same dimensionality as the input.
name: Name used to scope the operations that compute the sufficient stats.
Returns:
Four `Tensor` objects of the same type as `x`:
* the count (number of elements to average over).
* the (possibly shifted) sum of the elements in the array.
* the (possibly shifted) sum of squares of the elements in the array.
* the shift by which the mean must be corrected or None if `shift` is None.
"""
return sufficient_statistics(
x=x, axes=axes, shift=shift, keep_dims=keepdims, name=name)
@tf_export("nn.normalize_moments")
def normalize_moments(counts, mean_ss, variance_ss, shift, name=None):
"""Calculate the mean and variance of based on the sufficient statistics.
Args:
counts: A `Tensor` containing the total count of the data (one value).
mean_ss: A `Tensor` containing the mean sufficient statistics: the (possibly
shifted) sum of the elements to average over.
variance_ss: A `Tensor` containing the variance sufficient statistics: the
(possibly shifted) squared sum of the data to compute the variance over.
shift: A `Tensor` containing the value by which the data is shifted for
numerical stability, or `None` if no shift was performed.
name: Name used to scope the operations that compute the moments.
Returns:
Two `Tensor` objects: `mean` and `variance`.
"""
with ops.name_scope(name, "normalize", [counts, mean_ss, variance_ss, shift]):
divisor = math_ops.reciprocal(counts, name="divisor")
if shift is not None:
shifted_mean = math_ops.multiply(mean_ss, divisor, name="shifted_mean")
mean = math_ops.add(shifted_mean, shift, name="mean")
else: # no shift.
shifted_mean = math_ops.multiply(mean_ss, divisor, name="mean")
mean = shifted_mean
variance = math_ops.subtract(
math_ops.multiply(variance_ss, divisor),
math_ops.square(shifted_mean),
name="variance")
return (mean, variance)
@tf_export(v1=["nn.moments"])
def moments(
x,
axes,
shift=None, # pylint: disable=unused-argument
name=None,
keep_dims=None,
keepdims=None):
"""Calculate the mean and variance of `x`.
The mean and variance are calculated by aggregating the contents of `x`
across `axes`. If `x` is 1-D and `axes = [0]` this is just the mean
and variance of a vector.
Note: shift is currently not used; the true mean is computed and used.
When using these moments for batch normalization (see
`tf.nn.batch_normalization`):
* for so-called "global normalization", used with convolutional filters with
shape `[batch, height, width, depth]`, pass `axes=[0, 1, 2]`.
* for simple batch normalization pass `axes=[0]` (batch only).
Args:
x: A `Tensor`.
axes: Array of ints. Axes along which to compute mean and
variance.
shift: Not used in the current implementation
name: Name used to scope the operations that compute the moments.
keep_dims: produce moments with the same dimensionality as the input.
keepdims: Alias to keep_dims.
Returns:
Two `Tensor` objects: `mean` and `variance`.
"""
keep_dims = deprecated_argument_lookup(
"keepdims", keepdims, "keep_dims", keep_dims)
if keep_dims is None:
keep_dims = False
with ops.name_scope(name, "moments", [x, axes]):
# The dynamic range of fp16 is too limited to support the collection of
# sufficient statistics. As a workaround we simply perform the operations
# on 32-bit floats before converting the mean and variance back to fp16
y = math_ops.cast(x, dtypes.float32) if x.dtype == dtypes.float16 else x
# Compute true mean while keeping the dims for proper broadcasting.
mean = math_ops.reduce_mean(y, axes, keepdims=True, name="mean")
# sample variance, not unbiased variance
# Note: stop_gradient does not change the gradient that gets
# backpropagated to the mean from the variance calculation,
# because that gradient is zero
variance = math_ops.reduce_mean(
math_ops.squared_difference(y, array_ops.stop_gradient(mean)),
axes,
keepdims=True,
name="variance")
if not keep_dims:
mean = array_ops.squeeze(mean, axes)
variance = array_ops.squeeze(variance, axes)
if x.dtype == dtypes.float16:
return (math_ops.cast(mean, dtypes.float16),
math_ops.cast(variance, dtypes.float16))
else:
return (mean, variance)
@tf_export("nn.moments", v1=[])
def moments_v2(
x,
axes,
shift=None,
keepdims=False,
name=None):
"""Calculates the mean and variance of `x`.
The mean and variance are calculated by aggregating the contents of `x`
across `axes`. If `x` is 1-D and `axes = [0]` this is just the mean
and variance of a vector.
Note: shift is currently not used; the true mean is computed and used.
When using these moments for batch normalization (see
`tf.nn.batch_normalization`):
* for so-called "global normalization", used with convolutional filters with
shape `[batch, height, width, depth]`, pass `axes=[0, 1, 2]`.
* for simple batch normalization pass `axes=[0]` (batch only).
Args:
x: A `Tensor`.
axes: Array of ints. Axes along which to compute mean and
variance.
shift: Not used in the current implementation.
keepdims: produce moments with the same dimensionality as the input.
name: Name used to scope the operations that compute the moments.
Returns:
Two `Tensor` objects: `mean` and `variance`.
"""
return moments(x=x, axes=axes, shift=shift, name=name, keep_dims=keepdims)
@tf_export(v1=["nn.weighted_moments"])
def weighted_moments(x, axes, frequency_weights, name=None, keep_dims=None,
keepdims=None):
"""Returns the frequency-weighted mean and variance of `x`.
Args:
x: A tensor.
axes: 1-d tensor of int32 values; these are the axes along which
to compute mean and variance.
frequency_weights: A tensor of positive weights which can be
broadcast with x.
name: Name used to scope the operation.
keep_dims: Produce moments with the same dimensionality as the input.
keepdims: Alias of keep_dims.
Returns:
Two tensors: `weighted_mean` and `weighted_variance`.
"""
keep_dims = deprecated_argument_lookup(
"keepdims", keepdims, "keep_dims", keep_dims)
if keep_dims is None:
keep_dims = False
with ops.name_scope(name, "weighted_moments", [x, frequency_weights, axes]):
x = ops.convert_to_tensor(x, name="x")
frequency_weights = ops.convert_to_tensor(
frequency_weights, name="frequency_weights")
# Unlike moments(), this just uses a simpler two-pass method.
# See comment in moments() WRT precision; it applies here too.
needs_cast = x.dtype == dtypes.float16
if needs_cast:
x = math_ops.cast(x, dtypes.float32)
if frequency_weights.dtype != x.dtype:
frequency_weights = math_ops.cast(frequency_weights, x.dtype)
# Note that we use keep_dims=True for our reductions regardless of the arg;
# this is so that the results remain broadcast-compatible with the inputs.
weighted_input_sum = math_ops.reduce_sum(
frequency_weights * x, axes, name="weighted_input_sum", keepdims=True)
# The shape of the weights isn't necessarily the same as x's
# shape, just broadcast-compatible with it -- so this expression
# performs broadcasting to give a per-item weight, with the same
# shape as (freqency_weights * x). This avoids having to reason
# through all the broadcast logic to compute a correct
# sum_of_weights.
broadcasted_weights = frequency_weights + array_ops.zeros_like(x)
sum_of_weights = math_ops.reduce_sum(
broadcasted_weights, axes, name="sum_of_weights", keepdims=True)
divisor = math_ops.reciprocal(sum_of_weights, name="inv_weight_sum")
weighted_mean = math_ops.multiply(weighted_input_sum, divisor)
# Have the weighted mean; now on to variance:
weighted_distsq = math_ops.reduce_sum(
frequency_weights * math_ops.squared_difference(x, weighted_mean),
axes,
name="weighted_distsq",
keepdims=True)
weighted_variance = math_ops.multiply(weighted_distsq, divisor)
if not keep_dims:
weighted_mean = array_ops.squeeze(weighted_mean, axis=axes)
weighted_variance = array_ops.squeeze(
weighted_variance, axis=axes)
if needs_cast:
weighted_mean = math_ops.cast(weighted_mean, dtypes.float16)
weighted_variance = math_ops.cast(weighted_variance, dtypes.float16)
return weighted_mean, weighted_variance
@tf_export("nn.weighted_moments", v1=[])
def weighted_moments_v2(x, axes, frequency_weights, keepdims=False, name=None):
"""Returns the frequency-weighted mean and variance of `x`.
Args:
x: A tensor.
axes: 1-d tensor of int32 values; these are the axes along which
to compute mean and variance.
frequency_weights: A tensor of positive weights which can be
broadcast with x.
keepdims: Produce moments with the same dimensionality as the input.
name: Name used to scope the operation.
Returns:
Two tensors: `weighted_mean` and `weighted_variance`.
"""
return weighted_moments(
x=x,
axes=axes,
frequency_weights=frequency_weights,
name=name,
keep_dims=keepdims)
@tf_export("nn.batch_normalization")
def batch_normalization(x,
mean,
variance,
offset,
scale,
variance_epsilon,
name=None):
r"""Batch normalization.
Normalizes a tensor by `mean` and `variance`, and applies (optionally) a
`scale` \\(\gamma\\) to it, as well as an `offset` \\(\beta\\):
\\(\frac{\gamma(x-\mu)}{\sigma}+\beta\\)
`mean`, `variance`, `offset` and `scale` are all expected to be of one of two
shapes:
* In all generality, they can have the same number of dimensions as the
input `x`, with identical sizes as `x` for the dimensions that are not
normalized over (the 'depth' dimension(s)), and dimension 1 for the
others which are being normalized over.
`mean` and `variance` in this case would typically be the outputs of
`tf.nn.moments(..., keep_dims=True)` during training, or running averages
thereof during inference.
* In the common case where the 'depth' dimension is the last dimension in
the input tensor `x`, they may be one dimensional tensors of the same
size as the 'depth' dimension.
This is the case for example for the common `[batch, depth]` layout of
fully-connected layers, and `[batch, height, width, depth]` for
convolutions.
`mean` and `variance` in this case would typically be the outputs of
`tf.nn.moments(..., keep_dims=False)` during training, or running averages
thereof during inference.
See Source: [Batch Normalization: Accelerating Deep Network Training by
Reducing Internal Covariate Shift; S. Ioffe, C. Szegedy]
(http://arxiv.org/abs/1502.03167).
Args:
x: Input `Tensor` of arbitrary dimensionality.
mean: A mean `Tensor`.
variance: A variance `Tensor`.
offset: An offset `Tensor`, often denoted \\(\beta\\) in equations, or
None. If present, will be added to the normalized tensor.
scale: A scale `Tensor`, often denoted \\(\gamma\\) in equations, or
`None`. If present, the scale is applied to the normalized tensor.
variance_epsilon: A small float number to avoid dividing by 0.
name: A name for this operation (optional).
Returns:
the normalized, scaled, offset tensor.
"""
with ops.name_scope(name, "batchnorm", [x, mean, variance, scale, offset]):
inv = math_ops.rsqrt(variance + variance_epsilon)
if scale is not None:
inv *= scale
# Note: tensorflow/contrib/quantize/python/fold_batch_norms.py depends on
# the precise order of ops that are generated by the expression below.
return x * math_ops.cast(inv, x.dtype) + math_ops.cast(
offset - mean * inv if offset is not None else -mean * inv, x.dtype)
@tf_export(v1=["nn.fused_batch_norm"])
def fused_batch_norm(
x,
scale,
offset, # pylint: disable=invalid-name
mean=None,
variance=None,
epsilon=0.001,
data_format="NHWC",
is_training=True,
name=None):
r"""Batch normalization.
See Source: [Batch Normalization: Accelerating Deep Network Training by
Reducing Internal Covariate Shift; S. Ioffe, C. Szegedy]
(http://arxiv.org/abs/1502.03167).
Args:
x: Input `Tensor` of 4 dimensions.
scale: A `Tensor` of 1 dimension for scaling.
offset: A `Tensor` of 1 dimension for bias.
mean: A `Tensor` of 1 dimension for population mean used for inference.
variance: A `Tensor` of 1 dimension for population variance
used for inference.
epsilon: A small float number added to the variance of x.
data_format: The data format for x. Either "NHWC" (default) or "NCHW".
is_training: A bool value to specify if the operation is used for
training or inference.
name: A name for this operation (optional).
Returns:
y: A 4D Tensor for the normalized, scaled, offsetted x.
batch_mean: A 1D Tensor for the mean of x.
batch_var: A 1D Tensor for the variance of x.
Raises:
ValueError: If mean or variance is not None when is_training is True.
"""
x = ops.convert_to_tensor(x, name="input")
scale = ops.convert_to_tensor(scale, name="scale")
offset = ops.convert_to_tensor(offset, name="offset")
if is_training:
if (mean is not None) or (variance is not None):
raise ValueError("Both 'mean' and 'variance' must be None "
"if is_training is True.")
if mean is None:
mean = constant_op.constant([])
if variance is None:
variance = constant_op.constant([])
# Set a minimum epsilon to 1.001e-5, which is a requirement by CUDNN to
# prevent exception (see cudnn.h).
min_epsilon = 1.001e-5
epsilon = epsilon if epsilon > min_epsilon else min_epsilon
if compat.forward_compatible(2019, 6, 6):
y, batch_mean, batch_var, _, _, _ = gen_nn_ops.fused_batch_norm_v3(
x,
scale,
offset,
mean,
variance,
epsilon=epsilon,
data_format=data_format,
is_training=is_training,
name=name)
return y, batch_mean, batch_var
if x.dtype == dtypes.float16 or x.dtype == dtypes.bfloat16:
fused_batch_norm_func = gen_nn_ops.fused_batch_norm_v2
else:
fused_batch_norm_func = gen_nn_ops._fused_batch_norm # pylint: disable=protected-access
y, batch_mean, batch_var, _, _ = fused_batch_norm_func(
x,
scale,
offset,
mean,
variance,
epsilon=epsilon,
data_format=data_format,
is_training=is_training,
name=name)
return y, batch_mean, batch_var
@tf_export(v1=["nn.batch_norm_with_global_normalization"])
def batch_norm_with_global_normalization(t=None,
m=None,
v=None,
beta=None,
gamma=None,
variance_epsilon=None,
scale_after_normalization=None,
name=None,
input=None, # pylint: disable=redefined-builtin
mean=None,
variance=None):
"""Batch normalization.
This op is deprecated. See `tf.nn.batch_normalization`.
Args:
t: A 4D input Tensor.
m: A 1D mean Tensor with size matching the last dimension of t.
This is the first output from tf.nn.moments,
or a saved moving average thereof.
v: A 1D variance Tensor with size matching the last dimension of t.
This is the second output from tf.nn.moments,
or a saved moving average thereof.
beta: A 1D beta Tensor with size matching the last dimension of t.
An offset to be added to the normalized tensor.
gamma: A 1D gamma Tensor with size matching the last dimension of t.
If "scale_after_normalization" is true, this tensor will be multiplied
with the normalized tensor.
variance_epsilon: A small float number to avoid dividing by 0.
scale_after_normalization: A bool indicating whether the resulted tensor
needs to be multiplied with gamma.
name: A name for this operation (optional).
input: Alias for t.
mean: Alias for m.
variance: Alias for v.
Returns:
A batch-normalized `t`.
"""
t = deprecated_argument_lookup("input", input, "t", t)
m = deprecated_argument_lookup("mean", mean, "m", m)
v = deprecated_argument_lookup("variance", variance, "v", v)
return batch_normalization(t, m, v, beta, gamma if scale_after_normalization
else None, variance_epsilon, name)
# pylint: disable=redefined-builtin,line-too-long
@tf_export("nn.batch_norm_with_global_normalization", v1=[])
def batch_norm_with_global_normalization_v2(input,
mean,
variance,
beta,
gamma,
variance_epsilon,
scale_after_normalization,
name=None):
"""Batch normalization.
This op is deprecated. See `tf.nn.batch_normalization`.
Args:
input: A 4D input Tensor.
mean: A 1D mean Tensor with size matching the last dimension of t.
This is the first output from tf.nn.moments,
or a saved moving average thereof.
variance: A 1D variance Tensor with size matching the last dimension of t.
This is the second output from tf.nn.moments,
or a saved moving average thereof.
beta: A 1D beta Tensor with size matching the last dimension of t.
An offset to be added to the normalized tensor.
gamma: A 1D gamma Tensor with size matching the last dimension of t.
If "scale_after_normalization" is true, this tensor will be multiplied
with the normalized tensor.
variance_epsilon: A small float number to avoid dividing by 0.
scale_after_normalization: A bool indicating whether the resulted tensor
needs to be multiplied with gamma.
name: A name for this operation (optional).
Returns:
A batch-normalized `t`.
"""
return batch_norm_with_global_normalization(t=input,
m=mean,
v=variance,
beta=beta,
gamma=gamma,
variance_epsilon=variance_epsilon,
scale_after_normalization=scale_after_normalization,
name=name)
# pylint: enable=redefined-builtin,line-too-long
def _sum_rows(x):
"""Returns a vector summing up each row of the matrix x."""
# _sum_rows(x) is equivalent to math_ops.reduce_sum(x, 1) when x is
# a matrix. The gradient of _sum_rows(x) is more efficient than
# reduce_sum(x, 1)'s gradient in today's implementation. Therefore,
# we use _sum_rows(x) in the nce_loss() computation since the loss
# is mostly used for training.
cols = array_ops.shape(x)[1]
ones_shape = array_ops.stack([cols, 1])
ones = array_ops.ones(ones_shape, x.dtype)
return array_ops.reshape(math_ops.matmul(x, ones), [-1])
def _compute_sampled_logits(weights,
biases,
labels,
inputs,
num_sampled,
num_classes,
num_true=1,
sampled_values=None,
subtract_log_q=True,
remove_accidental_hits=False,
partition_strategy="mod",
name=None,
seed=None):
"""Helper function for nce_loss and sampled_softmax_loss functions.
Computes sampled output training logits and labels suitable for implementing
e.g. noise-contrastive estimation (see nce_loss) or sampled softmax (see
sampled_softmax_loss).
Note: In the case where num_true > 1, we assign to each target class
the target probability 1 / num_true so that the target probabilities
sum to 1 per-example.
Args:
weights: A `Tensor` of shape `[num_classes, dim]`, or a list of `Tensor`
objects whose concatenation along dimension 0 has shape
`[num_classes, dim]`. The (possibly-partitioned) class embeddings.
biases: A `Tensor` of shape `[num_classes]`. The (possibly-partitioned)
class biases.
labels: A `Tensor` of type `int64` and shape `[batch_size,
num_true]`. The target classes. Note that this format differs from
the `labels` argument of `nn.softmax_cross_entropy_with_logits`.
inputs: A `Tensor` of shape `[batch_size, dim]`. The forward
activations of the input network.
num_sampled: An `int`. The number of classes to randomly sample per batch.
num_classes: An `int`. The number of possible classes.
num_true: An `int`. The number of target classes per training example.
sampled_values: a tuple of (`sampled_candidates`, `true_expected_count`,
`sampled_expected_count`) returned by a `*_candidate_sampler` function.
(if None, we default to `log_uniform_candidate_sampler`)
subtract_log_q: A `bool`. whether to subtract the log expected count of
the labels in the sample to get the logits of the true labels.
Default is True. Turn off for Negative Sampling.
remove_accidental_hits: A `bool`. whether to remove "accidental hits"
where a sampled class equals one of the target classes. Default is
False.
partition_strategy: A string specifying the partitioning strategy, relevant
if `len(weights) > 1`. Currently `"div"` and `"mod"` are supported.
Default is `"mod"`. See `tf.nn.embedding_lookup` for more details.
name: A name for the operation (optional).
seed: random seed for candidate sampling. Default to None, which doesn't set
the op-level random seed for candidate sampling.
Returns:
out_logits: `Tensor` object with shape
`[batch_size, num_true + num_sampled]`, for passing to either
`nn.sigmoid_cross_entropy_with_logits` (NCE) or
`nn.softmax_cross_entropy_with_logits` (sampled softmax).
out_labels: A Tensor object with the same shape as `out_logits`.
"""
if isinstance(weights, variables.PartitionedVariable):
weights = list(weights)
if not isinstance(weights, list):
weights = [weights]
with ops.name_scope(name, "compute_sampled_logits",
weights + [biases, inputs, labels]):
if labels.dtype != dtypes.int64:
labels = math_ops.cast(labels, dtypes.int64)
labels_flat = array_ops.reshape(labels, [-1])
# Sample the negative labels.
# sampled shape: [num_sampled] tensor
# true_expected_count shape = [batch_size, 1] tensor
# sampled_expected_count shape = [num_sampled] tensor
if sampled_values is None:
sampled_values = candidate_sampling_ops.log_uniform_candidate_sampler(
true_classes=labels,
num_true=num_true,
num_sampled=num_sampled,
unique=True,
range_max=num_classes,
seed=seed)
# NOTE: pylint cannot tell that 'sampled_values' is a sequence
# pylint: disable=unpacking-non-sequence
sampled, true_expected_count, sampled_expected_count = (
array_ops.stop_gradient(s) for s in sampled_values)
# pylint: enable=unpacking-non-sequence
sampled = math_ops.cast(sampled, dtypes.int64)
# labels_flat is a [batch_size * num_true] tensor
# sampled is a [num_sampled] int tensor
all_ids = array_ops.concat([labels_flat, sampled], 0)
# Retrieve the true weights and the logits of the sampled weights.
# weights shape is [num_classes, dim]
all_w = embedding_ops.embedding_lookup(
weights, all_ids, partition_strategy=partition_strategy)
if all_w.dtype != inputs.dtype:
all_w = math_ops.cast(all_w, inputs.dtype)
# true_w shape is [batch_size * num_true, dim]
true_w = array_ops.slice(all_w, [0, 0],
array_ops.stack(
[array_ops.shape(labels_flat)[0], -1]))
sampled_w = array_ops.slice(
all_w, array_ops.stack([array_ops.shape(labels_flat)[0], 0]), [-1, -1])
# inputs has shape [batch_size, dim]
# sampled_w has shape [num_sampled, dim]
# Apply X*W', which yields [batch_size, num_sampled]
sampled_logits = math_ops.matmul(inputs, sampled_w, transpose_b=True)
# Retrieve the true and sampled biases, compute the true logits, and
# add the biases to the true and sampled logits.
all_b = embedding_ops.embedding_lookup(
biases, all_ids, partition_strategy=partition_strategy)
if all_b.dtype != inputs.dtype:
all_b = math_ops.cast(all_b, inputs.dtype)
# true_b is a [batch_size * num_true] tensor
# sampled_b is a [num_sampled] float tensor
true_b = array_ops.slice(all_b, [0], array_ops.shape(labels_flat))
sampled_b = array_ops.slice(all_b, array_ops.shape(labels_flat), [-1])
# inputs shape is [batch_size, dim]
# true_w shape is [batch_size * num_true, dim]
# row_wise_dots is [batch_size, num_true, dim]
dim = array_ops.shape(true_w)[1:2]
new_true_w_shape = array_ops.concat([[-1, num_true], dim], 0)
row_wise_dots = math_ops.multiply(
array_ops.expand_dims(inputs, 1),
array_ops.reshape(true_w, new_true_w_shape))
# We want the row-wise dot plus biases which yields a
# [batch_size, num_true] tensor of true_logits.
dots_as_matrix = array_ops.reshape(row_wise_dots,
array_ops.concat([[-1], dim], 0))
true_logits = array_ops.reshape(_sum_rows(dots_as_matrix), [-1, num_true])
true_b = array_ops.reshape(true_b, [-1, num_true])
true_logits += true_b
sampled_logits += sampled_b
if remove_accidental_hits:
acc_hits = candidate_sampling_ops.compute_accidental_hits(
labels, sampled, num_true=num_true)
acc_indices, acc_ids, acc_weights = acc_hits
# This is how SparseToDense expects the indices.
acc_indices_2d = array_ops.reshape(acc_indices, [-1, 1])
acc_ids_2d_int32 = array_ops.reshape(
math_ops.cast(acc_ids, dtypes.int32), [-1, 1])
sparse_indices = array_ops.concat([acc_indices_2d, acc_ids_2d_int32], 1,
"sparse_indices")
# Create sampled_logits_shape = [batch_size, num_sampled]
sampled_logits_shape = array_ops.concat(
[array_ops.shape(labels)[:1],
array_ops.expand_dims(num_sampled, 0)], 0)
if sampled_logits.dtype != acc_weights.dtype:
acc_weights = math_ops.cast(acc_weights, sampled_logits.dtype)
sampled_logits += gen_sparse_ops.sparse_to_dense(
sparse_indices,
sampled_logits_shape,
acc_weights,
default_value=0.0,
validate_indices=False)
if subtract_log_q:
# Subtract log of Q(l), prior probability that l appears in sampled.
true_logits -= math_ops.log(true_expected_count)
sampled_logits -= math_ops.log(sampled_expected_count)
# Construct output logits and labels. The true labels/logits start at col 0.
out_logits = array_ops.concat([true_logits, sampled_logits], 1)
# true_logits is a float tensor, ones_like(true_logits) is a float
# tensor of ones. We then divide by num_true to ensure the per-example
# labels sum to 1.0, i.e. form a proper probability distribution.
out_labels = array_ops.concat([
array_ops.ones_like(true_logits) / num_true,
array_ops.zeros_like(sampled_logits)
], 1)
return out_logits, out_labels
@tf_export("nn.nce_loss", v1=[])
def nce_loss_v2(weights,
biases,
labels,
inputs,
num_sampled,
num_classes,
num_true=1,
sampled_values=None,
remove_accidental_hits=False,
name="nce_loss"):
"""Computes and returns the noise-contrastive estimation training loss.
See [Noise-contrastive estimation: A new estimation principle for
unnormalized statistical
models](http://www.jmlr.org/proceedings/papers/v9/gutmann10a/gutmann10a.pdf).
Also see our [Candidate Sampling Algorithms
Reference](https://www.tensorflow.org/extras/candidate_sampling.pdf)
A common use case is to use this method for training, and calculate the full
sigmoid loss for evaluation or inference as in the following example:
```python
if mode == "train":
loss = tf.nn.nce_loss(
weights=weights,
biases=biases,
labels=labels,
inputs=inputs,
...)
elif mode == "eval":
logits = tf.matmul(inputs, tf.transpose(weights))
logits = tf.nn.bias_add(logits, biases)
labels_one_hot = tf.one_hot(labels, n_classes)
loss = tf.nn.sigmoid_cross_entropy_with_logits(
labels=labels_one_hot,
logits=logits)
loss = tf.reduce_sum(loss, axis=1)
```
Note: when doing embedding lookup on `weights` and `bias`, "div" partition
strategy will be used. Support for other partition strategy will be added
later.
Note: By default this uses a log-uniform (Zipfian) distribution for sampling,
so your labels must be sorted in order of decreasing frequency to achieve
good results. For more details, see
`tf.random.log_uniform_candidate_sampler`.
Note: In the case where `num_true` > 1, we assign to each target class
the target probability 1 / `num_true` so that the target probabilities
sum to 1 per-example.
Note: It would be useful to allow a variable number of target classes per
example. We hope to provide this functionality in a future release.
For now, if you have a variable number of target classes, you can pad them
out to a constant number by either repeating them or by padding
with an otherwise unused class.
Args:
weights: A `Tensor` of shape `[num_classes, dim]`, or a list of `Tensor`
objects whose concatenation along dimension 0 has shape [num_classes,
dim]. The (possibly-partitioned) class embeddings.
biases: A `Tensor` of shape `[num_classes]`. The class biases.
labels: A `Tensor` of type `int64` and shape `[batch_size, num_true]`. The
target classes.
inputs: A `Tensor` of shape `[batch_size, dim]`. The forward activations of
the input network.
num_sampled: An `int`. The number of negative classes to randomly sample
per batch. This single sample of negative classes is evaluated for each
element in the batch.
num_classes: An `int`. The number of possible classes.
num_true: An `int`. The number of target classes per training example.
sampled_values: a tuple of (`sampled_candidates`, `true_expected_count`,
`sampled_expected_count`) returned by a `*_candidate_sampler` function.
(if None, we default to `log_uniform_candidate_sampler`)
remove_accidental_hits: A `bool`. Whether to remove "accidental hits"
where a sampled class equals one of the target classes. If set to `True`,
this is a "Sampled Logistic" loss instead of NCE, and we are learning to
generate log-odds instead of log probabilities. See our [Candidate
Sampling Algorithms Reference]
(https://www.tensorflow.org/extras/candidate_sampling.pdf). Default is
False.
name: A name for the operation (optional).
Returns:
A `batch_size` 1-D tensor of per-example NCE losses.
"""
# TODO(yuefengz): get partition_strategy from either variables or distribution
# strategies.
return nce_loss(
weights,
biases,
labels,
inputs,
num_sampled,
num_classes,
num_true=num_true,
sampled_values=sampled_values,
remove_accidental_hits=remove_accidental_hits,
partition_strategy="div",
name=name)
@tf_export(v1=["nn.nce_loss"])
def nce_loss(weights,
biases,
labels,
inputs,
num_sampled,
num_classes,
num_true=1,
sampled_values=None,
remove_accidental_hits=False,
partition_strategy="mod",
name="nce_loss"):
"""Computes and returns the noise-contrastive estimation training loss.
See [Noise-contrastive estimation: A new estimation principle for
unnormalized statistical
models](http://www.jmlr.org/proceedings/papers/v9/gutmann10a/gutmann10a.pdf).
Also see our [Candidate Sampling Algorithms
Reference](https://www.tensorflow.org/extras/candidate_sampling.pdf)
A common use case is to use this method for training, and calculate the full
sigmoid loss for evaluation or inference. In this case, you must set
`partition_strategy="div"` for the two losses to be consistent, as in the
following example:
```python
if mode == "train":
loss = tf.nn.nce_loss(
weights=weights,
biases=biases,
labels=labels,
inputs=inputs,
...,
partition_strategy="div")
elif mode == "eval":
logits = tf.matmul(inputs, tf.transpose(weights))
logits = tf.nn.bias_add(logits, biases)
labels_one_hot = tf.one_hot(labels, n_classes)
loss = tf.nn.sigmoid_cross_entropy_with_logits(
labels=labels_one_hot,
logits=logits)
loss = tf.reduce_sum(loss, axis=1)
```
Note: By default this uses a log-uniform (Zipfian) distribution for sampling,
so your labels must be sorted in order of decreasing frequency to achieve
good results. For more details, see
`tf.random.log_uniform_candidate_sampler`.
Note: In the case where `num_true` > 1, we assign to each target class
the target probability 1 / `num_true` so that the target probabilities
sum to 1 per-example.
Note: It would be useful to allow a variable number of target classes per
example. We hope to provide this functionality in a future release.
For now, if you have a variable number of target classes, you can pad them
out to a constant number by either repeating them or by padding
with an otherwise unused class.
Args:
weights: A `Tensor` of shape `[num_classes, dim]`, or a list of `Tensor`
objects whose concatenation along dimension 0 has shape
[num_classes, dim]. The (possibly-partitioned) class embeddings.
biases: A `Tensor` of shape `[num_classes]`. The class biases.
labels: A `Tensor` of type `int64` and shape `[batch_size,
num_true]`. The target classes.
inputs: A `Tensor` of shape `[batch_size, dim]`. The forward
activations of the input network.
num_sampled: An `int`. The number of negative classes to randomly sample
per batch. This single sample of negative classes is evaluated for each
element in the batch.
num_classes: An `int`. The number of possible classes.
num_true: An `int`. The number of target classes per training example.
sampled_values: a tuple of (`sampled_candidates`, `true_expected_count`,
`sampled_expected_count`) returned by a `*_candidate_sampler` function.
(if None, we default to `log_uniform_candidate_sampler`)
remove_accidental_hits: A `bool`. Whether to remove "accidental hits"
where a sampled class equals one of the target classes. If set to
`True`, this is a "Sampled Logistic" loss instead of NCE, and we are
learning to generate log-odds instead of log probabilities. See
our [Candidate Sampling Algorithms Reference]
(https://www.tensorflow.org/extras/candidate_sampling.pdf).
Default is False.
partition_strategy: A string specifying the partitioning strategy, relevant
if `len(weights) > 1`. Currently `"div"` and `"mod"` are supported.
Default is `"mod"`. See `tf.nn.embedding_lookup` for more details.
name: A name for the operation (optional).
Returns:
A `batch_size` 1-D tensor of per-example NCE losses.
"""
logits, labels = _compute_sampled_logits(
weights=weights,
biases=biases,
labels=labels,
inputs=inputs,
num_sampled=num_sampled,
num_classes=num_classes,
num_true=num_true,
sampled_values=sampled_values,
subtract_log_q=True,
remove_accidental_hits=remove_accidental_hits,
partition_strategy=partition_strategy,
name=name)
sampled_losses = sigmoid_cross_entropy_with_logits(
labels=labels, logits=logits, name="sampled_losses")
# sampled_losses is batch_size x {true_loss, sampled_losses...}
# We sum out true and sampled losses.
return _sum_rows(sampled_losses)
@tf_export("nn.sampled_softmax_loss", v1=[])
def sampled_softmax_loss_v2(weights,
biases,
labels,
inputs,
num_sampled,
num_classes,
num_true=1,
sampled_values=None,
remove_accidental_hits=True,
seed=None,
name="sampled_softmax_loss"):
"""Computes and returns the sampled softmax training loss.
This is a faster way to train a softmax classifier over a huge number of
classes.
This operation is for training only. It is generally an underestimate of
the full softmax loss.
A common use case is to use this method for training, and calculate the full
sigmoid loss for evaluation or inference as in the following example:
```python
if mode == "train":
loss = tf.nn.sampled_softmax_loss(
weights=weights,
biases=biases,
labels=labels,
inputs=inputs,
...)
elif mode == "eval":
logits = tf.matmul(inputs, tf.transpose(weights))
logits = tf.nn.bias_add(logits, biases)
labels_one_hot = tf.one_hot(labels, n_classes)
loss = tf.nn.softmax_cross_entropy_with_logits(
labels=labels_one_hot,
logits=logits)
```
See our [Candidate Sampling Algorithms Reference]
(https://www.tensorflow.org/extras/candidate_sampling.pdf)
Also see Section 3 of [Jean et al., 2014](http://arxiv.org/abs/1412.2007)
([pdf](http://arxiv.org/pdf/1412.2007.pdf)) for the math.
Note: when doing embedding lookup on `weights` and `bias`, "div" partition
strategy will be used. Support for other partition strategy will be added
later.
Args:
weights: A `Tensor` of shape `[num_classes, dim]`, or a list of `Tensor`
objects whose concatenation along dimension 0 has shape [num_classes,
dim]. The (possibly-sharded) class embeddings.
biases: A `Tensor` of shape `[num_classes]`. The class biases.
labels: A `Tensor` of type `int64` and shape `[batch_size, num_true]`. The
target classes. Note that this format differs from the `labels` argument
of `nn.softmax_cross_entropy_with_logits`.
inputs: A `Tensor` of shape `[batch_size, dim]`. The forward activations of
the input network.
num_sampled: An `int`. The number of classes to randomly sample per batch.
num_classes: An `int`. The number of possible classes.
num_true: An `int`. The number of target classes per training example.
sampled_values: a tuple of (`sampled_candidates`, `true_expected_count`,
`sampled_expected_count`) returned by a `*_candidate_sampler` function.
(if None, we default to `log_uniform_candidate_sampler`)
remove_accidental_hits: A `bool`. whether to remove "accidental hits"
where a sampled class equals one of the target classes. Default is True.
seed: random seed for candidate sampling. Default to None, which doesn't set
the op-level random seed for candidate sampling.
name: A name for the operation (optional).
Returns:
A `batch_size` 1-D tensor of per-example sampled softmax losses.
"""
return sampled_softmax_loss(
weights,
biases,
labels,
inputs,
num_sampled,
num_classes,
num_true=num_true,
sampled_values=sampled_values,
remove_accidental_hits=remove_accidental_hits,
partition_strategy="div",
name=name,
seed=seed)
@tf_export(v1=["nn.sampled_softmax_loss"])
def sampled_softmax_loss(weights,
biases,
labels,
inputs,
num_sampled,
num_classes,
num_true=1,
sampled_values=None,
remove_accidental_hits=True,
partition_strategy="mod",
name="sampled_softmax_loss",
seed=None):
"""Computes and returns the sampled softmax training loss.
This is a faster way to train a softmax classifier over a huge number of
classes.
This operation is for training only. It is generally an underestimate of
the full softmax loss.
A common use case is to use this method for training, and calculate the full
softmax loss for evaluation or inference. In this case, you must set
`partition_strategy="div"` for the two losses to be consistent, as in the
following example:
```python
if mode == "train":
loss = tf.nn.sampled_softmax_loss(
weights=weights,
biases=biases,
labels=labels,
inputs=inputs,
...,
partition_strategy="div")
elif mode == "eval":
logits = tf.matmul(inputs, tf.transpose(weights))
logits = tf.nn.bias_add(logits, biases)
labels_one_hot = tf.one_hot(labels, n_classes)
loss = tf.nn.softmax_cross_entropy_with_logits(
labels=labels_one_hot,
logits=logits)
```
See our [Candidate Sampling Algorithms Reference]
(https://www.tensorflow.org/extras/candidate_sampling.pdf)
Also see Section 3 of [Jean et al., 2014](http://arxiv.org/abs/1412.2007)
([pdf](http://arxiv.org/pdf/1412.2007.pdf)) for the math.
Args:
weights: A `Tensor` of shape `[num_classes, dim]`, or a list of `Tensor`
objects whose concatenation along dimension 0 has shape
[num_classes, dim]. The (possibly-sharded) class embeddings.
biases: A `Tensor` of shape `[num_classes]`. The class biases.
labels: A `Tensor` of type `int64` and shape `[batch_size,
num_true]`. The target classes. Note that this format differs from
the `labels` argument of `nn.softmax_cross_entropy_with_logits`.
inputs: A `Tensor` of shape `[batch_size, dim]`. The forward
activations of the input network.
num_sampled: An `int`. The number of classes to randomly sample per batch.
num_classes: An `int`. The number of possible classes.
num_true: An `int`. The number of target classes per training example.
sampled_values: a tuple of (`sampled_candidates`, `true_expected_count`,
`sampled_expected_count`) returned by a `*_candidate_sampler` function.
(if None, we default to `log_uniform_candidate_sampler`)
remove_accidental_hits: A `bool`. whether to remove "accidental hits"
where a sampled class equals one of the target classes. Default is
True.
partition_strategy: A string specifying the partitioning strategy, relevant
if `len(weights) > 1`. Currently `"div"` and `"mod"` are supported.
Default is `"mod"`. See `tf.nn.embedding_lookup` for more details.
name: A name for the operation (optional).
seed: random seed for candidate sampling. Default to None, which doesn't set
the op-level random seed for candidate sampling.
Returns:
A `batch_size` 1-D tensor of per-example sampled softmax losses.
"""
logits, labels = _compute_sampled_logits(
weights=weights,
biases=biases,
labels=labels,
inputs=inputs,
num_sampled=num_sampled,
num_classes=num_classes,
num_true=num_true,
sampled_values=sampled_values,
subtract_log_q=True,
remove_accidental_hits=remove_accidental_hits,
partition_strategy=partition_strategy,
name=name,
seed=seed)
labels = array_ops.stop_gradient(labels, name="labels_stop_gradient")
sampled_losses = nn_ops.softmax_cross_entropy_with_logits_v2(
labels=labels, logits=logits)
# sampled_losses is a [batch_size] tensor.
return sampled_losses
|
tensorflow-master
|
tensorflow/python/ops/nn_impl.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for matmul_benchmark.py."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import itertools
import numpy as np
from tensorflow.core.framework import graph_pb2
from tensorflow.core.framework import node_def_pb2
from tensorflow.python.framework import ops
from tensorflow.python.ops import matmul_benchmark
from tensorflow.python.platform import test as googletest
from tensorflow.python.platform import tf_logging
def BuildGraphTest(n, m, k, transpose_a, transpose_b, dtype):
def Test(self):
if not googletest.is_gpu_available():
tf_logging.info("Skipping BuildGraphTest %s",
(n, m, k, transpose_a, transpose_b))
return
tf_logging.info("Testing BuildGraphTest %s",
(n, m, k, transpose_a, transpose_b))
self._VerifyBuildGraph(n, m, k, transpose_a, transpose_b, dtype)
return Test
def RunGraphTest(n, m, k, transpose_a, transpose_b, dtype):
def Test(self):
if not googletest.is_gpu_available():
tf_logging.info("Skipping RunGraphTest %s",
(n, m, k, transpose_a, transpose_b))
return
tf_logging.info("Testing RunGraphTest %s",
(n, m, k, transpose_a, transpose_b))
self._VerifyRunGraph(n, m, k, transpose_a, transpose_b, dtype)
return Test
class MatmulBenchmarkTest(googletest.TestCase):
def _StripNode(self, nd):
snode = node_def_pb2.NodeDef(name=nd.name, op=nd.op, input=nd.input)
if nd.device:
snode.device = nd.device
return snode
def _StripGraph(self, gd):
return graph_pb2.GraphDef(node=[self._StripNode(nd) for nd in gd.node])
def _VerifyBuildGraph(self, n, m, k, transpose_a, transpose_b, dtype):
graph = ops.Graph()
with graph.as_default():
matmul_benchmark.build_graph(googletest.gpu_device_name(), n, m, k,
transpose_a, transpose_b, dtype)
gd = graph.as_graph_def()
dev = googletest.gpu_device_name()
proto_expected = """
node { name: "random_uniform/shape" op: "Const" device: \"""" + dev + """\" }
node { name: "random_uniform/min" op: "Const" device: \"""" + dev + """\" }
node { name: "random_uniform/max" op: "Const" device: \"""" + dev + """\" }
node { name: "random_uniform/RandomUniform" op: "RandomUniform" input: "random_uniform/shape" device: \"""" + dev + """\" }
node { name: "random_uniform/sub" op: "Sub" input: "random_uniform/max" input: "random_uniform/min" device: \"""" + dev + """\" }
node { name: "random_uniform/mul" op: "Mul" input: "random_uniform/RandomUniform" input: "random_uniform/sub" device: \"""" + dev + """\" }
node { name: "random_uniform" op: "Add" input: "random_uniform/mul" input: "random_uniform/min" device: \"""" + dev + """\" }
node { name: "Variable" op: "VariableV2" device: \"""" + dev + """\" }
node { name: "Variable/Assign" op: "Assign" input: "Variable" input: "random_uniform" device: \"""" + dev + """\" }
node { name: "Variable/read" op: "Identity" input: "Variable" device: \"""" + dev + """\" }
node { name: "random_uniform_1/shape" op: "Const" device: \"""" + dev + """\" }
node { name: "random_uniform_1/min" op: "Const" device: \"""" + dev + """\" }
node { name: "random_uniform_1/max" op: "Const" device: \"""" + dev + """\" }
node { name: "random_uniform_1/RandomUniform" op: "RandomUniform" input: "random_uniform_1/shape" device: \"""" + dev + """\" }
node { name: "random_uniform_1/sub" op: "Sub" input: "random_uniform_1/max" input: "random_uniform_1/min" device: \"""" + dev + """\" }
node { name: "random_uniform_1/mul" op: "Mul" input: "random_uniform_1/RandomUniform" input: "random_uniform_1/sub" device: \"""" + dev + """\" }
node { name: "random_uniform_1" op: "Add" input: "random_uniform_1/mul" input: "random_uniform_1/min" device: \"""" + dev + """\" }
node { name: "Variable_1" op: "VariableV2" device: \"""" + dev + """\" }
node { name: "Variable_1/Assign" op: "Assign" input: "Variable_1" input: "random_uniform_1" device: \"""" + dev + """\" }
node { name: "Variable_1/read" op: "Identity" input: "Variable_1" device: \"""" + dev + """\" }
node { name: "MatMul" op: "MatMul" input: "Variable/read" input: "Variable_1/read" device: \"""" + dev + """\" }
node { name: "group_deps" op: "NoOp" input: "^MatMul" device: \"""" + dev + """\" }
"""
self.assertProtoEquals(str(proto_expected), self._StripGraph(gd))
def _VerifyRunGraph(self, n, m, k, transpose_a, transpose_b, dtype):
benchmark_instance = matmul_benchmark.MatmulBenchmark()
duration = benchmark_instance.run_graph(googletest.gpu_device_name(), n, m,
k, transpose_a, transpose_b, 1,
dtype)
self.assertTrue(duration > 1e-6)
if __name__ == "__main__":
dtypes = [np.float32, np.float64]
index = 0
for _dtype in dtypes:
for _n, _m, (_transpose_a, _transpose_b) in itertools.product(
[512, 1024], [1, 8, 16, 128], [(False, False), (True, False),
(False, True)]):
_k = _n
setattr(MatmulBenchmarkTest, "testBuildGraph_" + str(index),
BuildGraphTest(_n, _m, _k, _transpose_a, _transpose_b, _dtype))
setattr(MatmulBenchmarkTest, "testRunGraph_" + str(index),
RunGraphTest(_n, _m, _k, _transpose_a, _transpose_b, _dtype))
index += 1
googletest.main()
|
tensorflow-master
|
tensorflow/python/ops/matmul_benchmark_test.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Operations for clipping (gradient, weight) tensors to min/max values."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import six
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gen_array_ops
from tensorflow.python.ops import gen_nn_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.util import deprecation
from tensorflow.python.util import dispatch
from tensorflow.python.util.tf_export import tf_export
@tf_export("clip_by_value")
@dispatch.add_dispatch_support
def clip_by_value(t, clip_value_min, clip_value_max,
name=None):
"""Clips tensor values to a specified min and max.
Given a tensor `t`, this operation returns a tensor of the same type and
shape as `t` with its values clipped to `clip_value_min` and `clip_value_max`.
Any values less than `clip_value_min` are set to `clip_value_min`. Any values
greater than `clip_value_max` are set to `clip_value_max`.
Note: `clip_value_min` needs to be smaller or equal to `clip_value_max` for
correct results.
Args:
t: A `Tensor` or `IndexedSlices`.
clip_value_min: A 0-D (scalar) `Tensor`, or a `Tensor` with the same shape
as `t`. The minimum value to clip by.
clip_value_max: A 0-D (scalar) `Tensor`, or a `Tensor` with the same shape
as `t`. The maximum value to clip by.
name: A name for the operation (optional).
Returns:
A clipped `Tensor` or `IndexedSlices`.
Raises:
ValueError: If the clip tensors would trigger array broadcasting
that would make the returned tensor larger than the input.
"""
with ops.name_scope(name, "clip_by_value",
[t, clip_value_min, clip_value_max]) as name:
values = ops.convert_to_tensor(
t.values if isinstance(t, ops.IndexedSlices) else t, name="t")
# Go through list of tensors, for each value in each tensor clip
t_min = math_ops.minimum(values, clip_value_max)
# Assert that the shape is compatible with the initial shape,
# to prevent unintentional broadcasting.
_ = values.shape.merge_with(t_min.shape)
t_max = math_ops.maximum(t_min, clip_value_min, name=name)
_ = values.shape.merge_with(t_max.shape)
if isinstance(t, ops.IndexedSlices):
t_max = ops.IndexedSlices(t_max, t.indices, t.dense_shape)
return t_max
# TODO(scottzhu): switch to use new implmentation in 2 weeks.
# return gen_math_ops.clip_by_value(
# t, clip_value_min, clip_value_max, name=name)
# TODO(scottzhu): switch to use new implmentation in 2 weeks.
# @ops.RegisterGradient("ClipByValue")
def _clip_by_value_grad(op, grad):
"""Returns grad of clip_by_value."""
x = op.inputs[0]
y = op.inputs[1]
z = op.inputs[2]
gdtype = grad.dtype
sx = array_ops.shape(x)
sy = array_ops.shape(y)
sz = array_ops.shape(z)
gradshape = array_ops.shape(grad)
zeros = array_ops.zeros(gradshape, gdtype)
xymask = math_ops.less(x, y)
xzmask = math_ops.greater(x, z)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
rx, rz = gen_array_ops.broadcast_gradient_args(sx, sz)
xgrad = array_ops.where(math_ops.logical_or(xymask, xzmask), zeros, grad)
ygrad = array_ops.where(xymask, grad, zeros)
zgrad = array_ops.where(xzmask, grad, zeros)
gx = array_ops.reshape(math_ops.reduce_sum(xgrad, rx), sx)
gy = array_ops.reshape(math_ops.reduce_sum(ygrad, ry), sy)
gz = array_ops.reshape(math_ops.reduce_sum(zgrad, rz), sz)
return (gx, gy, gz)
@tf_export("clip_by_norm")
def clip_by_norm(t, clip_norm, axes=None, name=None):
"""Clips tensor values to a maximum L2-norm.
Given a tensor `t`, and a maximum clip value `clip_norm`, this operation
normalizes `t` so that its L2-norm is less than or equal to `clip_norm`,
along the dimensions given in `axes`. Specifically, in the default case
where all dimensions are used for calculation, if the L2-norm of `t` is
already less than or equal to `clip_norm`, then `t` is not modified. If
the L2-norm is greater than `clip_norm`, then this operation returns a
tensor of the same type and shape as `t` with its values set to:
`t * clip_norm / l2norm(t)`
In this case, the L2-norm of the output tensor is `clip_norm`.
As another example, if `t` is a matrix and `axes == [1]`, then each row
of the output will have L2-norm less than or equal to `clip_norm`. If
`axes == [0]` instead, each column of the output will be clipped.
This operation is typically used to clip gradients before applying them with
an optimizer.
Args:
t: A `Tensor` or `IndexedSlices`.
clip_norm: A 0-D (scalar) `Tensor` > 0. A maximum clipping value.
axes: A 1-D (vector) `Tensor` of type int32 containing the dimensions
to use for computing the L2-norm. If `None` (the default), uses all
dimensions.
name: A name for the operation (optional).
Returns:
A clipped `Tensor` or `IndexedSlices`.
"""
with ops.name_scope(name, "clip_by_norm", [t, clip_norm]) as name:
values = ops.convert_to_tensor(
t.values if isinstance(t, ops.IndexedSlices) else t, name="t")
# Calculate L2-norm, clip elements by ratio of clip_norm to L2-norm
l2sum = math_ops.reduce_sum(values * values, axes, keepdims=True)
pred = l2sum > 0
# Two-tap tf.where trick to bypass NaN gradients
l2sum_safe = array_ops.where(pred, l2sum, array_ops.ones_like(l2sum))
l2norm = array_ops.where(pred, math_ops.sqrt(l2sum_safe), l2sum)
intermediate = values * clip_norm
# Assert that the shape is compatible with the initial shape,
# to prevent unintentional broadcasting.
_ = values.shape.merge_with(intermediate.shape)
values_clip = array_ops.identity(
intermediate / math_ops.maximum(l2norm, clip_norm), name=name)
if isinstance(t, ops.IndexedSlices):
return ops.IndexedSlices(values_clip, t.indices, t.dense_shape)
return values_clip
@tf_export("linalg.global_norm", v1=["linalg.global_norm", "global_norm"])
@deprecation.deprecated_endpoints("global_norm")
def global_norm(t_list, name=None):
"""Computes the global norm of multiple tensors.
Given a tuple or list of tensors `t_list`, this operation returns the
global norm of the elements in all tensors in `t_list`. The global norm is
computed as:
`global_norm = sqrt(sum([l2norm(t)**2 for t in t_list]))`
Any entries in `t_list` that are of type None are ignored.
Args:
t_list: A tuple or list of mixed `Tensors`, `IndexedSlices`, or None.
name: A name for the operation (optional).
Returns:
A 0-D (scalar) `Tensor` of type `float`.
Raises:
TypeError: If `t_list` is not a sequence.
"""
if (not isinstance(t_list, collections.Sequence)
or isinstance(t_list, six.string_types)):
raise TypeError("t_list should be a sequence")
t_list = list(t_list)
with ops.name_scope(name, "global_norm", t_list) as name:
values = [
ops.convert_to_tensor(
t.values if isinstance(t, ops.IndexedSlices) else t,
name="t_%d" % i)
if t is not None else t
for i, t in enumerate(t_list)]
half_squared_norms = []
for v in values:
if v is not None:
with ops.colocate_with(v):
half_squared_norms.append(gen_nn_ops.l2_loss(v))
half_squared_norm = math_ops.reduce_sum(array_ops.stack(half_squared_norms))
norm = math_ops.sqrt(
half_squared_norm *
constant_op.constant(2.0, dtype=half_squared_norm.dtype),
name="global_norm")
return norm
@tf_export("clip_by_global_norm")
def clip_by_global_norm(t_list, clip_norm, use_norm=None, name=None):
"""Clips values of multiple tensors by the ratio of the sum of their norms.
Given a tuple or list of tensors `t_list`, and a clipping ratio `clip_norm`,
this operation returns a list of clipped tensors `list_clipped`
and the global norm (`global_norm`) of all tensors in `t_list`. Optionally,
if you've already computed the global norm for `t_list`, you can specify
the global norm with `use_norm`.
To perform the clipping, the values `t_list[i]` are set to:
t_list[i] * clip_norm / max(global_norm, clip_norm)
where:
global_norm = sqrt(sum([l2norm(t)**2 for t in t_list]))
If `clip_norm > global_norm` then the entries in `t_list` remain as they are,
otherwise they're all shrunk by the global ratio.
If `global_norm == infinity` then the entries in `t_list` are all set to `NaN`
to signal that an error occurred.
Any of the entries of `t_list` that are of type `None` are ignored.
This is the correct way to perform gradient clipping (for example, see
[Pascanu et al., 2012](http://arxiv.org/abs/1211.5063)
([pdf](http://arxiv.org/pdf/1211.5063.pdf))).
However, it is slower than `clip_by_norm()` because all the parameters must be
ready before the clipping operation can be performed.
Args:
t_list: A tuple or list of mixed `Tensors`, `IndexedSlices`, or None.
clip_norm: A 0-D (scalar) `Tensor` > 0. The clipping ratio.
use_norm: A 0-D (scalar) `Tensor` of type `float` (optional). The global
norm to use. If not provided, `global_norm()` is used to compute the norm.
name: A name for the operation (optional).
Returns:
list_clipped: A list of `Tensors` of the same type as `list_t`.
global_norm: A 0-D (scalar) `Tensor` representing the global norm.
Raises:
TypeError: If `t_list` is not a sequence.
"""
if (not isinstance(t_list, collections.Sequence)
or isinstance(t_list, six.string_types)):
raise TypeError("t_list should be a sequence")
t_list = list(t_list)
if use_norm is None:
use_norm = global_norm(t_list, name)
with ops.name_scope(name, "clip_by_global_norm",
t_list + [clip_norm]) as name:
# Calculate L2-norm, clip elements by ratio of clip_norm to L2-norm
scale_for_finite = clip_norm * math_ops.minimum(
1.0 / use_norm,
constant_op.constant(1.0, dtype=use_norm.dtype) / clip_norm)
scale = array_ops.where(
math_ops.is_finite(use_norm),
scale_for_finite,
# Return NaN if use_norm is not finite.
constant_op.constant(float("nan"), dtype=use_norm.dtype))
values = [
ops.convert_to_tensor(
t.values if isinstance(t, ops.IndexedSlices) else t,
name="t_%d" % i)
if t is not None else t
for i, t in enumerate(t_list)]
values_clipped = []
for i, v in enumerate(values):
if v is None:
values_clipped.append(None)
else:
with ops.colocate_with(v):
values_clipped.append(
array_ops.identity(v * scale, name="%s_%d" % (name, i)))
list_clipped = [
ops.IndexedSlices(c_v, t.indices, t.dense_shape)
if isinstance(t, ops.IndexedSlices)
else c_v
for (c_v, t) in zip(values_clipped, t_list)]
return list_clipped, use_norm
@deprecation.deprecated(
date=None,
instructions="clip_by_average_norm is deprecated in TensorFlow 2.0. Please "
"use clip_by_norm(t, clip_norm * tf.cast(tf.size(t), tf.float32), name) "
"instead.")
@tf_export(v1=["clip_by_average_norm"])
def clip_by_average_norm(t, clip_norm, name=None):
"""Clips tensor values to a maximum average L2-norm.
Given a tensor `t`, and a maximum clip value `clip_norm`, this operation
normalizes `t` so that its average L2-norm is less than or equal to
`clip_norm`. Specifically, if the average L2-norm is already less than or
equal to `clip_norm`, then `t` is not modified. If the average L2-norm is
greater than `clip_norm`, then this operation returns a tensor of the same
type and shape as `t` with its values set to:
`t * clip_norm / l2norm_avg(t)`
In this case, the average L2-norm of the output tensor is `clip_norm`.
This operation is typically used to clip gradients before applying them with
an optimizer.
Args:
t: A `Tensor`.
clip_norm: A 0-D (scalar) `Tensor` > 0. A maximum clipping value.
name: A name for the operation (optional).
Returns:
A clipped `Tensor`.
"""
with ops.name_scope(name, "clip_by_average_norm", [t, clip_norm]) as name:
t = ops.convert_to_tensor(t, name="t")
# Calculate L2-norm per element, clip elements by ratio of clip_norm to
# L2-norm per element
n_element = math_ops.cast(array_ops.size(t), dtypes.float32)
l2norm_inv = math_ops.rsqrt(
math_ops.reduce_sum(t * t, math_ops.range(array_ops.rank(t))))
tclip = array_ops.identity(
t * clip_norm * math_ops.minimum(
l2norm_inv * n_element, constant_op.constant(1.0) / clip_norm),
name=name)
return tclip
|
tensorflow-master
|
tensorflow/python/ops/clip_ops.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Utilities for calculating gradients."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import enum
from tensorflow.python.util.tf_export import tf_export
@tf_export("UnconnectedGradients")
class UnconnectedGradients(enum.Enum):
"""Controls how gradient computation behaves when y does not depend on x.
The gradient of y with respect to x can be zero in two different ways: there
could be no differentiable path in the graph connecting x to y (and so we can
statically prove that the gradient is zero) or it could be that runtime values
of tensors in a particular execution lead to a gradient of zero (say, if a
relu unit happens to not be activated). To allow you to distinguish between
these two cases you can choose what value gets returned for the gradient when
there is no path in the graph from x to y:
* `NONE`: Indicates that [None] will be returned if there is no path from x
to y
* `ZERO`: Indicates that a zero tensor will be returned in the shape of x.
"""
NONE = "none"
ZERO = "zero"
|
tensorflow-master
|
tensorflow/python/ops/unconnected_gradients.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#==============================================================================
"""Data Flow Operations."""
# pylint: disable=g-bad-name
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import hashlib
import threading
import six
from tensorflow.python.compat import compat
from tensorflow.python.eager import context
from tensorflow.python.framework import dtypes as _dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import random_seed
from tensorflow.python.framework import tensor_shape
from tensorflow.python.framework import tensor_util
from tensorflow.python.lib.io import python_io
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import gen_data_flow_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import resource_variable_ops
# go/tf-wildcard-import
# pylint: disable=wildcard-import
from tensorflow.python.ops.gen_data_flow_ops import *
from tensorflow.python.util import deprecation
from tensorflow.python.util.tf_export import tf_export
# pylint: enable=wildcard-import
def _as_type_list(dtypes):
"""Convert dtypes to a list of types."""
assert dtypes is not None
if not (isinstance(dtypes, list) or isinstance(dtypes, tuple)):
# We have a single type.
return [dtypes]
else:
# We have a list or tuple of types.
return list(dtypes)
def _as_shape_list(shapes,
dtypes,
unknown_dim_allowed=False,
unknown_rank_allowed=False):
"""Convert shapes to a list of tuples of int (or None)."""
del dtypes
if unknown_dim_allowed:
if (not isinstance(shapes, collections.Sequence) or not shapes or
any(shape is None or isinstance(shape, int) for shape in shapes)):
raise ValueError(
"When providing partial shapes, a list of shapes must be provided.")
if shapes is None:
return None
if isinstance(shapes, tensor_shape.TensorShape):
shapes = [shapes]
if not isinstance(shapes, (tuple, list)):
raise TypeError(
"shapes must be a TensorShape or a list or tuple of TensorShapes.")
if all(shape is None or isinstance(shape, int) for shape in shapes):
# We have a single shape.
shapes = [shapes]
shapes = [tensor_shape.as_shape(shape) for shape in shapes]
if not unknown_dim_allowed:
if any(not shape.is_fully_defined() for shape in shapes):
raise ValueError("All shapes must be fully defined: %s" % shapes)
if not unknown_rank_allowed:
if any([shape.dims is None for shape in shapes]):
raise ValueError("All shapes must have a defined rank: %s" % shapes)
return shapes
def _as_name_list(names, dtypes):
if names is None:
return None
if not isinstance(names, (list, tuple)):
names = [names]
if len(names) != len(dtypes):
raise ValueError("List of names must have the same length as the list "
"of dtypes")
return list(names)
def _shape_common(s1, s2):
"""The greatest lower bound (ordered by specificity) TensorShape."""
s1 = tensor_shape.TensorShape(s1)
s2 = tensor_shape.TensorShape(s2)
if s1.ndims is None or s2.ndims is None or s1.ndims != s2.ndims:
return tensor_shape.unknown_shape()
d = [
d1 if d1 is not None and d1 == d2 else None
for (d1, d2) in zip(s1.as_list(), s2.as_list())
]
return tensor_shape.TensorShape(d)
# pylint: disable=protected-access
@tf_export("queue.QueueBase",
v1=["queue.QueueBase", "io.QueueBase", "QueueBase"])
@deprecation.deprecated_endpoints(["io.QueueBase", "QueueBase"])
class QueueBase(object):
"""Base class for queue implementations.
A queue is a TensorFlow data structure that stores tensors across
multiple steps, and exposes operations that enqueue and dequeue
tensors.
Each queue element is a tuple of one or more tensors, where each
tuple component has a static dtype, and may have a static shape. The
queue implementations support versions of enqueue and dequeue that
handle single elements, versions that support enqueuing and
dequeuing a batch of elements at once.
See `tf.queue.FIFOQueue` and
`tf.queue.RandomShuffleQueue` for concrete
implementations of this class, and instructions on how to create
them.
"""
def __init__(self, dtypes, shapes, names, queue_ref):
"""Constructs a queue object from a queue reference.
The two optional lists, `shapes` and `names`, must be of the same length
as `dtypes` if provided. The values at a given index `i` indicate the
shape and name to use for the corresponding queue component in `dtypes`.
Args:
dtypes: A list of types. The length of dtypes must equal the number
of tensors in each element.
shapes: Constraints on the shapes of tensors in an element:
A list of shape tuples or None. This list is the same length
as dtypes. If the shape of any tensors in the element are constrained,
all must be; shapes can be None if the shapes should not be constrained.
names: Optional list of names. If provided, the `enqueue()` and
`dequeue()` methods will use dictionaries with these names as keys.
Must be None or a list or tuple of the same length as `dtypes`.
queue_ref: The queue reference, i.e. the output of the queue op.
Raises:
ValueError: If one of the arguments is invalid.
"""
self._dtypes = dtypes
if shapes is not None:
if len(shapes) != len(dtypes):
raise ValueError("Queue shapes must have the same length as dtypes")
self._shapes = [tensor_shape.TensorShape(s) for s in shapes]
else:
self._shapes = [tensor_shape.unknown_shape() for _ in self._dtypes]
if names is not None:
if len(names) != len(dtypes):
raise ValueError("Queue names must have the same length as dtypes")
self._names = names
else:
self._names = None
self._queue_ref = queue_ref
if context.executing_eagerly():
if context.context().scope_name:
self._name = context.context().scope_name
else:
self._name = "Empty"
self._resource_deleter = resource_variable_ops.EagerResourceDeleter(
queue_ref, None)
else:
self._name = self._queue_ref.op.name.split("/")[-1]
@staticmethod
def from_list(index, queues):
"""Create a queue using the queue reference from `queues[index]`.
Args:
index: An integer scalar tensor that determines the input that gets
selected.
queues: A list of `QueueBase` objects.
Returns:
A `QueueBase` object.
Raises:
TypeError: When `queues` is not a list of `QueueBase` objects,
or when the data types of `queues` are not all the same.
"""
if ((not queues) or (not isinstance(queues, list)) or
(not all(isinstance(x, QueueBase) for x in queues))):
raise TypeError("A list of queues expected")
dtypes = queues[0].dtypes
if not all(dtypes == q.dtypes for q in queues[1:]):
raise TypeError("Queues do not have matching component dtypes.")
names = queues[0].names
if not all(names == q.names for q in queues[1:]):
raise TypeError("Queues do not have matching component names.")
queue_shapes = [q.shapes for q in queues]
reduced_shapes = [
six.moves.reduce(_shape_common, s) for s in zip(*queue_shapes)
]
queue_refs = array_ops.stack([x.queue_ref for x in queues])
selected_queue = array_ops.gather(queue_refs, index)
return QueueBase(
dtypes=dtypes,
shapes=reduced_shapes,
names=names,
queue_ref=selected_queue)
@property
def queue_ref(self):
"""The underlying queue reference."""
return self._queue_ref
@property
def name(self):
"""The name of the underlying queue."""
if context.executing_eagerly():
return self._name
return self._queue_ref.op.name
@property
def dtypes(self):
"""The list of dtypes for each component of a queue element."""
return self._dtypes
@property
def shapes(self):
"""The list of shapes for each component of a queue element."""
return self._shapes
@property
def names(self):
"""The list of names for each component of a queue element."""
return self._names
def _check_enqueue_dtypes(self, vals):
"""Validate and convert `vals` to a list of `Tensor`s.
The `vals` argument can be a Tensor, a list or tuple of tensors, or a
dictionary with tensor values.
If it is a dictionary, the queue must have been constructed with a
`names` attribute and the dictionary keys must match the queue names.
If the queue was constructed with a `names` attribute, `vals` must
be a dictionary.
Args:
vals: A tensor, a list or tuple of tensors, or a dictionary..
Returns:
A list of `Tensor` objects.
Raises:
ValueError: If `vals` is invalid.
"""
if isinstance(vals, dict):
if not self._names:
raise ValueError("Queue must have names to enqueue a dictionary")
if sorted(self._names, key=str) != sorted(vals.keys(), key=str):
raise ValueError("Keys in dictionary to enqueue do not match "
"names of Queue. Dictionary: (%s), Queue: (%s)" %
(sorted(vals.keys()), sorted(self._names)))
# The order of values in `self._names` indicates the order in which the
# tensors in the dictionary `vals` must be listed.
vals = [vals[k] for k in self._names]
else:
if self._names:
raise ValueError("You must enqueue a dictionary in a Queue with names")
if not isinstance(vals, (list, tuple)):
vals = [vals]
tensors = []
for i, (val, dtype) in enumerate(zip(vals, self._dtypes)):
tensors.append(
ops.convert_to_tensor(val, dtype=dtype, name="component_%d" % i))
return tensors
def _scope_vals(self, vals):
"""Return a list of values to pass to `name_scope()`.
Args:
vals: A tensor, a list or tuple of tensors, or a dictionary.
Returns:
The values in vals as a list.
"""
if isinstance(vals, (list, tuple)):
return vals
elif isinstance(vals, dict):
return vals.values()
else:
return [vals]
def enqueue(self, vals, name=None):
"""Enqueues one element to this queue.
If the queue is full when this operation executes, it will block
until the element has been enqueued.
At runtime, this operation may raise an error if the queue is
`tf.QueueBase.close` before or during its execution. If the
queue is closed before this operation runs,
`tf.errors.CancelledError` will be raised. If this operation is
blocked, and either (i) the queue is closed by a close operation
with `cancel_pending_enqueues=True`, or (ii) the session is
`tf.Session.close`,
`tf.errors.CancelledError` will be raised.
Args:
vals: A tensor, a list or tuple of tensors, or a dictionary containing
the values to enqueue.
name: A name for the operation (optional).
Returns:
The operation that enqueues a new tuple of tensors to the queue.
"""
with ops.name_scope(name, "%s_enqueue" % self._name,
self._scope_vals(vals)) as scope:
vals = self._check_enqueue_dtypes(vals)
# NOTE(mrry): Not using a shape function because we need access to
# the `QueueBase` object.
for val, shape in zip(vals, self._shapes):
val.get_shape().assert_is_compatible_with(shape)
if self._queue_ref.dtype == _dtypes.resource:
return gen_data_flow_ops.queue_enqueue_v2(
self._queue_ref, vals, name=scope)
else:
return gen_data_flow_ops.queue_enqueue(
self._queue_ref, vals, name=scope)
def enqueue_many(self, vals, name=None):
"""Enqueues zero or more elements to this queue.
This operation slices each component tensor along the 0th dimension to
make multiple queue elements. All of the tensors in `vals` must have the
same size in the 0th dimension.
If the queue is full when this operation executes, it will block
until all of the elements have been enqueued.
At runtime, this operation may raise an error if the queue is
`tf.QueueBase.close` before or during its execution. If the
queue is closed before this operation runs,
`tf.errors.CancelledError` will be raised. If this operation is
blocked, and either (i) the queue is closed by a close operation
with `cancel_pending_enqueues=True`, or (ii) the session is
`tf.Session.close`,
`tf.errors.CancelledError` will be raised.
Args:
vals: A tensor, a list or tuple of tensors, or a dictionary
from which the queue elements are taken.
name: A name for the operation (optional).
Returns:
The operation that enqueues a batch of tuples of tensors to the queue.
"""
with ops.name_scope(name, "%s_EnqueueMany" % self._name,
self._scope_vals(vals)) as scope:
vals = self._check_enqueue_dtypes(vals)
# NOTE(mrry): Not using a shape function because we need access to
# the `QueueBase` object.
# NOTE(fchollet): the code that follow is verbose because it needs to be
# compatible with both TF v1 TensorShape behavior and TF v2 behavior.
batch_dim = tensor_shape.dimension_value(
vals[0].get_shape().with_rank_at_least(1)[0])
batch_dim = tensor_shape.Dimension(batch_dim)
for val, shape in zip(vals, self._shapes):
val_batch_dim = tensor_shape.dimension_value(
val.get_shape().with_rank_at_least(1)[0])
val_batch_dim = tensor_shape.Dimension(val_batch_dim)
batch_dim = batch_dim.merge_with(val_batch_dim)
val.get_shape()[1:].assert_is_compatible_with(shape)
return gen_data_flow_ops.queue_enqueue_many_v2(
self._queue_ref, vals, name=scope)
def _dequeue_return_value(self, tensors):
"""Return the value to return from a dequeue op.
If the queue has names, return a dictionary with the
names as keys. Otherwise return either a single tensor
or a list of tensors depending on the length of `tensors`.
Args:
tensors: List of tensors from the dequeue op.
Returns:
A single tensor, a list of tensors, or a dictionary
of tensors.
"""
if self._names:
# The returned values in `tensors` are in the same order as
# the names in `self._names`.
return {n: tensors[i] for i, n in enumerate(self._names)}
elif len(tensors) == 1:
return tensors[0]
else:
return tensors
def dequeue(self, name=None):
"""Dequeues one element from this queue.
If the queue is empty when this operation executes, it will block
until there is an element to dequeue.
At runtime, this operation may raise an error if the queue is
`tf.QueueBase.close` before or during its execution. If the
queue is closed, the queue is empty, and there are no pending
enqueue operations that can fulfill this request,
`tf.errors.OutOfRangeError` will be raised. If the session is
`tf.Session.close`,
`tf.errors.CancelledError` will be raised.
Args:
name: A name for the operation (optional).
Returns:
The tuple of tensors that was dequeued.
"""
if name is None:
name = "%s_Dequeue" % self._name
if self._queue_ref.dtype == _dtypes.resource:
ret = gen_data_flow_ops.queue_dequeue_v2(
self._queue_ref, self._dtypes, name=name)
else:
ret = gen_data_flow_ops.queue_dequeue(
self._queue_ref, self._dtypes, name=name)
# NOTE(mrry): Not using a shape function because we need access to
# the `QueueBase` object.
if not context.executing_eagerly():
op = ret[0].op
for output, shape in zip(op.values(), self._shapes):
output.set_shape(shape)
return self._dequeue_return_value(ret)
def dequeue_many(self, n, name=None):
"""Dequeues and concatenates `n` elements from this queue.
This operation concatenates queue-element component tensors along
the 0th dimension to make a single component tensor. All of the
components in the dequeued tuple will have size `n` in the 0th dimension.
If the queue is closed and there are less than `n` elements left, then an
`OutOfRange` exception is raised.
At runtime, this operation may raise an error if the queue is
`tf.QueueBase.close` before or during its execution. If the
queue is closed, the queue contains fewer than `n` elements, and
there are no pending enqueue operations that can fulfill this
request, `tf.errors.OutOfRangeError` will be raised. If the
session is `tf.Session.close`,
`tf.errors.CancelledError` will be raised.
Args:
n: A scalar `Tensor` containing the number of elements to dequeue.
name: A name for the operation (optional).
Returns:
The list of concatenated tensors that was dequeued.
"""
if name is None:
name = "%s_DequeueMany" % self._name
ret = gen_data_flow_ops.queue_dequeue_many_v2(
self._queue_ref, n=n, component_types=self._dtypes, name=name)
# NOTE(mrry): Not using a shape function because we need access to
# the Queue object.
if not context.executing_eagerly():
op = ret[0].op
batch_dim = tensor_shape.Dimension(
tensor_util.constant_value(op.inputs[1]))
for output, shape in zip(op.values(), self._shapes):
output.set_shape(
tensor_shape.TensorShape([batch_dim]).concatenate(shape))
return self._dequeue_return_value(ret)
def dequeue_up_to(self, n, name=None):
"""Dequeues and concatenates `n` elements from this queue.
**Note** This operation is not supported by all queues. If a queue does not
support DequeueUpTo, then a `tf.errors.UnimplementedError` is raised.
This operation concatenates queue-element component tensors along
the 0th dimension to make a single component tensor. If the queue
has not been closed, all of the components in the dequeued tuple
will have size `n` in the 0th dimension.
If the queue is closed and there are more than `0` but fewer than
`n` elements remaining, then instead of raising a
`tf.errors.OutOfRangeError` like `tf.QueueBase.dequeue_many`,
less than `n` elements are returned immediately. If the queue is
closed and there are `0` elements left in the queue, then a
`tf.errors.OutOfRangeError` is raised just like in `dequeue_many`.
Otherwise the behavior is identical to `dequeue_many`.
Args:
n: A scalar `Tensor` containing the number of elements to dequeue.
name: A name for the operation (optional).
Returns:
The tuple of concatenated tensors that was dequeued.
"""
if name is None:
name = "%s_DequeueUpTo" % self._name
ret = gen_data_flow_ops.queue_dequeue_up_to_v2(
self._queue_ref, n=n, component_types=self._dtypes, name=name)
# NOTE(mrry): Not using a shape function because we need access to
# the Queue object.
if not context.executing_eagerly():
op = ret[0].op
for output, shape in zip(op.values(), self._shapes):
output.set_shape(tensor_shape.TensorShape([None]).concatenate(shape))
return self._dequeue_return_value(ret)
def close(self, cancel_pending_enqueues=False, name=None):
"""Closes this queue.
This operation signals that no more elements will be enqueued in
the given queue. Subsequent `enqueue` and `enqueue_many`
operations will fail. Subsequent `dequeue` and `dequeue_many`
operations will continue to succeed if sufficient elements remain
in the queue. Subsequently dequeue and dequeue_many operations
that would otherwise block waiting for more elements (if close
hadn't been called) will now fail immediately.
If `cancel_pending_enqueues` is `True`, all pending requests will also
be canceled.
Args:
cancel_pending_enqueues: (Optional.) A boolean, defaulting to
`False` (described above).
name: A name for the operation (optional).
Returns:
The operation that closes the queue.
"""
if name is None:
name = "%s_Close" % self._name
if self._queue_ref.dtype == _dtypes.resource:
return gen_data_flow_ops.queue_close_v2(
self._queue_ref,
cancel_pending_enqueues=cancel_pending_enqueues,
name=name)
else:
return gen_data_flow_ops.queue_close(
self._queue_ref,
cancel_pending_enqueues=cancel_pending_enqueues,
name=name)
def is_closed(self, name=None):
"""Returns true if queue is closed.
This operation returns true if the queue is closed and false if the queue
is open.
Args:
name: A name for the operation (optional).
Returns:
True if the queue is closed and false if the queue is open.
"""
if name is None:
name = "%s_Is_Closed" % self._name
if self._queue_ref.dtype == _dtypes.resource:
return gen_data_flow_ops.queue_is_closed_v2(self._queue_ref, name=name)
else:
return gen_data_flow_ops.queue_is_closed_(self._queue_ref, name=name)
def size(self, name=None):
"""Compute the number of elements in this queue.
Args:
name: A name for the operation (optional).
Returns:
A scalar tensor containing the number of elements in this queue.
"""
if name is None:
name = "%s_Size" % self._name
if self._queue_ref.dtype == _dtypes.resource:
return gen_data_flow_ops.queue_size_v2(self._queue_ref, name=name)
else:
return gen_data_flow_ops.queue_size(self._queue_ref, name=name)
def _shared_name(shared_name):
if context.executing_eagerly():
return str(ops.uid())
return shared_name
@tf_export(
"queue.RandomShuffleQueue",
v1=["queue.RandomShuffleQueue",
"io.RandomShuffleQueue", "RandomShuffleQueue"])
@deprecation.deprecated_endpoints(
["io.RandomShuffleQueue", "RandomShuffleQueue"])
class RandomShuffleQueue(QueueBase):
"""A queue implementation that dequeues elements in a random order.
See `tf.queue.QueueBase` for a description of the methods on
this class.
"""
def __init__(self,
capacity,
min_after_dequeue,
dtypes,
shapes=None,
names=None,
seed=None,
shared_name=None,
name="random_shuffle_queue"):
"""Create a queue that dequeues elements in a random order.
A `RandomShuffleQueue` has bounded capacity; supports multiple
concurrent producers and consumers; and provides exactly-once
delivery.
A `RandomShuffleQueue` holds a list of up to `capacity`
elements. Each element is a fixed-length tuple of tensors whose
dtypes are described by `dtypes`, and whose shapes are optionally
described by the `shapes` argument.
If the `shapes` argument is specified, each component of a queue
element must have the respective fixed shape. If it is
unspecified, different queue elements may have different shapes,
but the use of `dequeue_many` is disallowed.
The `min_after_dequeue` argument allows the caller to specify a
minimum number of elements that will remain in the queue after a
`dequeue` or `dequeue_many` operation completes, to ensure a
minimum level of mixing of elements. This invariant is maintained
by blocking those operations until sufficient elements have been
enqueued. The `min_after_dequeue` argument is ignored after the
queue has been closed.
Args:
capacity: An integer. The upper bound on the number of elements
that may be stored in this queue.
min_after_dequeue: An integer (described above).
dtypes: A list of `DType` objects. The length of `dtypes` must equal
the number of tensors in each queue element.
shapes: (Optional.) A list of fully-defined `TensorShape` objects
with the same length as `dtypes`, or `None`.
names: (Optional.) A list of string naming the components in the queue
with the same length as `dtypes`, or `None`. If specified the dequeue
methods return a dictionary with the names as keys.
seed: A Python integer. Used to create a random seed. See
`tf.compat.v1.set_random_seed`
for behavior.
shared_name: (Optional.) If non-empty, this queue will be shared under
the given name across multiple sessions.
name: Optional name for the queue operation.
"""
dtypes = _as_type_list(dtypes)
shapes = _as_shape_list(shapes, dtypes)
names = _as_name_list(names, dtypes)
seed1, seed2 = random_seed.get_seed(seed)
if seed1 is None and seed2 is None:
seed1, seed2 = 0, 0
elif seed is None and shared_name is not None:
# This means that graph seed is provided but op seed is not provided.
# If shared_name is also provided, make seed2 depend only on the graph
# seed and shared_name. (seed2 from get_seed() is generally dependent on
# the id of the last op created.)
string = (str(seed1) + shared_name).encode("utf-8")
seed2 = int(hashlib.md5(string).hexdigest()[:8], 16) & 0x7FFFFFFF
queue_ref = gen_data_flow_ops.random_shuffle_queue_v2(
component_types=dtypes,
shapes=shapes,
capacity=capacity,
min_after_dequeue=min_after_dequeue,
seed=seed1,
seed2=seed2,
shared_name=_shared_name(shared_name),
name=name)
super(RandomShuffleQueue, self).__init__(dtypes, shapes, names, queue_ref)
@tf_export("queue.FIFOQueue", v1=["queue.FIFOQueue", "FIFOQueue"])
@deprecation.deprecated_endpoints("FIFOQueue")
class FIFOQueue(QueueBase):
"""A queue implementation that dequeues elements in first-in first-out order.
See `tf.queue.QueueBase` for a description of the methods on
this class.
"""
def __init__(self,
capacity,
dtypes,
shapes=None,
names=None,
shared_name=None,
name="fifo_queue"):
"""Creates a queue that dequeues elements in a first-in first-out order.
A `FIFOQueue` has bounded capacity; supports multiple concurrent
producers and consumers; and provides exactly-once delivery.
A `FIFOQueue` holds a list of up to `capacity` elements. Each
element is a fixed-length tuple of tensors whose dtypes are
described by `dtypes`, and whose shapes are optionally described
by the `shapes` argument.
If the `shapes` argument is specified, each component of a queue
element must have the respective fixed shape. If it is
unspecified, different queue elements may have different shapes,
but the use of `dequeue_many` is disallowed.
Args:
capacity: An integer. The upper bound on the number of elements
that may be stored in this queue.
dtypes: A list of `DType` objects. The length of `dtypes` must equal
the number of tensors in each queue element.
shapes: (Optional.) A list of fully-defined `TensorShape` objects
with the same length as `dtypes`, or `None`.
names: (Optional.) A list of string naming the components in the queue
with the same length as `dtypes`, or `None`. If specified the dequeue
methods return a dictionary with the names as keys.
shared_name: (Optional.) If non-empty, this queue will be shared under
the given name across multiple sessions.
name: Optional name for the queue operation.
"""
dtypes = _as_type_list(dtypes)
shapes = _as_shape_list(shapes, dtypes)
names = _as_name_list(names, dtypes)
queue_ref = gen_data_flow_ops.fifo_queue_v2(
component_types=dtypes,
shapes=shapes,
capacity=capacity,
shared_name=_shared_name(shared_name),
name=name)
super(FIFOQueue, self).__init__(dtypes, shapes, names, queue_ref)
@tf_export(
"queue.PaddingFIFOQueue",
v1=["queue.PaddingFIFOQueue", "io.PaddingFIFOQueue", "PaddingFIFOQueue"])
@deprecation.deprecated_endpoints(["io.PaddingFIFOQueue", "PaddingFIFOQueue"])
class PaddingFIFOQueue(QueueBase):
"""A FIFOQueue that supports batching variable-sized tensors by padding.
A `PaddingFIFOQueue` may contain components with dynamic shape, while also
supporting `dequeue_many`. See the constructor for more details.
See `tf.queue.QueueBase` for a description of the methods on
this class.
"""
def __init__(self,
capacity,
dtypes,
shapes,
names=None,
shared_name=None,
name="padding_fifo_queue"):
"""Creates a queue that dequeues elements in a first-in first-out order.
A `PaddingFIFOQueue` has bounded capacity; supports multiple concurrent
producers and consumers; and provides exactly-once delivery.
A `PaddingFIFOQueue` holds a list of up to `capacity` elements. Each
element is a fixed-length tuple of tensors whose dtypes are
described by `dtypes`, and whose shapes are described by the `shapes`
argument.
The `shapes` argument must be specified; each component of a queue
element must have the respective shape. Shapes of fixed
rank but variable size are allowed by setting any shape dimension to None.
In this case, the inputs' shape may vary along the given dimension, and
`dequeue_many` will pad the given dimension with zeros up to the maximum
shape of all elements in the given batch.
Args:
capacity: An integer. The upper bound on the number of elements
that may be stored in this queue.
dtypes: A list of `DType` objects. The length of `dtypes` must equal
the number of tensors in each queue element.
shapes: A list of `TensorShape` objects, with the same length as
`dtypes`. Any dimension in the `TensorShape` containing value
`None` is dynamic and allows values to be enqueued with
variable size in that dimension.
names: (Optional.) A list of string naming the components in the queue
with the same length as `dtypes`, or `None`. If specified the dequeue
methods return a dictionary with the names as keys.
shared_name: (Optional.) If non-empty, this queue will be shared under
the given name across multiple sessions.
name: Optional name for the queue operation.
Raises:
ValueError: If shapes is not a list of shapes, or the lengths of dtypes
and shapes do not match, or if names is specified and the lengths of
dtypes and names do not match.
"""
dtypes = _as_type_list(dtypes)
shapes = _as_shape_list(shapes, dtypes, unknown_dim_allowed=True)
names = _as_name_list(names, dtypes)
if len(dtypes) != len(shapes):
raise ValueError("Shapes must be provided for all components, "
"but received %d dtypes and %d shapes." % (len(dtypes),
len(shapes)))
queue_ref = gen_data_flow_ops.padding_fifo_queue_v2(
component_types=dtypes,
shapes=shapes,
capacity=capacity,
shared_name=_shared_name(shared_name),
name=name)
super(PaddingFIFOQueue, self).__init__(dtypes, shapes, names, queue_ref)
@tf_export("queue.PriorityQueue",
v1=["queue.PriorityQueue", "io.PriorityQueue", "PriorityQueue"])
@deprecation.deprecated_endpoints(["io.PriorityQueue", "PriorityQueue"])
class PriorityQueue(QueueBase):
"""A queue implementation that dequeues elements in prioritized order.
See `tf.queue.QueueBase` for a description of the methods on
this class.
"""
def __init__(self,
capacity,
types,
shapes=None,
names=None,
shared_name=None,
name="priority_queue"):
"""Creates a queue that dequeues elements in a first-in first-out order.
A `PriorityQueue` has bounded capacity; supports multiple concurrent
producers and consumers; and provides exactly-once delivery.
A `PriorityQueue` holds a list of up to `capacity` elements. Each
element is a fixed-length tuple of tensors whose dtypes are
described by `types`, and whose shapes are optionally described
by the `shapes` argument.
If the `shapes` argument is specified, each component of a queue
element must have the respective fixed shape. If it is
unspecified, different queue elements may have different shapes,
but the use of `dequeue_many` is disallowed.
Enqueues and Dequeues to the `PriorityQueue` must include an additional
tuple entry at the beginning: the `priority`. The priority must be
an int64 scalar (for `enqueue`) or an int64 vector (for `enqueue_many`).
Args:
capacity: An integer. The upper bound on the number of elements
that may be stored in this queue.
types: A list of `DType` objects. The length of `types` must equal
the number of tensors in each queue element, except the first priority
element. The first tensor in each element is the priority,
which must be type int64.
shapes: (Optional.) A list of fully-defined `TensorShape` objects,
with the same length as `types`, or `None`.
names: (Optional.) A list of strings naming the components in the queue
with the same length as `dtypes`, or `None`. If specified, the dequeue
methods return a dictionary with the names as keys.
shared_name: (Optional.) If non-empty, this queue will be shared under
the given name across multiple sessions.
name: Optional name for the queue operation.
"""
types = _as_type_list(types)
shapes = _as_shape_list(shapes, types)
queue_ref = gen_data_flow_ops.priority_queue_v2(
component_types=types,
shapes=shapes,
capacity=capacity,
shared_name=_shared_name(shared_name),
name=name)
priority_dtypes = [_dtypes.int64] + types
priority_shapes = [()] + shapes if shapes else shapes
super(PriorityQueue, self).__init__(priority_dtypes, priority_shapes, names,
queue_ref)
# TODO(josh11b): class BatchQueue(QueueBase):
class Barrier(object):
"""Represents a key-value map that persists across graph executions."""
def __init__(self, types, shapes=None, shared_name=None, name="barrier"):
"""Creates a barrier that persists across different graph executions.
A barrier represents a key-value map, where each key is a string, and
each value is a tuple of tensors.
At runtime, the barrier contains 'complete' and 'incomplete'
elements. A complete element has defined tensors for all
components of its value tuple, and may be accessed using
take_many. An incomplete element has some undefined components in
its value tuple, and may be updated using insert_many.
The barrier call `take_many` outputs values in a particular order.
First, it only outputs completed values. Second, the order in which
completed values are returned matches the order in which their very
first component was inserted into the barrier. So, for example, for this
sequence of insertions and removals:
barrier = Barrier((tf.string, tf.int32), shapes=((), ()))
barrier.insert_many(0, keys=["k1", "k2"], values=["a", "b"]).run()
barrier.insert_many(1, keys=["k1"], values=[1]).run()
barrier.insert_many(0, keys=["k3"], values=["c"]).run()
barrier.insert_many(1, keys=["k3"], values=[3]).run()
barrier.insert_many(1, keys=["k2"], values=[2]).run()
(indices, keys, values) = barrier.take_many(2)
(indices_val, keys_val, values0_val, values1_val) =
session.run([indices, keys, values[0], values[1]])
The output will be (up to permutation of "k1" and "k2"):
indices_val == (-2**63, -2**63)
keys_val == ("k1", "k2")
values0_val == ("a", "b")
values1_val == (1, 2)
Note the key "k2" was inserted into the barrier before "k3". Even though
"k3" was completed first, both are complete by the time
take_many is called. As a result, "k2" is prioritized and "k1" and "k2"
are returned first. "k3" remains in the barrier until the next execution
of `take_many`. Since "k1" and "k2" had their first insertions into
the barrier together, their indices are the same (-2**63). The index
of "k3" will be -2**63 + 1, because it was the next new inserted key.
Args:
types: A single dtype or a tuple of dtypes, corresponding to the
dtypes of the tensor elements that comprise a value in this barrier.
shapes: Optional. Constraints on the shapes of tensors in the values:
a single tensor shape tuple; a tuple of tensor shape tuples
for each barrier-element tuple component; or None if the shape should
not be constrained.
shared_name: Optional. If non-empty, this barrier will be shared under
the given name across multiple sessions.
name: Optional name for the barrier op.
Raises:
ValueError: If one of the `shapes` indicate no elements.
"""
self._types = _as_type_list(types)
if shapes is not None:
shapes = _as_shape_list(shapes, self._types)
self._shapes = [tensor_shape.TensorShape(s) for s in shapes]
for i, shape in enumerate(self._shapes):
if shape.num_elements() == 0:
raise ValueError("Empty tensors are not supported, but received "
"shape '%s' at index %d" % (shape, i))
else:
self._shapes = [tensor_shape.unknown_shape() for _ in self._types]
self._barrier_ref = gen_data_flow_ops.barrier(
component_types=self._types,
shapes=self._shapes,
shared_name=shared_name,
name=name)
if context.executing_eagerly():
self._name = context.context().scope_name
else:
self._name = self._barrier_ref.op.name.split("/")[-1]
@property
def barrier_ref(self):
"""Get the underlying barrier reference."""
return self._barrier_ref
@property
def name(self):
"""The name of the underlying barrier."""
if context.executing_eagerly():
return self._name
return self._barrier_ref.op.name
def insert_many(self, component_index, keys, values, name=None):
"""For each key, assigns the respective value to the specified component.
This operation updates each element at component_index.
Args:
component_index: The component of the value that is being assigned.
keys: A vector of keys, with length n.
values: An any-dimensional tensor of values, which are associated with the
respective keys. The first dimension must have length n.
name: Optional name for the op.
Returns:
The operation that performs the insertion.
Raises:
InvalidArgumentsError: If inserting keys and values without elements.
"""
if name is None:
name = "%s_BarrierInsertMany" % self._name
return gen_data_flow_ops.barrier_insert_many(
self._barrier_ref, keys, values, component_index, name=name)
def take_many(self,
num_elements,
allow_small_batch=False,
timeout=None,
name=None):
"""Takes the given number of completed elements from this barrier.
This operation concatenates completed-element component tensors along
the 0th dimension to make a single component tensor.
If barrier has no completed elements, this operation will block
until there are 'num_elements' elements to take.
TODO(b/25743580): the semantics of `allow_small_batch` are experimental
and may be extended to other cases in the future.
TODO(ebrevdo): If a take_many(allow_small_batch=True) is blocking
already when the barrier is closed, it will block for ever. Fix this
by using asynchronous operations.
Args:
num_elements: The number of elements to take.
allow_small_batch: If the barrier is closed, don't block if there are less
completed elements than requested, but instead return all available
completed elements.
timeout: This specifies the number of milliseconds to block
before returning with DEADLINE_EXCEEDED. (This option is not
supported yet.)
name: A name for the operation (optional).
Returns:
A tuple of (index, key, value_list).
"index" is a int64 tensor of length num_elements containing the
index of the insert_many call for which the very first component of
the given element was inserted into the Barrier, starting with
the value -2**63. Note, this value is different from the
index of the insert_many call for which the element was completed.
"key" is a string tensor of length num_elements containing the keys.
"value_list" is a tuple of tensors, each one with size num_elements
in the 0th dimension for each component in the barrier's values.
"""
if name is None:
name = "%s_BarrierTakeMany" % self._name
ret = gen_data_flow_ops.barrier_take_many(
self._barrier_ref,
num_elements,
self._types,
allow_small_batch,
timeout,
name=name)
# NOTE(mrry): Not using a shape function because we need access to
# the Barrier object.
if not context.executing_eagerly():
op = ret[0].op
if allow_small_batch:
batch_dim = None
else:
batch_dim = tensor_shape.Dimension(
tensor_util.constant_value(op.inputs[1]))
op.outputs[0].set_shape(tensor_shape.vector(batch_dim)) # indices
op.outputs[1].set_shape(tensor_shape.vector(batch_dim)) # keys
for output, shape in zip(op.outputs[2:], self._shapes): # value_list
output.set_shape(
tensor_shape.TensorShape([batch_dim]).concatenate(shape))
return ret
def close(self, cancel_pending_enqueues=False, name=None):
"""Closes this barrier.
This operation signals that no more new key values will be inserted in the
given barrier. Subsequent InsertMany operations with new keys will fail.
InsertMany operations that just complement already existing keys with other
components, will continue to succeed. Subsequent TakeMany operations will
continue to succeed if sufficient elements remain in the barrier. Subsequent
TakeMany operations that would block will fail immediately.
If `cancel_pending_enqueues` is `True`, all pending requests to the
underlying queue will also be canceled, and completing of already
started values is also not acceptable anymore.
Args:
cancel_pending_enqueues: (Optional.) A boolean, defaulting to
`False` (described above).
name: Optional name for the op.
Returns:
The operation that closes the barrier.
"""
if name is None:
name = "%s_BarrierClose" % self._name
return gen_data_flow_ops.barrier_close(
self._barrier_ref,
cancel_pending_enqueues=cancel_pending_enqueues,
name=name)
def ready_size(self, name=None):
"""Compute the number of complete elements in the given barrier.
Args:
name: A name for the operation (optional).
Returns:
A single-element tensor containing the number of complete elements in the
given barrier.
"""
if name is None:
name = "%s_BarrierReadySize" % self._name
return gen_data_flow_ops.barrier_ready_size(self._barrier_ref, name=name)
def incomplete_size(self, name=None):
"""Compute the number of incomplete elements in the given barrier.
Args:
name: A name for the operation (optional).
Returns:
A single-element tensor containing the number of incomplete elements in
the given barrier.
"""
if name is None:
name = "%s_BarrierIncompleteSize" % self._name
return gen_data_flow_ops.barrier_incomplete_size(
self._barrier_ref, name=name)
@tf_export(v1=["ConditionalAccumulatorBase"])
class ConditionalAccumulatorBase(object):
"""A conditional accumulator for aggregating gradients.
Up-to-date gradients (i.e., time step at which gradient was computed is
equal to the accumulator's time step) are added to the accumulator.
Extraction of the average gradient is blocked until the required number of
gradients has been accumulated.
"""
def __init__(self, dtype, shape, accumulator_ref):
"""Creates a new ConditionalAccumulator.
Args:
dtype: Datatype of the accumulated gradients.
shape: Shape of the accumulated gradients.
accumulator_ref: A handle to the conditional accumulator, created by sub-
classes
"""
self._dtype = dtype
if shape is not None:
self._shape = tensor_shape.TensorShape(shape)
else:
self._shape = tensor_shape.unknown_shape()
self._accumulator_ref = accumulator_ref
if context.executing_eagerly():
self._name = context.context().scope_name
else:
self._name = self._accumulator_ref.op.name.split("/")[-1]
@property
def accumulator_ref(self):
"""The underlying accumulator reference."""
return self._accumulator_ref
@property
def name(self):
"""The name of the underlying accumulator."""
return self._name
@property
def dtype(self):
"""The datatype of the gradients accumulated by this accumulator."""
return self._dtype
def num_accumulated(self, name=None):
"""Number of gradients that have currently been aggregated in accumulator.
Args:
name: Optional name for the operation.
Returns:
Number of accumulated gradients currently in accumulator.
"""
if name is None:
name = "%s_NumAccumulated" % self._name
if compat.forward_compatible(2019, 7, 8):
return gen_data_flow_ops.resource_accumulator_num_accumulated(
self._accumulator_ref, name=name)
return gen_data_flow_ops.accumulator_num_accumulated(
self._accumulator_ref, name=name)
def set_global_step(self, new_global_step, name=None):
"""Sets the global time step of the accumulator.
The operation logs a warning if we attempt to set to a time step that is
lower than the accumulator's own time step.
Args:
new_global_step: Value of new time step. Can be a variable or a constant
name: Optional name for the operation.
Returns:
Operation that sets the accumulator's time step.
"""
if compat.forward_compatible(2019, 7, 8):
return gen_data_flow_ops.resource_accumulator_set_global_step(
self._accumulator_ref,
math_ops.cast(ops.convert_to_tensor(new_global_step), _dtypes.int64),
name=name)
return gen_data_flow_ops.accumulator_set_global_step(
self._accumulator_ref,
math_ops.cast(ops.convert_to_tensor(new_global_step), _dtypes.int64),
name=name)
@tf_export(v1=["ConditionalAccumulator"])
class ConditionalAccumulator(ConditionalAccumulatorBase):
"""A conditional accumulator for aggregating gradients.
Up-to-date gradients (i.e., time step at which gradient was computed is
equal to the accumulator's time step) are added to the accumulator.
Extraction of the average gradient is blocked until the required number of
gradients has been accumulated.
"""
def __init__(self,
dtype,
shape=None,
shared_name=None,
name="conditional_accumulator",
reduction_type="MEAN"):
"""Creates a new ConditionalAccumulator.
Args:
dtype: Datatype of the accumulated gradients.
shape: Shape of the accumulated gradients.
shared_name: Optional. If non-empty, this accumulator will be shared under
the given name across multiple sessions.
name: Optional name for the accumulator.
reduction_type: Reduction type to use when taking the gradient.
"""
if compat.forward_compatible(2019, 7, 8):
accumulator_ref = gen_data_flow_ops.resource_conditional_accumulator(
dtype=dtype,
shape=shape,
shared_name=shared_name,
name=name,
reduction_type=reduction_type)
self._resource_deleter = resource_variable_ops.EagerResourceDeleter(
handle=accumulator_ref, handle_device=context.context().device_name)
else:
accumulator_ref = gen_data_flow_ops.conditional_accumulator(
dtype=dtype,
shape=shape,
shared_name=shared_name,
name=name,
reduction_type=reduction_type)
super(ConditionalAccumulator, self).__init__(dtype, shape, accumulator_ref)
def apply_grad(self, grad, local_step=0, name=None):
"""Attempts to apply a gradient to the accumulator.
The attempt is silently dropped if the gradient is stale, i.e., local_step
is less than the accumulator's global time step.
Args:
grad: The gradient tensor to be applied.
local_step: Time step at which the gradient was computed.
name: Optional name for the operation.
Returns:
The operation that (conditionally) applies a gradient to the accumulator.
Raises:
ValueError: If grad is of the wrong shape
"""
grad = ops.convert_to_tensor(grad, self._dtype)
grad.get_shape().assert_is_compatible_with(self._shape)
local_step = math_ops.cast(ops.convert_to_tensor(local_step), _dtypes.int64)
if compat.forward_compatible(2019, 7, 8):
return gen_data_flow_ops.resource_accumulator_apply_gradient(
self._accumulator_ref,
local_step=local_step,
gradient=grad,
name=name)
return gen_data_flow_ops.accumulator_apply_gradient(
self._accumulator_ref, local_step=local_step, gradient=grad, name=name)
def take_grad(self, num_required, name=None):
"""Attempts to extract the average gradient from the accumulator.
The operation blocks until sufficient number of gradients have been
successfully applied to the accumulator.
Once successful, the following actions are also triggered:
- Counter of accumulated gradients is reset to 0.
- Aggregated gradient is reset to 0 tensor.
- Accumulator's internal time step is incremented by 1.
Args:
num_required: Number of gradients that needs to have been aggregated
name: Optional name for the operation
Returns:
A tensor holding the value of the average gradient.
Raises:
InvalidArgumentError: If num_required < 1
"""
if compat.forward_compatible(2019, 7, 8):
out = gen_data_flow_ops.resource_accumulator_take_gradient(
self._accumulator_ref, num_required, dtype=self._dtype, name=name)
else:
out = gen_data_flow_ops.accumulator_take_gradient(
self._accumulator_ref, num_required, dtype=self._dtype, name=name)
out.set_shape(self._shape)
return out
@tf_export(
v1=["sparse.SparseConditionalAccumulator", "SparseConditionalAccumulator"])
class SparseConditionalAccumulator(ConditionalAccumulatorBase):
"""A conditional accumulator for aggregating sparse gradients.
Sparse gradients are represented by `IndexedSlices`.
Up-to-date gradients (i.e., time step at which gradient was computed is
equal to the accumulator's time step) are added to the accumulator.
Extraction of the average gradient is blocked until the required number of
gradients has been accumulated.
Args:
dtype: Datatype of the accumulated gradients.
shape: Shape of the accumulated gradients.
shared_name: Optional. If non-empty, this accumulator will be shared under
the given name across multiple sessions.
name: Optional name for the accumulator.
reduction_type: Reduction type to use when taking the gradient.
"""
def __init__(self,
dtype,
shape=None,
shared_name=None,
name="sparse_conditional_accumulator",
reduction_type="MEAN"):
accumulator_ref = gen_data_flow_ops.sparse_conditional_accumulator(
dtype=dtype,
shape=shape,
shared_name=shared_name,
name=name,
reduction_type=reduction_type)
super(SparseConditionalAccumulator, self).__init__(dtype, shape,
accumulator_ref)
def apply_indexed_slices_grad(self, grad, local_step=0, name=None):
"""Attempts to apply a gradient to the accumulator.
The attempt is silently dropped if the gradient is stale, i.e., `local_step`
is less than the accumulator's global time step.
Args:
grad: The gradient `IndexedSlices` to be applied.
local_step: Time step at which the gradient was computed.
name: Optional name for the operation.
Returns:
The operation that (conditionally) applies a gradient to the accumulator.
Raises:
InvalidArgumentError: If grad is of the wrong shape
"""
return self.apply_grad(
grad_indices=grad.indices,
grad_values=grad.values,
grad_shape=grad.dense_shape,
local_step=local_step,
name=name)
def apply_grad(self,
grad_indices,
grad_values,
grad_shape=None,
local_step=0,
name=None):
"""Attempts to apply a sparse gradient to the accumulator.
The attempt is silently dropped if the gradient is stale, i.e., `local_step`
is less than the accumulator's global time step.
A sparse gradient is represented by its indices, values and possibly empty
or None shape. Indices must be a vector representing the locations of
non-zero entries in the tensor. Values are the non-zero slices of the
gradient, and must have the same first dimension as indices, i.e., the nnz
represented by indices and values must be consistent. Shape, if not empty or
None, must be consistent with the accumulator's shape (if also provided).
Example:
A tensor [[0, 0], [0, 1], [2, 3]] can be represented
indices: [1,2]
values: [[0,1],[2,3]]
shape: [3, 2]
Args:
grad_indices: Indices of the sparse gradient to be applied.
grad_values: Values of the sparse gradient to be applied.
grad_shape: Shape of the sparse gradient to be applied.
local_step: Time step at which the gradient was computed.
name: Optional name for the operation.
Returns:
The operation that (conditionally) applies a gradient to the accumulator.
Raises:
InvalidArgumentError: If grad is of the wrong shape
"""
local_step = math_ops.cast(ops.convert_to_tensor(local_step), _dtypes.int64)
return gen_data_flow_ops.sparse_accumulator_apply_gradient(
self._accumulator_ref,
local_step=local_step,
gradient_indices=math_ops.cast(grad_indices, _dtypes.int64),
gradient_values=grad_values,
gradient_shape=math_ops.cast(
[] if grad_shape is None else grad_shape, _dtypes.int64),
has_known_shape=(grad_shape is not None),
name=name)
def take_grad(self, num_required, name=None):
"""Attempts to extract the average gradient from the accumulator.
The operation blocks until sufficient number of gradients have been
successfully applied to the accumulator.
Once successful, the following actions are also triggered:
- Counter of accumulated gradients is reset to 0.
- Aggregated gradient is reset to 0 tensor.
- Accumulator's internal time step is incremented by 1.
Args:
num_required: Number of gradients that needs to have been aggregated
name: Optional name for the operation
Returns:
A tuple of indices, values, and shape representing the average gradient.
Raises:
InvalidArgumentError: If `num_required` < 1
"""
return gen_data_flow_ops.sparse_accumulator_take_gradient(
self._accumulator_ref, num_required, dtype=self._dtype, name=name)
def take_indexed_slices_grad(self, num_required, name=None):
"""Attempts to extract the average gradient from the accumulator.
The operation blocks until sufficient number of gradients have been
successfully applied to the accumulator.
Once successful, the following actions are also triggered:
- Counter of accumulated gradients is reset to 0.
- Aggregated gradient is reset to 0 tensor.
- Accumulator's internal time step is incremented by 1.
Args:
num_required: Number of gradients that needs to have been aggregated
name: Optional name for the operation
Returns:
An `IndexedSlices` holding the value of the average gradient.
Raises:
InvalidArgumentError: If `num_required` < 1
"""
return_val = gen_data_flow_ops.sparse_accumulator_take_gradient(
self._accumulator_ref, num_required, dtype=self._dtype, name=name)
return ops.IndexedSlices(
indices=return_val.indices,
values=return_val.values,
dense_shape=return_val.shape)
class BaseStagingArea(object):
"""Base class for Staging Areas."""
_identifier = 0
_lock = threading.Lock()
def __init__(self,
dtypes,
shapes=None,
names=None,
shared_name=None,
capacity=0,
memory_limit=0):
if shared_name is None:
self._name = (
ops.get_default_graph().unique_name(self.__class__.__name__))
elif isinstance(shared_name, six.string_types):
self._name = shared_name
else:
raise ValueError("shared_name must be a string")
self._dtypes = dtypes
if shapes is not None:
if len(shapes) != len(dtypes):
raise ValueError("StagingArea shapes must be the same length as dtypes")
self._shapes = [tensor_shape.TensorShape(s) for s in shapes]
else:
self._shapes = [tensor_shape.unknown_shape() for _ in self._dtypes]
if names is not None:
if len(names) != len(dtypes):
raise ValueError("StagingArea names must be the same length as dtypes")
self._names = names
else:
self._names = None
self._capacity = capacity
self._memory_limit = memory_limit
# all get and put ops must colocate with this op
with ops.name_scope("%s_root" % self._name):
self._coloc_op = control_flow_ops.no_op()
@property
def name(self):
"""The name of the staging area."""
return self._name
@property
def dtypes(self):
"""The list of dtypes for each component of a staging area element."""
return self._dtypes
@property
def shapes(self):
"""The list of shapes for each component of a staging area element."""
return self._shapes
@property
def names(self):
"""The list of names for each component of a staging area element."""
return self._names
@property
def capacity(self):
"""The maximum number of elements of this staging area."""
return self._capacity
@property
def memory_limit(self):
"""The maximum number of bytes of this staging area."""
return self._memory_limit
def _check_put_dtypes(self, vals, indices=None):
"""Validate and convert `vals` to a list of `Tensor`s.
The `vals` argument can be a Tensor, a list or tuple of tensors, or a
dictionary with tensor values.
If `vals` is a list, then the appropriate indices associated with the
values must be provided.
If it is a dictionary, the staging area must have been constructed with a
`names` attribute and the dictionary keys must match the staging area names.
`indices` will be inferred from the dictionary keys.
If the staging area was constructed with a `names` attribute, `vals` must
be a dictionary.
Checks that the dtype and shape of each value matches that
of the staging area.
Args:
vals: A tensor, a list or tuple of tensors, or a dictionary.
Returns:
A (tensors, indices) tuple where `tensors` is a list of `Tensor` objects
and `indices` is a list of indices associed with the tensors.
Raises:
ValueError: If `vals` or `indices` is invalid.
"""
if isinstance(vals, dict):
if not self._names:
raise ValueError(
"Staging areas must have names to enqueue a dictionary")
if not set(vals.keys()).issubset(self._names):
raise ValueError("Keys in dictionary to put do not match names "
"of staging area. Dictionary: (%s), Queue: (%s)" %
(sorted(vals.keys()), sorted(self._names)))
# The order of values in `self._names` indicates the order in which the
# tensors in the dictionary `vals` must be listed.
vals, indices, _ = zip(*[(vals[k], i, k)
for i, k in enumerate(self._names)
if k in vals])
else:
if self._names:
raise ValueError("You must enqueue a dictionary in a staging area "
"with names")
if indices is None:
raise ValueError("Indices must be supplied when inserting a list "
"of tensors")
if len(indices) != len(vals):
raise ValueError("Number of indices '%s' doesn't match "
"number of values '%s'")
if not isinstance(vals, (list, tuple)):
vals = [vals]
indices = [0]
# Sanity check number of values
if not len(vals) <= len(self._dtypes):
raise ValueError("Unexpected number of inputs '%s' vs '%s'" %
(len(vals), len(self._dtypes)))
tensors = []
for val, i in zip(vals, indices):
dtype, shape = self._dtypes[i], self._shapes[i]
# Check dtype
if val.dtype != dtype:
raise ValueError("Datatypes do not match. '%s' != '%s'" %
(str(val.dtype), str(dtype)))
# Check shape
val.get_shape().assert_is_compatible_with(shape)
tensors.append(
ops.convert_to_tensor(val, dtype=dtype, name="component_%d" % i))
return tensors, indices
def _create_device_transfers(self, tensors):
"""Encode inter-device transfers if the current device
is not the same as the Staging Area's device.
"""
if not isinstance(tensors, (tuple, list)):
tensors = [tensors]
curr_device_scope = control_flow_ops.no_op().device
if curr_device_scope != self._coloc_op.device:
tensors = [array_ops.identity(t) for t in tensors]
return tensors
def _get_return_value(self, tensors, indices):
"""Return the value to return from a get op.
If the staging area has names, return a dictionary with the
names as keys. Otherwise return either a single tensor
or a list of tensors depending on the length of `tensors`.
Args:
tensors: List of tensors from the get op.
indices: Indices of associated names and shapes
Returns:
A single tensor, a list of tensors, or a dictionary
of tensors.
"""
tensors = self._create_device_transfers(tensors)
# Sets shape
for output, i in zip(tensors, indices):
output.set_shape(self._shapes[i])
if self._names:
# The returned values in `tensors` are in the same order as
# the names in `self._names`.
return {self._names[i]: t for t, i in zip(tensors, indices)}
return tensors
def _scope_vals(self, vals):
"""Return a list of values to pass to `name_scope()`.
Args:
vals: A tensor, a list or tuple of tensors, or a dictionary.
Returns:
The values in vals as a list.
"""
if isinstance(vals, (list, tuple)):
return vals
elif isinstance(vals, dict):
return vals.values()
else:
return [vals]
class StagingArea(BaseStagingArea):
"""Class for staging inputs. No ordering guarantees.
A `StagingArea` is a TensorFlow data structure that stores tensors across
multiple steps, and exposes operations that can put and get tensors.
Each `StagingArea` element is a tuple of one or more tensors, where each
tuple component has a static dtype, and may have a static shape.
The capacity of a `StagingArea` may be bounded or unbounded.
It supports multiple concurrent producers and consumers; and
provides exactly-once delivery.
Each element of a `StagingArea` is a fixed-length tuple of tensors whose
dtypes are described by `dtypes`, and whose shapes are optionally described
by the `shapes` argument.
If the `shapes` argument is specified, each component of a staging area
element must have the respective fixed shape. If it is
unspecified, different elements may have different shapes,
It can be configured with a capacity in which case
put(values) will block until space becomes available.
Similarly, it can be configured with a memory limit which
will block put(values) until space is available.
This is mostly useful for limiting the number of tensors on
devices such as GPUs.
All get() and peek() commands block if the requested data
is not present in the Staging Area.
"""
def __init__(self,
dtypes,
shapes=None,
names=None,
shared_name=None,
capacity=0,
memory_limit=0):
"""Constructs a staging area object.
The two optional lists, `shapes` and `names`, must be of the same length
as `dtypes` if provided. The values at a given index `i` indicate the
shape and name to use for the corresponding queue component in `dtypes`.
The device scope at the time of object creation determines where the
storage for the `StagingArea` will reside. Calls to `put` will incur a copy
to this memory space, if necessary. Tensors returned by `get` will be
placed according to the device scope when `get` is called.
Args:
dtypes: A list of types. The length of dtypes must equal the number
of tensors in each element.
shapes: (Optional.) Constraints on the shapes of tensors in an element.
A list of shape tuples or None. This list is the same length
as dtypes. If the shape of any tensors in the element are constrained,
all must be; shapes can be None if the shapes should not be constrained.
names: (Optional.) If provided, the `get()` and
`put()` methods will use dictionaries with these names as keys.
Must be None or a list or tuple of the same length as `dtypes`.
shared_name: (Optional.) A name to be used for the shared object. By
passing the same name to two different python objects they will share
the underlying staging area. Must be a string.
capacity: (Optional.) Maximum number of elements.
An integer. If zero, the Staging Area is unbounded
memory_limit: (Optional.) Maximum number of bytes of all tensors
in the Staging Area.
An integer. If zero, the Staging Area is unbounded
Raises:
ValueError: If one of the arguments is invalid.
"""
super(StagingArea, self).__init__(dtypes, shapes, names, shared_name,
capacity, memory_limit)
def put(self, values, name=None):
"""Create an op that places a value into the staging area.
This operation will block if the `StagingArea` has reached
its capacity.
Args:
values: A single tensor, a list or tuple of tensors, or a dictionary with
tensor values. The number of elements must match the length of the
list provided to the dtypes argument when creating the StagingArea.
name: A name for the operation (optional).
Returns:
The created op.
Raises:
ValueError: If the number or type of inputs don't match the staging area.
"""
with ops.name_scope(name, "%s_put" % self._name,
self._scope_vals(values)) as scope:
if not isinstance(values, (list, tuple, dict)):
values = [values]
# Hard-code indices for this staging area
indices = list(six.moves.range(len(values)))
vals, _ = self._check_put_dtypes(values, indices)
with ops.colocate_with(self._coloc_op):
op = gen_data_flow_ops.stage(
values=vals,
shared_name=self._name,
name=scope,
capacity=self._capacity,
memory_limit=self._memory_limit)
return op
def __internal_get(self, get_fn, name):
with ops.colocate_with(self._coloc_op):
ret = get_fn()
indices = list(six.moves.range(len(self._dtypes))) # Hard coded
return self._get_return_value(ret, indices)
def get(self, name=None):
"""Gets one element from this staging area.
If the staging area is empty when this operation executes, it will block
until there is an element to dequeue.
Note that unlike others ops that can block, like the queue Dequeue
operations, this can stop other work from happening. To avoid this, the
intended use is for this to be called only when there will be an element
already available. One method for doing this in a training loop would be to
run a `put()` call during a warmup session.run call, and then call both
`get()` and `put()` in each subsequent step.
The placement of the returned tensor will be determined by the current
device scope when this function is called.
Args:
name: A name for the operation (optional).
Returns:
The tuple of tensors that was gotten.
"""
if name is None:
name = "%s_get" % self._name
# pylint: disable=bad-continuation
fn = lambda: gen_data_flow_ops.unstage(dtypes=self._dtypes,
shared_name=self._name, name=name,
capacity=self._capacity,
memory_limit=self._memory_limit)
# pylint: enable=bad-continuation
return self.__internal_get(fn, name)
def peek(self, index, name=None):
"""Peeks at an element in the staging area.
If the staging area is too small to contain the element at
the specified index, it will block until enough elements
are inserted to complete the operation.
The placement of the returned tensor will be determined by
the current device scope when this function is called.
Args:
index: The index of the tensor within the staging area
to look up.
name: A name for the operation (optional).
Returns:
The tuple of tensors that was gotten.
"""
if name is None:
name = "%s_peek" % self._name
# pylint: disable=bad-continuation
fn = lambda: gen_data_flow_ops.stage_peek(index,
dtypes=self._dtypes, shared_name=self._name,
name=name, capacity=self._capacity,
memory_limit=self._memory_limit)
# pylint: enable=bad-continuation
return self.__internal_get(fn, name)
def size(self, name=None):
"""Returns the number of elements in the staging area.
Args:
name: A name for the operation (optional)
Returns:
The created op
"""
if name is None:
name = "%s_size" % self._name
return gen_data_flow_ops.stage_size(
name=name,
shared_name=self._name,
dtypes=self._dtypes,
capacity=self._capacity,
memory_limit=self._memory_limit)
def clear(self, name=None):
"""Clears the staging area.
Args:
name: A name for the operation (optional)
Returns:
The created op
"""
if name is None:
name = "%s_clear" % self._name
return gen_data_flow_ops.stage_clear(
name=name,
shared_name=self._name,
dtypes=self._dtypes,
capacity=self._capacity,
memory_limit=self._memory_limit)
class MapStagingArea(BaseStagingArea):
"""A `MapStagingArea` is a TensorFlow data structure that stores tensors
across multiple steps, and exposes operations that can put and get tensors.
Each `MapStagingArea` element is a (key, value) pair.
Only int64 keys are supported, other types should be
hashed to produce a key.
Values are a tuple of one or more tensors.
Each tuple component has a static dtype,
and may have a static shape.
The capacity of a `MapStagingArea` may be bounded or unbounded.
It supports multiple concurrent producers and consumers; and
provides exactly-once delivery.
Each value tuple of a `MapStagingArea` is a fixed-length tuple of tensors
whose
dtypes are described by `dtypes`, and whose shapes are optionally described
by the `shapes` argument.
If the `shapes` argument is specified, each component of a staging area
element must have the respective fixed shape. If it is
unspecified, different elements may have different shapes,
It behaves like an associative container with support for:
- put(key, values)
- peek(key) like dict.get(key)
- get(key) like dict.pop(key)
- get(key=None) like dict.popitem()
- size()
- clear()
If ordered a tree structure ordered by key will be used and
get(key=None) will remove (key, value) pairs in increasing key order.
Otherwise a hashtable
It can be configured with a capacity in which case
put(key, values) will block until space becomes available.
Similarly, it can be configured with a memory limit which
will block put(key, values) until space is available.
This is mostly useful for limiting the number of tensors on
devices such as GPUs.
All get() and peek() commands block if the requested
(key, value) pair is not present in the staging area.
Partial puts are supported and will be placed in an incomplete
map until such time as all values associated with the key have
been inserted. Once completed, this (key, value) pair will be
inserted into the map. Data in the incomplete map
counts towards the memory limit, but not towards capacity limit.
Partial gets from the map are also supported.
This removes the partially requested tensors from the entry,
but the entry is only removed from the map once all tensors
associated with it are removed.
"""
def __init__(self,
dtypes,
shapes=None,
names=None,
shared_name=None,
ordered=False,
capacity=0,
memory_limit=0):
"""Args:
dtypes: A list of types. The length of dtypes must equal the number
of tensors in each element.
capacity: (Optional.) Maximum number of elements.
An integer. If zero, the Staging Area is unbounded
memory_limit: (Optional.) Maximum number of bytes of all tensors
in the Staging Area (excluding keys).
An integer. If zero, the Staging Area is unbounded
ordered: (Optional.) If True the underlying data structure
is a tree ordered on key. Otherwise assume a hashtable.
shapes: (Optional.) Constraints on the shapes of tensors in an element.
A list of shape tuples or None. This list is the same length
as dtypes. If the shape of any tensors in the element are constrained,
all must be; shapes can be None if the shapes should not be constrained.
names: (Optional.) If provided, the `get()` and
`put()` methods will use dictionaries with these names as keys.
Must be None or a list or tuple of the same length as `dtypes`.
shared_name: (Optional.) A name to be used for the shared object. By
passing the same name to two different python objects they will share
the underlying staging area. Must be a string.
Raises:
ValueError: If one of the arguments is invalid.
"""
super(MapStagingArea, self).__init__(dtypes, shapes, names, shared_name,
capacity, memory_limit)
# Defer to different methods depending if the map is ordered
self._ordered = ordered
if ordered:
self._put_fn = gen_data_flow_ops.ordered_map_stage
self._pop_fn = gen_data_flow_ops.ordered_map_unstage
self._popitem_fn = gen_data_flow_ops.ordered_map_unstage_no_key
self._peek_fn = gen_data_flow_ops.ordered_map_peek
self._size_fn = gen_data_flow_ops.ordered_map_size
self._incomplete_size_fn = gen_data_flow_ops.ordered_map_incomplete_size
self._clear_fn = gen_data_flow_ops.ordered_map_clear
else:
self._put_fn = gen_data_flow_ops.map_stage
self._pop_fn = gen_data_flow_ops.map_unstage
self._popitem_fn = gen_data_flow_ops.map_unstage_no_key
self._peek_fn = gen_data_flow_ops.map_peek
self._size_fn = gen_data_flow_ops.map_size
self._incomplete_size_fn = gen_data_flow_ops.map_incomplete_size
self._clear_fn = gen_data_flow_ops.map_clear
def put(self, key, vals, indices=None, name=None):
"""Create an op that stores the (key, vals) pair in the staging area.
Incomplete puts are possible, preferably using a dictionary for vals
as the appropriate dtypes and shapes can be inferred from the value names
dictionary key values. If vals is a list or tuple, indices must
also be specified so that the op knows at which element position
to perform the insert.
This operation will block if the capacity or memory limit of this
container is reached.
Args:
key: Key associated with the data
vals: Tensor (or a dict/tuple of Tensors) to place
into the staging area.
indices: (Optional) if vals is a tuple/list, this is required.
name: A name for the operation (optional)
Returns:
The created op
Raises:
ValueError: If the number or type of inputs don't match the staging
area.
"""
with ops.name_scope(name, "%s_put" % self._name,
self._scope_vals(vals)) as scope:
vals, indices = self._check_put_dtypes(vals, indices)
with ops.colocate_with(self._coloc_op):
op = self._put_fn(
key,
indices,
vals,
dtypes=self._dtypes,
shared_name=self._name,
name=scope,
capacity=self._capacity,
memory_limit=self._memory_limit)
return op
def _get_indices_and_dtypes(self, indices=None):
if indices is None:
indices = list(six.moves.range(len(self._dtypes)))
if not isinstance(indices, (tuple, list)):
raise TypeError("Invalid indices type '%s'" % type(indices))
if len(indices) == 0:
raise ValueError("Empty indices")
if all(isinstance(i, str) for i in indices):
if self._names is None:
raise ValueError("String indices provided '%s', but this Staging Area "
"was not created with names." % indices)
try:
indices = [self._names.index(n) for n in indices]
except ValueError:
raise ValueError("Named index '%s' not in "
"Staging Area names '%s'" % (n, self._names))
elif all(isinstance(i, int) for i in indices):
pass
else:
raise TypeError("Mixed types in indices '%s'. "
"May only be str or int" % indices)
dtypes = [self._dtypes[i] for i in indices]
return indices, dtypes
def peek(self, key, indices=None, name=None):
"""Peeks at staging area data associated with the key.
If the key is not in the staging area, it will block
until the associated (key, value) is inserted.
Args:
key: Key associated with the required data
indices: Partial list of tensors to retrieve (optional).
A list of integer or string indices.
String indices are only valid if the Staging Area
has names associated with it.
name: A name for the operation (optional)
Returns:
The created op
"""
if name is None:
name = "%s_pop" % self._name
indices, dtypes = self._get_indices_and_dtypes(indices)
with ops.colocate_with(self._coloc_op):
result = self._peek_fn(
key,
shared_name=self._name,
indices=indices,
dtypes=dtypes,
name=name,
capacity=self._capacity,
memory_limit=self._memory_limit)
return self._get_return_value(result, indices)
def get(self, key=None, indices=None, name=None):
"""If the key is provided, the associated (key, value) is returned from the staging area.
If the key is not in the staging area, this method will block until
the associated (key, value) is inserted.
If no key is provided and the staging area is ordered,
the (key, value) with the smallest key will be returned.
Otherwise, a random (key, value) will be returned.
If the staging area is empty when this operation executes,
it will block until there is an element to dequeue.
Args:
key: Key associated with the required data (Optional)
indices: Partial list of tensors to retrieve (optional).
A list of integer or string indices.
String indices are only valid if the Staging Area
has names associated with it.
name: A name for the operation (optional)
Returns:
The created op
"""
if key is None:
return self._popitem(indices=indices, name=name)
else:
return self._pop(key, indices=indices, name=name)
def _pop(self, key, indices=None, name=None):
"""Remove and return the associated (key, value) is returned from the staging area.
If the key is not in the staging area, this method will block until
the associated (key, value) is inserted.
Args:
key: Key associated with the required data
indices: Partial list of tensors to retrieve (optional).
A list of integer or string indices.
String indices are only valid if the Staging Area
has names associated with it.
name: A name for the operation (optional)
Returns:
The created op
"""
if name is None:
name = "%s_get" % self._name
indices, dtypes = self._get_indices_and_dtypes(indices)
with ops.colocate_with(self._coloc_op):
result = self._pop_fn(
key,
shared_name=self._name,
indices=indices,
dtypes=dtypes,
name=name,
capacity=self._capacity,
memory_limit=self._memory_limit)
return key, self._get_return_value(result, indices)
def _popitem(self, indices=None, name=None):
"""If the staging area is ordered, the (key, value) with the smallest key will be returned.
Otherwise, a random (key, value) will be returned.
If the staging area is empty when this operation executes,
it will block until there is an element to dequeue.
Args:
key: Key associated with the required data
indices: Partial list of tensors to retrieve (optional).
A list of integer or string indices.
String indices are only valid if the Staging Area
has names associated with it.
name: A name for the operation (optional)
Returns:
The created op
"""
if name is None:
name = "%s_get_nokey" % self._name
indices, dtypes = self._get_indices_and_dtypes(indices)
with ops.colocate_with(self._coloc_op):
key, result = self._popitem_fn(
shared_name=self._name,
indices=indices,
dtypes=dtypes,
name=name,
capacity=self._capacity,
memory_limit=self._memory_limit)
# Separate keys and results out from
# underlying namedtuple
key = self._create_device_transfers(key)[0]
result = self._get_return_value(result, indices)
return key, result
def size(self, name=None):
"""Returns the number of elements in the staging area.
Args:
name: A name for the operation (optional)
Returns:
The created op
"""
if name is None:
name = "%s_size" % self._name
return self._size_fn(
shared_name=self._name,
name=name,
dtypes=self._dtypes,
capacity=self._capacity,
memory_limit=self._memory_limit)
def incomplete_size(self, name=None):
"""Returns the number of incomplete elements in the staging area.
Args:
name: A name for the operation (optional)
Returns:
The created op
"""
if name is None:
name = "%s_incomplete_size" % self._name
return self._incomplete_size_fn(
shared_name=self._name,
name=name,
dtypes=self._dtypes,
capacity=self._capacity,
memory_limit=self._memory_limit)
def clear(self, name=None):
"""Clears the staging area.
Args:
name: A name for the operation (optional)
Returns:
The created op
"""
if name is None:
name = "%s_clear" % self._name
return self._clear_fn(
shared_name=self._name,
name=name,
dtypes=self._dtypes,
capacity=self._capacity,
memory_limit=self._memory_limit)
class RecordInput(object):
"""RecordInput asynchronously reads and randomly yields TFRecords.
A RecordInput Op will continuously read a batch of records asynchronously
into a buffer of some fixed capacity. It can also asynchronously yield
random records from this buffer.
It will not start yielding until at least `buffer_size / 2` elements have been
placed into the buffer so that sufficient randomization can take place.
The order the files are read will be shifted each epoch by `shift_amount` so
that the data is presented in a different order every epoch.
"""
def __init__(self,
file_pattern,
batch_size=1,
buffer_size=1,
parallelism=1,
shift_ratio=0,
seed=0,
name=None,
batches=None,
compression_type=None):
"""Constructs a RecordInput Op.
Args:
file_pattern: File path to the dataset, possibly containing wildcards.
All matching files will be iterated over each epoch.
batch_size: How many records to return at a time.
buffer_size: The maximum number of records the buffer will contain.
parallelism: How many reader threads to use for reading from files.
shift_ratio: What percentage of the total number files to move the start
file forward by each epoch.
seed: Specify the random number seed used by generator that randomizes
records.
name: Optional name for the operation.
batches: None by default, creating a single batch op. Otherwise specifies
how many batches to create, which are returned as a list when
`get_yield_op()` is called. An example use case is to split processing
between devices on one computer.
compression_type: The type of compression for the file. Currently ZLIB and
GZIP are supported. Defaults to none.
Raises:
ValueError: If one of the arguments is invalid.
"""
self._batch_size = batch_size
if batches is not None:
self._batch_size *= batches
self._batches = batches
self._file_pattern = file_pattern
self._buffer_size = buffer_size
self._parallelism = parallelism
self._shift_ratio = shift_ratio
self._seed = seed
self._name = name
self._compression_type = python_io.TFRecordCompressionType.NONE
if compression_type is not None:
self._compression_type = compression_type
def get_yield_op(self):
"""Adds a node that yields a group of records every time it is executed.
If RecordInput `batches` parameter is not None, it yields a list of
record batches with the specified `batch_size`.
"""
compression_type = python_io.TFRecordOptions.get_compression_type_string(
python_io.TFRecordOptions(self._compression_type))
records = gen_data_flow_ops.record_input(
file_pattern=self._file_pattern,
file_buffer_size=self._buffer_size,
file_parallelism=self._parallelism,
file_shuffle_shift_ratio=self._shift_ratio,
batch_size=self._batch_size,
file_random_seed=self._seed,
compression_type=compression_type,
name=self._name)
if self._batches is None:
return records
else:
with ops.name_scope(self._name):
batch_list = [[] for _ in six.moves.range(self._batches)]
records = array_ops.split(records, self._batch_size, 0)
records = [array_ops.reshape(record, []) for record in records]
for index, protobuf in zip(six.moves.range(len(records)), records):
batch_index = index % self._batches
batch_list[batch_index].append(protobuf)
return batch_list
|
tensorflow-master
|
tensorflow/python/ops/data_flow_ops.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Inplace operations.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gen_array_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.util import deprecation
def _inplace_helper(x, i, v, op):
"""Applies an inplace op on (x, i, v).
op is one of gen_array_ops.alias_inplace_update,
gen_array_ops.alias_inplace_add, or gen_array_ops.alias_inplace_sub.
If i is None, x and v must be the same shape. Computes
x op v;
If i is a scalar, x has a rank 1 higher than v's. Computes
x[i, :] op v;
Otherwise, x and v must have the same rank. Computes
x[i, :] op v;
Args:
x: A Tensor.
i: None, a scalar or a vector.
v: A Tensor.
op: alias_inplace_update, alias_inplace_add, or alias_inplace_sub.
Returns:
Returns x.
"""
x = ops.convert_to_tensor(x)
v = ops.convert_to_tensor(v, x.dtype)
if i is None:
# Full tensor.
return array_ops.reshape(
op(array_ops.reshape(x, [1, -1]), [0], array_ops.reshape(v, [1, -1])),
array_ops.shape(x))
i = math_ops.cast(i, dtypes.int32)
if i.get_shape().ndims == 0:
# Single 0-dim update.
return op(x, array_ops.reshape(i, [1]), array_ops.expand_dims(v, 0))
return op(x, i, v)
@deprecation.deprecated(
None,
('Prefer tf.tensor_scatter_nd_update, which offers the same functionality '
'with well-defined read-write semantics.'))
def alias_inplace_update(x, i, v):
"""Applies an inplace update on input x at index i with value v. Aliases x.
If i is None, x and v must be the same shape. Computes
x = v;
If i is a scalar, x has a rank 1 higher than v's. Computes
x[i, :] = v;
Otherwise, x and v must have the same rank. Computes
x[i, :] = v;
Args:
x: A Tensor.
i: None, a scalar or a vector.
v: A Tensor.
Returns:
Returns x.
"""
return _inplace_helper(x, i, v, gen_array_ops.inplace_update)
@deprecation.deprecated(
None,
('Prefer tf.tensor_scatter_nd_add, which offers the same functionality '
'with well-defined read-write semantics.'))
def alias_inplace_add(x, i, v):
"""Applies an inplace add on input x at index i with value v. Aliases x.
If i is None, x and v must be the same shape. Computes
x += v;
If i is a scalar, x has a rank 1 higher than v's. Computes
x[i, :] += v;
Otherwise, x and v must have the same rank. Computes
x[i, :] += v;
Args:
x: A Tensor.
i: None, a scalar or a vector.
v: A Tensor.
Returns:
Returns x.
"""
return _inplace_helper(x, i, v, gen_array_ops.inplace_add)
@deprecation.deprecated(
None,
('Prefer tf.tensor_scatter_nd_sub, which offers the same functionality '
'with well-defined read-write semantics.'))
def alias_inplace_sub(x, i, v):
"""Applies an inplace sub on input x at index i with value v. Aliases x.
If i is None, x and v must be the same shape. Computes
x -= v;
If i is a scalar, x has a rank 1 higher than v's. Computes
x[i, :] -= v;
Otherwise, x and v must have the same rank. Computes
x[i, :] -= v;
Args:
x: A Tensor.
i: None, a scalar or a vector.
v: A Tensor.
Returns:
Returns x.
"""
return _inplace_helper(x, i, v, gen_array_ops.inplace_sub)
def empty_like(x, init=None):
"""Returns a non-initialized tensor with the same shape and dtype as x.
Args:
x: A Tensor.
init: Initialize the returned tensor with the default value of
x.dtype(), if True. Otherwise, do not initialize. Defaults to
None.
Returns:
A tensor y, whose dtype and shape are the same as those of x.
y is guaranteed not to be an alias of x. Upon return, y may contain
arbitrary data.
"""
x = ops.convert_to_tensor(x)
return gen_array_ops.empty(array_ops.shape(x), x.dtype, init=init)
@deprecation.deprecated(
None,
('Prefer tf.tensor_scatter_nd_update, which offers the same functionality '
'with well-defined read-write semantics.'))
def inplace_update(x, i, v):
"""Applies an inplace update on input x at index i with value v.
Note that this function is not actually inplace - it allocates
a copy of x. The utility is not avoiding memory copies but rather
specifying a sparse update.
If i is None, x and v must be the same shape. Computes
y = x; y = v;
If i is a scalar, x has a rank 1 higher than v's. Computes
y = x; y[i, :] = v;
Otherwise, x and v must have the same rank. Computes
y = x; y[i, :] = v;
Args:
x: A Tensor.
i: None, a scalar or a vector.
v: A Tensor.
Returns:
Returns y, which is guaranteed not to be an alias of x.
"""
return alias_inplace_update(gen_array_ops.deep_copy(x), i, v)
@deprecation.deprecated(
None,
('Prefer tf.tensor_scatter_nd_add, which offers the same functionality '
'with well-defined read-write semantics.'))
def inplace_add(x, i, v):
"""Applies an inplace add on input x at index i with value v.
Note that this function is not actually inplace - it allocates
a copy of x. The utility is not avoiding memory copies but rather
specifying a sparse update.
If i is None, x and v must be the same shape. Computes
y = x; y += v;
If i is a scalar, x has a rank 1 higher than v's. Computes
y = x; y[i, :] += v;
Otherwise, x and v must have the same rank. Computes
y = x; y[i, :] += v;
Args:
x: A Tensor.
i: None, a scalar or a vector.
v: A Tensor.
Returns:
Returns y, which is guaranteed not to be an alias of x.
"""
return alias_inplace_add(gen_array_ops.deep_copy(x), i, v)
@deprecation.deprecated(
None,
('Prefer tf.tensor_scatter_nd_sub, which offers the same functionality '
'with well-defined read-write semantics.'))
def inplace_sub(x, i, v):
"""Applies an inplace sub on input x at index i with value v.
Note that this function is not actually inplace - it allocates
a copy of x. The utility is not avoiding memory copies but rather
specifying a sparse update.
If i is None, x and v must be the same shape. Computes
y = x; y -= v;
If i is a scalar, x has a rank 1 higher than v's. Computes
y = x; y[i, :] -= v;
Otherwise, x and v must have the same rank. Computes
y = x; y[i, :] -= v;
Args:
x: A Tensor.
i: None, a scalar or a vector.
v: A Tensor.
Returns:
Returns y, which is guaranteed not to be an alias of x.
"""
return alias_inplace_sub(gen_array_ops.deep_copy(x), i, v)
empty = gen_array_ops.empty
|
tensorflow-master
|
tensorflow/python/ops/inplace_ops.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for Dequantize Operations."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.ops import array_ops
from tensorflow.python.platform import test
class DequantizeOpTest(test.TestCase):
def __init__(self, method_name="runTest"):
super(DequantizeOpTest, self).__init__(method_name)
def _testDequantizeOp(self, inputs, min_range, max_range, dtype):
with self.cached_session():
input_op = constant_op.constant(inputs, shape=[len(inputs)], dtype=dtype)
dequantized = array_ops.dequantize(input_op, min_range, max_range)
tf_ans = self.evaluate(dequantized)
# TODO(vrv): Add support for DT_QINT32 quantization if needed.
type_dict = {
dtypes.quint8: np.uint8,
dtypes.qint8: np.int8,
dtypes.quint16: np.uint16,
dtypes.qint16: np.int16
}
self.assertTrue(dtype in type_dict.keys())
v_max = np.iinfo(type_dict[dtype]).max
v_min = np.iinfo(type_dict[dtype]).min
self.assertTrue(min_range >= v_min)
self.assertTrue(max_range <= v_max)
type_range = v_max - v_min
if v_min < 0:
half_range = (type_range + 1) / 2
else:
half_range = 0.0
np_ans = ((inputs.astype(np.float32) + half_range) *
(max_range - min_range) / type_range) + min_range
self.assertAllClose(tf_ans, np_ans, rtol=1e-5, atol=1e-5)
def testBasicQuint8(self):
self._testDequantizeOp(np.array([0, 128, 255]), 0.0, 6.0, dtypes.quint8)
self._testDequantizeOp(np.array([0, 128, 255]), 0.0, 123.456, dtypes.quint8)
self._testDequantizeOp(
np.array([0, 4, 42, 108, 243]), 5.0, 200.2, dtypes.quint8)
def testBasicQint8(self):
self._testDequantizeOp(np.array([-128, 0, 127]), -1.0, 2.0, dtypes.qint8)
self._testDequantizeOp(np.array([-2, 4, -17]), -5.0, -3.0, dtypes.qint8)
self._testDequantizeOp(np.array([0, -4, 42, -108]), 5.0, 40.0, dtypes.qint8)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/ops/dequantize_op_test.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# =============================================================================
# pylint: disable=unused-import,g-bad-import-order
"""Neural network support.
See the [Neural network](https://tensorflow.org/api_guides/python/nn) guide.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import sys as _sys
# pylint: disable=unused-import
from tensorflow.python.ops import ctc_ops as _ctc_ops
from tensorflow.python.ops import embedding_ops as _embedding_ops
from tensorflow.python.ops import nn_grad as _nn_grad
from tensorflow.python.ops import nn_ops as _nn_ops
from tensorflow.python.ops.math_ops import sigmoid
from tensorflow.python.ops.math_ops import tanh
# pylint: enable=unused-import
# Bring more nn-associated functionality into this package.
# go/tf-wildcard-import
# pylint: disable=wildcard-import,unused-import
from tensorflow.python.ops.ctc_ops import *
from tensorflow.python.ops.nn_impl import *
from tensorflow.python.ops.nn_ops import *
from tensorflow.python.ops.candidate_sampling_ops import *
from tensorflow.python.ops.embedding_ops import *
# pylint: enable=wildcard-import,unused-import
|
tensorflow-master
|
tensorflow/python/ops/nn.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Variable class."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import abc
import enum # pylint: disable=g-bad-import-order
import itertools
import functools
import os
import six
from tensorflow.core.framework import attr_value_pb2
from tensorflow.core.framework import variable_pb2
from tensorflow.python.eager import context
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import tensor_shape
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import gen_array_ops
from tensorflow.python.ops import gen_state_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import state_ops
from tensorflow.python.platform import tf_logging as logging
from tensorflow.python.training.tracking import base as trackable
from tensorflow.python.util import compat
from tensorflow.python.util import tf_should_use
from tensorflow.python.util.deprecation import deprecated
from tensorflow.python.util.tf_export import tf_export
def default_variable_creator(_, **kwds):
del kwds
raise NotImplementedError("variable_scope needs to be imported")
def default_variable_creator_v2(_, **kwds):
del kwds
raise NotImplementedError("variable_scope needs to be imported")
def _make_getter(captured_getter, captured_previous):
"""To avoid capturing loop variables."""
def getter(**kwargs):
return captured_getter(captured_previous, **kwargs)
return getter
@tf_export("VariableSynchronization")
class VariableSynchronization(enum.Enum):
"""Indicates when a distributed variable will be synced.
* `AUTO`: Indicates that the synchronization will be determined by the current
`DistributionStrategy` (eg. With `MirroredStrategy` this would be
`ON_WRITE`).
* `NONE`: Indicates that there will only be one copy of the variable, so
there is no need to sync.
* `ON_WRITE`: Indicates that the variable will be updated across devices
every time it is written.
* `ON_READ`: Indicates that the variable will be aggregated across devices
when it is read (eg. when checkpointing or when evaluating an op that uses
the variable).
"""
AUTO = 0
NONE = 1
ON_WRITE = 2
ON_READ = 3
# LINT.IfChange
@tf_export("VariableAggregation", v1=[])
class VariableAggregationV2(enum.Enum):
"""Indicates how a distributed variable will be aggregated.
`tf.distribute.Strategy` distributes a model by making multiple copies
(called "replicas") acting data-parallel on different elements of the input
batch. When performing some variable-update operation, say
`var.assign_add(x)`, in a model, we need to resolve how to combine the
different values for `x` computed in the different replicas.
* `NONE`: This is the default, giving an error if you use a
variable-update operation with multiple replicas.
* `SUM`: Add the updates across replicas.
* `MEAN`: Take the arithmetic mean ("average") of the updates across replicas.
* `ONLY_FIRST_REPLICA`: This is for when every replica is performing the same
update, but we only want to perform the update once. Used, e.g., for the
global step counter.
"""
NONE = 0
SUM = 1
MEAN = 2
ONLY_FIRST_REPLICA = 3
def __hash__(self):
return hash(self.value)
def __eq__(self, other):
if self is other:
return True
elif isinstance(other, VariableAggregation):
return int(self.value) == int(other.value)
else:
return False
@tf_export(v1=["VariableAggregation"])
class VariableAggregation(enum.Enum):
NONE = 0
SUM = 1
MEAN = 2
ONLY_FIRST_REPLICA = 3
ONLY_FIRST_TOWER = 3 # DEPRECATED
def __hash__(self):
return hash(self.value)
# LINT.ThenChange(//tensorflow/core/framework/variable.proto)
#
# Note that we are currently relying on the integer values of the Python enums
# matching the integer values of the proto enums.
VariableAggregation.__doc__ = (
VariableAggregationV2.__doc__ +
"* `ONLY_FIRST_TOWER`: Deprecated alias for `ONLY_FIRST_REPLICA`.\n ")
def validate_synchronization_aggregation_trainable(
synchronization, aggregation, trainable, name):
"""Given user-provided variable properties, sets defaults and validates."""
if aggregation is None:
aggregation = VariableAggregation.NONE
else:
if not isinstance(aggregation,
(VariableAggregation, VariableAggregationV2)):
try:
aggregation = VariableAggregationV2(aggregation)
except ValueError:
raise ValueError(
"Invalid variable aggregation mode: {} for variable: {}".format(
aggregation, name))
if synchronization is None:
synchronization = VariableSynchronization.AUTO
else:
try:
synchronization = VariableSynchronization(synchronization)
except ValueError:
raise ValueError(
"Invalid variable synchronization mode: {} for variable: {}".format(
synchronization, name))
if trainable is None:
trainable = synchronization != VariableSynchronization.ON_READ
return synchronization, aggregation, trainable
class VariableMetaclass(type):
"""Metaclass to allow construction of tf.Variable to be overridden."""
def _variable_v1_call(cls,
initial_value=None,
trainable=None,
collections=None,
validate_shape=True,
caching_device=None,
name=None,
variable_def=None,
dtype=None,
expected_shape=None,
import_scope=None,
constraint=None,
use_resource=None,
synchronization=VariableSynchronization.AUTO,
aggregation=VariableAggregation.NONE,
shape=None):
"""Call on Variable class. Useful to force the signature."""
previous_getter = lambda **kwargs: default_variable_creator(None, **kwargs)
for _, getter in ops.get_default_graph()._variable_creator_stack: # pylint: disable=protected-access
previous_getter = _make_getter(getter, previous_getter)
# Reset `aggregation` that is explicitly set as `None` to the enum NONE.
if aggregation is None:
aggregation = VariableAggregation.NONE
return previous_getter(
initial_value=initial_value,
trainable=trainable,
collections=collections,
validate_shape=validate_shape,
caching_device=caching_device,
name=name,
variable_def=variable_def,
dtype=dtype,
expected_shape=expected_shape,
import_scope=import_scope,
constraint=constraint,
use_resource=use_resource,
synchronization=synchronization,
aggregation=aggregation,
shape=shape)
def _variable_v2_call(cls,
initial_value=None,
trainable=None,
validate_shape=True,
caching_device=None,
name=None,
variable_def=None,
dtype=None,
import_scope=None,
constraint=None,
synchronization=VariableSynchronization.AUTO,
aggregation=VariableAggregation.NONE,
shape=None):
"""Call on Variable class. Useful to force the signature."""
previous_getter = lambda **kws: default_variable_creator_v2(None, **kws)
for _, getter in ops.get_default_graph()._variable_creator_stack: # pylint: disable=protected-access
previous_getter = _make_getter(getter, previous_getter)
# Reset `aggregation` that is explicitly set as `None` to the enum NONE.
if aggregation is None:
aggregation = VariableAggregation.NONE
return previous_getter(
initial_value=initial_value,
trainable=trainable,
validate_shape=validate_shape,
caching_device=caching_device,
name=name,
variable_def=variable_def,
dtype=dtype,
import_scope=import_scope,
constraint=constraint,
synchronization=synchronization,
aggregation=aggregation,
shape=shape)
def __call__(cls, *args, **kwargs):
if cls is VariableV1:
return cls._variable_v1_call(*args, **kwargs)
elif cls is Variable:
return cls._variable_v2_call(*args, **kwargs)
else:
return super(VariableMetaclass, cls).__call__(*args, **kwargs)
@tf_export("Variable", v1=[])
class Variable(six.with_metaclass(VariableMetaclass,
trackable.Trackable)):
"""See the [Variables Guide](https://tensorflow.org/guide/variables).
A variable maintains state in the graph across calls to `run()`. You add a
variable to the graph by constructing an instance of the class `Variable`.
The `Variable()` constructor requires an initial value for the variable,
which can be a `Tensor` of any type and shape. The initial value defines the
type and shape of the variable. After construction, the type and shape of
the variable are fixed. The value can be changed using one of the assign
methods.
If you want to change the shape of a variable later you have to use an
`assign` Op with `validate_shape=False`.
Just like any `Tensor`, variables created with `Variable()` can be used as
inputs for other Ops in the graph. Additionally, all the operators
overloaded for the `Tensor` class are carried over to variables, so you can
also add nodes to the graph by just doing arithmetic on variables.
```python
import tensorflow as tf
# Create a variable.
w = tf.Variable(<initial-value>, name=<optional-name>)
# Use the variable in the graph like any Tensor.
y = tf.matmul(w, ...another variable or tensor...)
# The overloaded operators are available too.
z = tf.sigmoid(w + y)
# Assign a new value to the variable with `assign()` or a related method.
w.assign(w + 1.0)
w.assign_add(1.0)
```
When you launch the graph, variables have to be explicitly initialized before
you can run Ops that use their value. You can initialize a variable by
running its *initializer op*, restoring the variable from a save file, or
simply running an `assign` Op that assigns a value to the variable. In fact,
the variable *initializer op* is just an `assign` Op that assigns the
variable's initial value to the variable itself.
```python
# Launch the graph in a session.
with tf.compat.v1.Session() as sess:
# Run the variable initializer.
sess.run(w.initializer)
# ...you now can run ops that use the value of 'w'...
```
The most common initialization pattern is to use the convenience function
`global_variables_initializer()` to add an Op to the graph that initializes
all the variables. You then run that Op after launching the graph.
```python
# Add an Op to initialize global variables.
init_op = tf.compat.v1.global_variables_initializer()
# Launch the graph in a session.
with tf.compat.v1.Session() as sess:
# Run the Op that initializes global variables.
sess.run(init_op)
# ...you can now run any Op that uses variable values...
```
If you need to create a variable with an initial value dependent on another
variable, use the other variable's `initialized_value()`. This ensures that
variables are initialized in the right order.
All variables are automatically collected in the graph where they are
created. By default, the constructor adds the new variable to the graph
collection `GraphKeys.GLOBAL_VARIABLES`. The convenience function
`global_variables()` returns the contents of that collection.
When building a machine learning model it is often convenient to distinguish
between variables holding the trainable model parameters and other variables
such as a `global step` variable used to count training steps. To make this
easier, the variable constructor supports a `trainable=<bool>` parameter. If
`True`, the new variable is also added to the graph collection
`GraphKeys.TRAINABLE_VARIABLES`. The convenience function
`trainable_variables()` returns the contents of this collection. The
various `Optimizer` classes use this collection as the default list of
variables to optimize.
WARNING: tf.Variable objects by default have a non-intuitive memory model. A
Variable is represented internally as a mutable Tensor which can
non-deterministically alias other Tensors in a graph. The set of operations
which consume a Variable and can lead to aliasing is undetermined and can
change across TensorFlow versions. Avoid writing code which relies on the
value of a Variable either changing or not changing as other operations
happen. For example, using Variable objects or simple functions thereof as
predicates in a `tf.cond` is dangerous and error-prone:
```
v = tf.Variable(True)
tf.cond(v, lambda: v.assign(False), my_false_fn) # Note: this is broken.
```
Here, adding `use_resource=True` when constructing the variable will
fix any nondeterminism issues:
```
v = tf.Variable(True, use_resource=True)
tf.cond(v, lambda: v.assign(False), my_false_fn)
```
To use the replacement for variables which does
not have these issues:
* Add `use_resource=True` when constructing `tf.Variable`;
* Call `tf.compat.v1.get_variable_scope().set_use_resource(True)` inside a
`tf.compat.v1.variable_scope` before the `tf.compat.v1.get_variable()` call.
"""
def __init__(self,
initial_value=None,
trainable=None,
validate_shape=True,
caching_device=None,
name=None,
variable_def=None,
dtype=None,
import_scope=None,
constraint=None,
synchronization=VariableSynchronization.AUTO,
aggregation=VariableAggregation.NONE,
shape=None):
"""Creates a new variable with value `initial_value`.
The new variable is added to the graph collections listed in `collections`,
which defaults to `[GraphKeys.GLOBAL_VARIABLES]`.
If `trainable` is `True` the variable is also added to the graph collection
`GraphKeys.TRAINABLE_VARIABLES`.
This constructor creates both a `variable` Op and an `assign` Op to set the
variable to its initial value.
Args:
initial_value: A `Tensor`, or Python object convertible to a `Tensor`,
which is the initial value for the Variable. The initial value must have
a shape specified unless `validate_shape` is set to False. Can also be a
callable with no argument that returns the initial value when called. In
that case, `dtype` must be specified. (Note that initializer functions
from init_ops.py must first be bound to a shape before being used here.)
trainable: If `True`, GradientTapes automatically watch uses
of this variable. Defaults to `True`, unless `synchronization` is
set to `ON_READ`, in which case it defaults to `False`.
validate_shape: If `False`, allows the variable to be initialized with a
value of unknown shape. If `True`, the default, the shape of
`initial_value` must be known.
caching_device: Optional device string describing where the Variable
should be cached for reading. Defaults to the Variable's device.
If not `None`, caches on another device. Typical use is to cache
on the device where the Ops using the Variable reside, to deduplicate
copying through `Switch` and other conditional statements.
name: Optional name for the variable. Defaults to `'Variable'` and gets
uniquified automatically.
variable_def: `VariableDef` protocol buffer. If not `None`, recreates
the Variable object with its contents, referencing the variable's nodes
in the graph, which must already exist. The graph is not changed.
`variable_def` and the other arguments are mutually exclusive.
dtype: If set, initial_value will be converted to the given type.
If `None`, either the datatype will be kept (if `initial_value` is
a Tensor), or `convert_to_tensor` will decide.
import_scope: Optional `string`. Name scope to add to the
`Variable.` Only used when initializing from protocol buffer.
constraint: An optional projection function to be applied to the variable
after being updated by an `Optimizer` (e.g. used to implement norm
constraints or value constraints for layer weights). The function must
take as input the unprojected Tensor representing the value of the
variable and return the Tensor for the projected value
(which must have the same shape). Constraints are not safe to
use when doing asynchronous distributed training.
synchronization: Indicates when a distributed a variable will be
aggregated. Accepted values are constants defined in the class
`tf.VariableSynchronization`. By default the synchronization is set to
`AUTO` and the current `DistributionStrategy` chooses
when to synchronize.
aggregation: Indicates how a distributed variable will be aggregated.
Accepted values are constants defined in the class
`tf.VariableAggregation`.
shape: (optional) The shape of this variable. If None, the shape of
`initial_value` will be used. When setting this argument to
`tf.TensorShape(None)` (representing an unspecified shape), the variable
can be assigned with values of different shapes.
Raises:
ValueError: If both `variable_def` and initial_value are specified.
ValueError: If the initial value is not specified, or does not have a
shape and `validate_shape` is `True`.
RuntimeError: If eager execution is enabled.
"""
raise NotImplementedError
def __repr__(self):
raise NotImplementedError
def value(self):
"""Returns the last snapshot of this variable.
You usually do not need to call this method as all ops that need the value
of the variable call it automatically through a `convert_to_tensor()` call.
Returns a `Tensor` which holds the value of the variable. You can not
assign a new value to this tensor as it is not a reference to the variable.
To avoid copies, if the consumer of the returned value is on the same device
as the variable, this actually returns the live value of the variable, not
a copy. Updates to the variable are seen by the consumer. If the consumer
is on a different device it will get a copy of the variable.
Returns:
A `Tensor` containing the value of the variable.
"""
raise NotImplementedError
def read_value(self):
"""Returns the value of this variable, read in the current context.
Can be different from value() if it's on another device, with control
dependencies, etc.
Returns:
A `Tensor` containing the value of the variable.
"""
raise NotImplementedError
def set_shape(self, shape):
"""Overrides the shape for this variable.
Args:
shape: the `TensorShape` representing the overridden shape.
"""
raise NotImplementedError
@property
def trainable(self):
raise NotImplementedError
@property
def synchronization(self):
raise NotImplementedError
@property
def aggregation(self):
raise NotImplementedError
def eval(self, session=None):
"""In a session, computes and returns the value of this variable.
This is not a graph construction method, it does not add ops to the graph.
This convenience method requires a session where the graph
containing this variable has been launched. If no session is
passed, the default session is used. See `tf.compat.v1.Session` for more
information on launching a graph and on sessions.
```python
v = tf.Variable([1, 2])
init = tf.compat.v1.global_variables_initializer()
with tf.compat.v1.Session() as sess:
sess.run(init)
# Usage passing the session explicitly.
print(v.eval(sess))
# Usage with the default session. The 'with' block
# above makes 'sess' the default session.
print(v.eval())
```
Args:
session: The session to use to evaluate this variable. If
none, the default session is used.
Returns:
A numpy `ndarray` with a copy of the value of this variable.
"""
raise NotImplementedError
@deprecated(
None,
"Use Variable.read_value. Variables in 2.X are initialized "
"automatically both in eager and graph (inside tf.defun) contexts.")
def initialized_value(self):
"""Returns the value of the initialized variable.
You should use this instead of the variable itself to initialize another
variable with a value that depends on the value of this variable.
```python
# Initialize 'v' with a random tensor.
v = tf.Variable(tf.random.truncated_normal([10, 40]))
# Use `initialized_value` to guarantee that `v` has been
# initialized before its value is used to initialize `w`.
# The random values are picked only once.
w = tf.Variable(v.initialized_value() * 2.0)
```
Returns:
A `Tensor` holding the value of this variable after its initializer
has run.
"""
with ops.init_scope():
return control_flow_ops.cond(is_variable_initialized(self),
self.read_value,
lambda: self.initial_value)
@property
def initial_value(self):
"""Returns the Tensor used as the initial value for the variable.
Note that this is different from `initialized_value()` which runs
the op that initializes the variable before returning its value.
This method returns the tensor that is used by the op that initializes
the variable.
Returns:
A `Tensor`.
"""
raise NotImplementedError
@property
def constraint(self):
"""Returns the constraint function associated with this variable.
Returns:
The constraint function that was passed to the variable constructor.
Can be `None` if no constraint was passed.
"""
raise NotImplementedError
def assign(self, value, use_locking=False, name=None, read_value=True):
"""Assigns a new value to the variable.
This is essentially a shortcut for `assign(self, value)`.
Args:
value: A `Tensor`. The new value for this variable.
use_locking: If `True`, use locking during the assignment.
name: The name of the operation to be created
read_value: if True, will return something which evaluates to the
new value of the variable; if False will return the assign op.
Returns:
A `Tensor` that will hold the new value of this variable after
the assignment has completed.
"""
raise NotImplementedError
def assign_add(self, delta, use_locking=False, name=None, read_value=True):
"""Adds a value to this variable.
This is essentially a shortcut for `assign_add(self, delta)`.
Args:
delta: A `Tensor`. The value to add to this variable.
use_locking: If `True`, use locking during the operation.
name: The name of the operation to be created
read_value: if True, will return something which evaluates to the
new value of the variable; if False will return the assign op.
Returns:
A `Tensor` that will hold the new value of this variable after
the addition has completed.
"""
raise NotImplementedError
def assign_sub(self, delta, use_locking=False, name=None, read_value=True):
"""Subtracts a value from this variable.
This is essentially a shortcut for `assign_sub(self, delta)`.
Args:
delta: A `Tensor`. The value to subtract from this variable.
use_locking: If `True`, use locking during the operation.
name: The name of the operation to be created
read_value: if True, will return something which evaluates to the
new value of the variable; if False will return the assign op.
Returns:
A `Tensor` that will hold the new value of this variable after
the subtraction has completed.
"""
raise NotImplementedError
def scatter_sub(self, sparse_delta, use_locking=False, name=None):
"""Subtracts `tf.IndexedSlices` from this variable.
Args:
sparse_delta: `tf.IndexedSlices` to be subtracted from this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered subtraction has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
raise NotImplementedError
def scatter_add(self, sparse_delta, use_locking=False, name=None):
"""Adds `tf.IndexedSlices` to this variable.
Args:
sparse_delta: `tf.IndexedSlices` to be added to this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered addition has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
raise NotImplementedError
def scatter_max(self, sparse_delta, use_locking=False, name=None):
"""Updates this variable with the max of `tf.IndexedSlices` and itself.
Args:
sparse_delta: `tf.IndexedSlices` to use as an argument of max
with this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered maximization has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
raise NotImplementedError
def scatter_min(self, sparse_delta, use_locking=False, name=None):
"""Updates this variable with the min of `tf.IndexedSlices` and itself.
Args:
sparse_delta: `tf.IndexedSlices` to use as an argument of min
with this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered minimization has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
raise NotImplementedError
def scatter_mul(self, sparse_delta, use_locking=False, name=None):
"""Multiply this variable by `tf.IndexedSlices`.
Args:
sparse_delta: `tf.IndexedSlices` to multiply this variable by.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered multiplication has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
raise NotImplementedError
def scatter_div(self, sparse_delta, use_locking=False, name=None):
"""Divide this variable by `tf.IndexedSlices`.
Args:
sparse_delta: `tf.IndexedSlices` to divide this variable by.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered division has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
raise NotImplementedError
def scatter_update(self, sparse_delta, use_locking=False, name=None):
"""Assigns `tf.IndexedSlices` to this variable.
Args:
sparse_delta: `tf.IndexedSlices` to be assigned to this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered assignment has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
raise NotImplementedError
def batch_scatter_update(self, sparse_delta, use_locking=False, name=None):
"""Assigns `tf.IndexedSlices` to this variable batch-wise.
Analogous to `batch_gather`. This assumes that this variable and the
sparse_delta IndexedSlices have a series of leading dimensions that are the
same for all of them, and the updates are performed on the last dimension of
indices. In other words, the dimensions should be the following:
`num_prefix_dims = sparse_delta.indices.ndims - 1`
`batch_dim = num_prefix_dims + 1`
`sparse_delta.updates.shape = sparse_delta.indices.shape + var.shape[
batch_dim:]`
where
`sparse_delta.updates.shape[:num_prefix_dims]`
`== sparse_delta.indices.shape[:num_prefix_dims]`
`== var.shape[:num_prefix_dims]`
And the operation performed can be expressed as:
`var[i_1, ..., i_n,
sparse_delta.indices[i_1, ..., i_n, j]] = sparse_delta.updates[
i_1, ..., i_n, j]`
When sparse_delta.indices is a 1D tensor, this operation is equivalent to
`scatter_update`.
To avoid this operation one can looping over the first `ndims` of the
variable and using `scatter_update` on the subtensors that result of slicing
the first dimension. This is a valid option for `ndims = 1`, but less
efficient than this implementation.
Args:
sparse_delta: `tf.IndexedSlices` to be assigned to this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered assignment has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
raise NotImplementedError
def scatter_nd_sub(self, indices, updates, name=None):
"""Applies sparse subtraction to individual values or slices in a Variable.
Assuming the variable has rank `P` and `indices` is a `Tensor` of rank `Q`.
`indices` must be integer tensor, containing indices into self.
It must be shape `[d_0, ..., d_{Q-2}, K]` where `0 < K <= P`.
The innermost dimension of `indices` (with length `K`) corresponds to
indices into elements (if `K = P`) or slices (if `K < P`) along the `K`th
dimension of self.
`updates` is `Tensor` of rank `Q-1+P-K` with shape:
```
[d_0, ..., d_{Q-2}, self.shape[K], ..., self.shape[P-1]].
```
For example, say we want to add 4 scattered elements to a rank-1 tensor to
8 elements. In Python, that update would look like this:
```python
v = tf.Variable([1, 2, 3, 4, 5, 6, 7, 8])
indices = tf.constant([[4], [3], [1] ,[7]])
updates = tf.constant([9, 10, 11, 12])
op = v.scatter_nd_sub(indices, updates)
with tf.compat.v1.Session() as sess:
print sess.run(op)
```
The resulting update to v would look like this:
[1, -9, 3, -6, -6, 6, 7, -4]
See `tf.scatter_nd` for more details about how to make updates to
slices.
Args:
indices: The indices to be used in the operation.
updates: The values to be used in the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered subtraction has completed.
"""
raise NotImplementedError
def scatter_nd_add(self, indices, updates, name=None):
"""Applies sparse addition to individual values or slices in a Variable.
The Variable has rank `P` and `indices` is a `Tensor` of rank `Q`.
`indices` must be integer tensor, containing indices into self.
It must be shape `[d_0, ..., d_{Q-2}, K]` where `0 < K <= P`.
The innermost dimension of `indices` (with length `K`) corresponds to
indices into elements (if `K = P`) or slices (if `K < P`) along the `K`th
dimension of self.
`updates` is `Tensor` of rank `Q-1+P-K` with shape:
```
[d_0, ..., d_{Q-2}, self.shape[K], ..., self.shape[P-1]].
```
For example, say we want to add 4 scattered elements to a rank-1 tensor to
8 elements. In Python, that update would look like this:
```python
v = tf.Variable([1, 2, 3, 4, 5, 6, 7, 8])
indices = tf.constant([[4], [3], [1] ,[7]])
updates = tf.constant([9, 10, 11, 12])
add = v.scatter_nd_add(indices, updates)
with tf.compat.v1.Session() as sess:
print sess.run(add)
```
The resulting update to v would look like this:
[1, 13, 3, 14, 14, 6, 7, 20]
See `tf.scatter_nd` for more details about how to make updates to
slices.
Args:
indices: The indices to be used in the operation.
updates: The values to be used in the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered addition has completed.
"""
raise NotImplementedError
def scatter_nd_update(self, indices, updates, name=None):
"""Applies sparse assignment to individual values or slices in a Variable.
The Variable has rank `P` and `indices` is a `Tensor` of rank `Q`.
`indices` must be integer tensor, containing indices into self.
It must be shape `[d_0, ..., d_{Q-2}, K]` where `0 < K <= P`.
The innermost dimension of `indices` (with length `K`) corresponds to
indices into elements (if `K = P`) or slices (if `K < P`) along the `K`th
dimension of self.
`updates` is `Tensor` of rank `Q-1+P-K` with shape:
```
[d_0, ..., d_{Q-2}, self.shape[K], ..., self.shape[P-1]].
```
For example, say we want to add 4 scattered elements to a rank-1 tensor to
8 elements. In Python, that update would look like this:
```python
v = tf.Variable([1, 2, 3, 4, 5, 6, 7, 8])
indices = tf.constant([[4], [3], [1] ,[7]])
updates = tf.constant([9, 10, 11, 12])
op = v.scatter_nd_assign(indices, updates)
with tf.compat.v1.Session() as sess:
print sess.run(op)
```
The resulting update to v would look like this:
[1, 11, 3, 10, 9, 6, 7, 12]
See `tf.scatter_nd` for more details about how to make updates to
slices.
Args:
indices: The indices to be used in the operation.
updates: The values to be used in the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered assignment has completed.
"""
raise NotImplementedError
def sparse_read(self, indices, name=None):
r"""Gather slices from params axis axis according to indices.
This function supports a subset of tf.gather, see tf.gather for details on
usage.
Args:
indices: The index `Tensor`. Must be one of the following types: `int32`,
`int64`. Must be in range `[0, params.shape[axis])`.
name: A name for the operation (optional).
Returns:
A `Tensor`. Has the same type as `params`.
"""
raise AttributeError
def gather_nd(self, indices, name=None):
r"""Gather slices from `params` into a Tensor with shape specified by `indices`.
See tf.gather_nd for details.
Args:
indices: A `Tensor`. Must be one of the following types: `int32`, `int64`.
Index tensor.
name: A name for the operation (optional).
Returns:
A `Tensor`. Has the same type as `params`.
"""
raise AttributeError
@deprecated(None, "Prefer Dataset.range instead.")
def count_up_to(self, limit):
"""Increments this variable until it reaches `limit`.
When that Op is run it tries to increment the variable by `1`. If
incrementing the variable would bring it above `limit` then the Op raises
the exception `OutOfRangeError`.
If no error is raised, the Op outputs the value of the variable before
the increment.
This is essentially a shortcut for `count_up_to(self, limit)`.
Args:
limit: value at which incrementing the variable raises an error.
Returns:
A `Tensor` that will hold the variable value before the increment. If no
other Op modifies this variable, the values produced will all be
distinct.
"""
raise NotImplementedError
@deprecated(
None,
"Prefer Variable.assign which has equivalent behavior in 2.X.")
def load(self, value, session=None):
"""Load new value into this variable.
Writes new value to variable's memory. Doesn't add ops to the graph.
This convenience method requires a session where the graph
containing this variable has been launched. If no session is
passed, the default session is used. See `tf.compat.v1.Session` for more
information on launching a graph and on sessions.
```python
v = tf.Variable([1, 2])
init = tf.compat.v1.global_variables_initializer()
with tf.compat.v1.Session() as sess:
sess.run(init)
# Usage passing the session explicitly.
v.load([2, 3], sess)
print(v.eval(sess)) # prints [2 3]
# Usage with the default session. The 'with' block
# above makes 'sess' the default session.
v.load([3, 4], sess)
print(v.eval()) # prints [3 4]
```
Args:
value: New variable value
session: The session to use to evaluate this variable. If
none, the default session is used.
Raises:
ValueError: Session is not passed and no default session
"""
if context.executing_eagerly():
self.assign(value)
else:
session = session or ops.get_default_session()
if session is None:
raise ValueError(
"Either session argument should be provided or default session "
"should be established")
session.run(self.initializer, {self.initializer.inputs[1]: value})
# Conversion to tensor.
@staticmethod
def _TensorConversionFunction(v, dtype=None, name=None, as_ref=False): # pylint: disable=invalid-name
"""Utility function for converting a Variable to a Tensor."""
_ = name
if dtype and not dtype.is_compatible_with(v.dtype):
raise ValueError(
"Incompatible type conversion requested to type '%s' for variable "
"of type '%s'" % (dtype.name, v.dtype.name))
if as_ref:
return v._ref() # pylint: disable=protected-access
else:
return v.value()
@classmethod
def _OverloadAllOperators(cls): # pylint: disable=invalid-name
"""Register overloads for all operators."""
for operator in ops.Tensor.OVERLOADABLE_OPERATORS:
cls._OverloadOperator(operator)
# For slicing, bind getitem differently than a tensor (use SliceHelperVar
# instead)
# pylint: disable=protected-access
setattr(cls, "__getitem__", array_ops._SliceHelperVar)
@classmethod
def _OverloadOperator(cls, operator): # pylint: disable=invalid-name
"""Defer an operator overload to `ops.Tensor`.
We pull the operator out of ops.Tensor dynamically to avoid ordering issues.
Args:
operator: string. The operator name.
"""
tensor_oper = getattr(ops.Tensor, operator)
def _run_op(a, *args, **kwargs):
# pylint: disable=protected-access
return tensor_oper(a.value(), *args, **kwargs)
functools.update_wrapper(_run_op, tensor_oper)
setattr(cls, operator, _run_op)
def __iter__(self):
"""Dummy method to prevent iteration. Do not call.
NOTE(mrry): If we register __getitem__ as an overloaded operator,
Python will valiantly attempt to iterate over the variable's Tensor from 0
to infinity. Declaring this method prevents this unintended behavior.
Raises:
TypeError: when invoked.
"""
raise TypeError("'Variable' object is not iterable.")
# NOTE(mrry): This enables the Variable's overloaded "right" binary
# operators to run when the left operand is an ndarray, because it
# accords the Variable class higher priority than an ndarray, or a
# numpy matrix.
# TODO(mrry): Convert this to using numpy's __numpy_ufunc__
# mechanism, which allows more control over how Variables interact
# with ndarrays.
__array_priority__ = 100
@property
def name(self):
"""The name of this variable."""
raise NotImplementedError
@property
def _shared_name(self):
"""The shared name of the variable.
Unlike name(), shared_name doesn't have ":0" suffix. It is user-specified
name with name scope prefix.
Returns:
variable name.
"""
return self.name[:self.name.index(":")]
@property
def initializer(self):
"""The initializer operation for this variable."""
raise NotImplementedError
@property
def device(self):
"""The device of this variable."""
raise NotImplementedError
@property
def dtype(self):
"""The `DType` of this variable."""
raise NotImplementedError
@property
def op(self):
"""The `Operation` of this variable."""
raise NotImplementedError
@property
def graph(self):
"""The `Graph` of this variable."""
raise NotImplementedError
@property
def shape(self):
"""The `TensorShape` of this variable.
Returns:
A `TensorShape`.
"""
raise NotImplementedError
def get_shape(self):
"""Alias of `Variable.shape`."""
return self.shape
def _gather_saveables_for_checkpoint(self):
"""For implementing `Trackable`. This object is saveable on its own."""
return {trackable.VARIABLE_VALUE_KEY: self}
def to_proto(self, export_scope=None):
"""Converts a `Variable` to a `VariableDef` protocol buffer.
Args:
export_scope: Optional `string`. Name scope to remove.
Returns:
A `VariableDef` protocol buffer, or `None` if the `Variable` is not
in the specified name scope.
"""
raise NotImplementedError
@staticmethod
def from_proto(variable_def, import_scope=None):
"""Returns a `Variable` object created from `variable_def`."""
return RefVariable(variable_def=variable_def,
import_scope=import_scope)
def _set_save_slice_info(self, save_slice_info):
"""Sets the slice info for this `Variable`.
Args:
save_slice_info: A `Variable.SaveSliceInfo` object.
"""
self._save_slice_info = save_slice_info
def _get_save_slice_info(self):
return self._save_slice_info
class SaveSliceInfo(object):
"""Information on how to save this Variable as a slice.
Provides internal support for saving variables as slices of a larger
variable. This API is not public and is subject to change.
Available properties:
* full_name
* full_shape
* var_offset
* var_shape
"""
def __init__(self,
full_name=None,
full_shape=None,
var_offset=None,
var_shape=None,
save_slice_info_def=None,
import_scope=None):
"""Create a `SaveSliceInfo`.
Args:
full_name: Name of the full variable of which this `Variable` is a
slice.
full_shape: Shape of the full variable, as a list of int.
var_offset: Offset of this `Variable` into the full variable, as a
list of int.
var_shape: Shape of this `Variable`, as a list of int.
save_slice_info_def: `SaveSliceInfoDef` protocol buffer. If not `None`,
recreates the SaveSliceInfo object its contents.
`save_slice_info_def` and other arguments are mutually
exclusive.
import_scope: Optional `string`. Name scope to add. Only used
when initializing from protocol buffer.
"""
if save_slice_info_def:
assert isinstance(save_slice_info_def, variable_pb2.SaveSliceInfoDef)
self.full_name = ops.prepend_name_scope(
save_slice_info_def.full_name, import_scope=import_scope)
self.full_shape = [i for i in save_slice_info_def.full_shape]
self.var_offset = [i for i in save_slice_info_def.var_offset]
self.var_shape = [i for i in save_slice_info_def.var_shape]
else:
self.full_name = full_name
self.full_shape = full_shape
self.var_offset = var_offset
self.var_shape = var_shape
@property
def spec(self):
"""Computes the spec string used for saving."""
full_shape_str = " ".join(["%d" % d for d in self.full_shape]) + " "
sl_spec = ":".join([
"%d,%d" % (o, s) for o, s in zip(self.var_offset, self.var_shape)
])
return full_shape_str + sl_spec
def to_proto(self, export_scope=None):
"""Returns a SaveSliceInfoDef() proto.
Args:
export_scope: Optional `string`. Name scope to remove.
Returns:
A `SaveSliceInfoDef` protocol buffer, or None if the `Variable` is not
in the specified name scope.
"""
if (export_scope is None or
self.full_name.startswith(export_scope)):
save_slice_info_def = variable_pb2.SaveSliceInfoDef()
save_slice_info_def.full_name = ops.strip_name_scope(
self.full_name, export_scope)
for i in self.full_shape:
save_slice_info_def.full_shape.append(i)
for i in self.var_offset:
save_slice_info_def.var_offset.append(i)
for i in self.var_shape:
save_slice_info_def.var_shape.append(i)
return save_slice_info_def
else:
return None
Variable._OverloadAllOperators() # pylint: disable=protected-access
@tf_export(v1=["Variable"])
class VariableV1(Variable):
"""See the [Variables Guide](https://tensorflow.org/guide/variables).
A variable maintains state in the graph across calls to `run()`. You add a
variable to the graph by constructing an instance of the class `Variable`.
The `Variable()` constructor requires an initial value for the variable,
which can be a `Tensor` of any type and shape. The initial value defines the
type and shape of the variable. After construction, the type and shape of
the variable are fixed. The value can be changed using one of the assign
methods.
If you want to change the shape of a variable later you have to use an
`assign` Op with `validate_shape=False`.
Just like any `Tensor`, variables created with `Variable()` can be used as
inputs for other Ops in the graph. Additionally, all the operators
overloaded for the `Tensor` class are carried over to variables, so you can
also add nodes to the graph by just doing arithmetic on variables.
```python
import tensorflow as tf
# Create a variable.
w = tf.Variable(<initial-value>, name=<optional-name>)
# Use the variable in the graph like any Tensor.
y = tf.matmul(w, ...another variable or tensor...)
# The overloaded operators are available too.
z = tf.sigmoid(w + y)
# Assign a new value to the variable with `assign()` or a related method.
w.assign(w + 1.0)
w.assign_add(1.0)
```
When you launch the graph, variables have to be explicitly initialized before
you can run Ops that use their value. You can initialize a variable by
running its *initializer op*, restoring the variable from a save file, or
simply running an `assign` Op that assigns a value to the variable. In fact,
the variable *initializer op* is just an `assign` Op that assigns the
variable's initial value to the variable itself.
```python
# Launch the graph in a session.
with tf.compat.v1.Session() as sess:
# Run the variable initializer.
sess.run(w.initializer)
# ...you now can run ops that use the value of 'w'...
```
The most common initialization pattern is to use the convenience function
`global_variables_initializer()` to add an Op to the graph that initializes
all the variables. You then run that Op after launching the graph.
```python
# Add an Op to initialize global variables.
init_op = tf.compat.v1.global_variables_initializer()
# Launch the graph in a session.
with tf.compat.v1.Session() as sess:
# Run the Op that initializes global variables.
sess.run(init_op)
# ...you can now run any Op that uses variable values...
```
If you need to create a variable with an initial value dependent on another
variable, use the other variable's `initialized_value()`. This ensures that
variables are initialized in the right order.
All variables are automatically collected in the graph where they are
created. By default, the constructor adds the new variable to the graph
collection `GraphKeys.GLOBAL_VARIABLES`. The convenience function
`global_variables()` returns the contents of that collection.
When building a machine learning model it is often convenient to distinguish
between variables holding the trainable model parameters and other variables
such as a `global step` variable used to count training steps. To make this
easier, the variable constructor supports a `trainable=<bool>` parameter. If
`True`, the new variable is also added to the graph collection
`GraphKeys.TRAINABLE_VARIABLES`. The convenience function
`trainable_variables()` returns the contents of this collection. The
various `Optimizer` classes use this collection as the default list of
variables to optimize.
WARNING: tf.Variable objects by default have a non-intuitive memory model. A
Variable is represented internally as a mutable Tensor which can
non-deterministically alias other Tensors in a graph. The set of operations
which consume a Variable and can lead to aliasing is undetermined and can
change across TensorFlow versions. Avoid writing code which relies on the
value of a Variable either changing or not changing as other operations
happen. For example, using Variable objects or simple functions thereof as
predicates in a `tf.cond` is dangerous and error-prone:
```
v = tf.Variable(True)
tf.cond(v, lambda: v.assign(False), my_false_fn) # Note: this is broken.
```
Here, adding `use_resource=True` when constructing the variable will
fix any nondeterminism issues:
```
v = tf.Variable(True, use_resource=True)
tf.cond(v, lambda: v.assign(False), my_false_fn)
```
To use the replacement for variables which does
not have these issues:
* Add `use_resource=True` when constructing `tf.Variable`;
* Call `tf.compat.v1.get_variable_scope().set_use_resource(True)` inside a
`tf.compat.v1.variable_scope` before the `tf.compat.v1.get_variable()` call.
"""
def __init__(self, # pylint: disable=super-init-not-called
initial_value=None,
trainable=None,
collections=None,
validate_shape=True,
caching_device=None,
name=None,
variable_def=None,
dtype=None,
expected_shape=None,
import_scope=None,
constraint=None,
use_resource=None,
synchronization=VariableSynchronization.AUTO,
aggregation=VariableAggregation.NONE,
shape=None):
"""Creates a new variable with value `initial_value`.
The new variable is added to the graph collections listed in `collections`,
which defaults to `[GraphKeys.GLOBAL_VARIABLES]`.
If `trainable` is `True` the variable is also added to the graph collection
`GraphKeys.TRAINABLE_VARIABLES`.
This constructor creates both a `variable` Op and an `assign` Op to set the
variable to its initial value.
Args:
initial_value: A `Tensor`, or Python object convertible to a `Tensor`,
which is the initial value for the Variable. The initial value must have
a shape specified unless `validate_shape` is set to False. Can also be a
callable with no argument that returns the initial value when called. In
that case, `dtype` must be specified. (Note that initializer functions
from init_ops.py must first be bound to a shape before being used here.)
trainable: If `True`, also adds the variable to the graph
collection `GraphKeys.TRAINABLE_VARIABLES`. This collection is used as
the default list of variables to use by the `Optimizer` classes.
Defaults to `True`, unless `synchronization` is set to `ON_READ`, in
which case it defaults to `False`.
collections: List of graph collections keys. The new variable is added to
these collections. Defaults to `[GraphKeys.GLOBAL_VARIABLES]`.
validate_shape: If `False`, allows the variable to be initialized with a
value of unknown shape. If `True`, the default, the shape of
`initial_value` must be known.
caching_device: Optional device string describing where the Variable
should be cached for reading. Defaults to the Variable's device.
If not `None`, caches on another device. Typical use is to cache
on the device where the Ops using the Variable reside, to deduplicate
copying through `Switch` and other conditional statements.
name: Optional name for the variable. Defaults to `'Variable'` and gets
uniquified automatically.
variable_def: `VariableDef` protocol buffer. If not `None`, recreates
the Variable object with its contents, referencing the variable's nodes
in the graph, which must already exist. The graph is not changed.
`variable_def` and the other arguments are mutually exclusive.
dtype: If set, initial_value will be converted to the given type.
If `None`, either the datatype will be kept (if `initial_value` is
a Tensor), or `convert_to_tensor` will decide.
expected_shape: A TensorShape. If set, initial_value is expected
to have this shape.
import_scope: Optional `string`. Name scope to add to the
`Variable.` Only used when initializing from protocol buffer.
constraint: An optional projection function to be applied to the variable
after being updated by an `Optimizer` (e.g. used to implement norm
constraints or value constraints for layer weights). The function must
take as input the unprojected Tensor representing the value of the
variable and return the Tensor for the projected value
(which must have the same shape). Constraints are not safe to
use when doing asynchronous distributed training.
use_resource: whether to use resource variables.
synchronization: Indicates when a distributed a variable will be
aggregated. Accepted values are constants defined in the class
`tf.VariableSynchronization`. By default the synchronization is set to
`AUTO` and the current `DistributionStrategy` chooses
when to synchronize.
aggregation: Indicates how a distributed variable will be aggregated.
Accepted values are constants defined in the class
`tf.VariableAggregation`.
shape: (optional) The shape of this variable. If None, the shape of
`initial_value` will be used. When setting this argument to
`tf.TensorShape(None)` (representing an unspecified shape), the variable
can be assigned with values of different shapes.
Raises:
ValueError: If both `variable_def` and initial_value are specified.
ValueError: If the initial value is not specified, or does not have a
shape and `validate_shape` is `True`.
RuntimeError: If eager execution is enabled.
"""
SaveSliceInfo = Variable.SaveSliceInfo
# TODO(apassos): do not repeat all comments here
class RefVariable(VariableV1):
"""Ref-based implementation of variables."""
def __init__(self, # pylint: disable=super-init-not-called
initial_value=None,
trainable=None,
collections=None,
validate_shape=True,
caching_device=None,
name=None,
variable_def=None,
dtype=None,
expected_shape=None,
import_scope=None,
constraint=None,
synchronization=None,
aggregation=None,
shape=None):
"""Creates a new variable with value `initial_value`.
The new variable is added to the graph collections listed in `collections`,
which defaults to `[GraphKeys.GLOBAL_VARIABLES]`.
If `trainable` is `True` the variable is also added to the graph collection
`GraphKeys.TRAINABLE_VARIABLES`.
This constructor creates both a `variable` Op and an `assign` Op to set the
variable to its initial value.
Args:
initial_value: A `Tensor`, or Python object convertible to a `Tensor`,
which is the initial value for the Variable. The initial value must have
a shape specified unless `validate_shape` is set to False. Can also be a
callable with no argument that returns the initial value when called. In
that case, `dtype` must be specified. (Note that initializer functions
from init_ops.py must first be bound to a shape before being used here.)
trainable: If `True`, also adds the variable to the graph
collection `GraphKeys.TRAINABLE_VARIABLES`. This collection is used as
the default list of variables to use by the `Optimizer` classes.
Defaults to `True`, unless `synchronization` is set to `ON_READ`, in
which case it defaults to `False`.
collections: List of graph collections keys. The new variable is added to
these collections. Defaults to `[GraphKeys.GLOBAL_VARIABLES]`.
validate_shape: If `False`, allows the variable to be initialized with a
value of unknown shape. If `True`, the default, the shape of
`initial_value` must be known.
caching_device: Optional device string describing where the Variable
should be cached for reading. Defaults to the Variable's device.
If not `None`, caches on another device. Typical use is to cache
on the device where the Ops using the Variable reside, to deduplicate
copying through `Switch` and other conditional statements.
name: Optional name for the variable. Defaults to `'Variable'` and gets
uniquified automatically.
variable_def: `VariableDef` protocol buffer. If not `None`, recreates
the Variable object with its contents, referencing the variable's nodes
in the graph, which must already exist. The graph is not changed.
`variable_def` and the other arguments are mutually exclusive.
dtype: If set, initial_value will be converted to the given type.
If `None`, either the datatype will be kept (if `initial_value` is
a Tensor), or `convert_to_tensor` will decide.
expected_shape: A TensorShape. If set, initial_value is expected
to have this shape.
import_scope: Optional `string`. Name scope to add to the
`Variable.` Only used when initializing from protocol buffer.
constraint: An optional projection function to be applied to the variable
after being updated by an `Optimizer` (e.g. used to implement norm
constraints or value constraints for layer weights). The function must
take as input the unprojected Tensor representing the value of the
variable and return the Tensor for the projected value
(which must have the same shape). Constraints are not safe to
use when doing asynchronous distributed training.
synchronization: Indicates when a distributed a variable will be
aggregated. Accepted values are constants defined in the class
`tf.VariableSynchronization`. By default the synchronization is set to
`AUTO` and the current `DistributionStrategy` chooses
when to synchronize.
aggregation: Indicates how a distributed variable will be aggregated.
Accepted values are constants defined in the class
`tf.VariableAggregation`.
shape: (optional) The shape of this variable. If None, the shape of
`initial_value` will be used. When setting this argument to
`tf.TensorShape(None)` (representing an unspecified shape), the variable
can be assigned with values of different shapes.
Raises:
ValueError: If both `variable_def` and initial_value are specified.
ValueError: If the initial value is not specified, or does not have a
shape and `validate_shape` is `True`.
RuntimeError: If eager execution is enabled.
"""
self._in_graph_mode = True
if variable_def:
# If variable_def is provided, recreates the variable from its fields.
if initial_value:
raise ValueError("variable_def and initial_value are mutually "
"exclusive.")
self._init_from_proto(variable_def, import_scope=import_scope)
else:
# Create from initial_value.
self._init_from_args(
initial_value=initial_value,
trainable=trainable,
collections=collections,
validate_shape=validate_shape,
caching_device=caching_device,
name=name,
dtype=dtype,
expected_shape=expected_shape,
constraint=constraint,
synchronization=synchronization,
aggregation=aggregation,
shape=shape)
def __repr__(self):
if context.executing_eagerly() and not self._in_graph_mode:
return "<tf.Variable '%s' shape=%s dtype=%s, numpy=%s>" % (
self.name, self.get_shape(), self.dtype.name,
ops.numpy_text(self.read_value(), is_repr=True))
else:
return "<tf.Variable '%s' shape=%s dtype=%s>" % (
self.name, self.get_shape(), self.dtype.name)
def _init_from_args(self,
initial_value=None,
trainable=None,
collections=None,
validate_shape=True,
caching_device=None,
name=None,
dtype=None,
expected_shape=None,
constraint=None,
synchronization=None,
aggregation=None,
shape=None):
"""Creates a new variable from arguments.
Args:
initial_value: A `Tensor`, or Python object convertible to a `Tensor`,
which is the initial value for the Variable. The initial value must have
a shape specified unless `validate_shape` is set to False. Can also be a
callable with no argument that returns the initial value when called.
(Note that initializer functions from init_ops.py must first be bound
to a shape before being used here.)
trainable: If `True`, also adds the variable to the graph
collection `GraphKeys.TRAINABLE_VARIABLES`. This collection is used as
the default list of variables to use by the `Optimizer` classes.
Defaults to `True`, unless `synchronization` is set to `ON_READ`, in
which case it defaults to `False`.
collections: List of graph collections keys. The new variable is added to
these collections. Defaults to `[GraphKeys.GLOBAL_VARIABLES]`.
validate_shape: If `False`, allows the variable to be initialized with a
value of unknown shape. If `True`, the default, the shape of
`initial_value` must be known.
caching_device: Optional device string or function describing where the
Variable should be cached for reading. Defaults to the Variable's
device. If not `None`, caches on another device. Typical use is to
cache on the device where the Ops using the Variable reside, to
deduplicate copying through `Switch` and other conditional statements.
name: Optional name for the variable. Defaults to `'Variable'` and gets
uniquified automatically.
dtype: If set, initial_value will be converted to the given type.
If None, either the datatype will be kept (if initial_value is
a Tensor) or float32 will be used (if it is a Python object convertible
to a Tensor).
expected_shape: Deprecated. Ignored.
constraint: An optional projection function to be applied to the variable
after being updated by an `Optimizer` (e.g. used to implement norm
constraints or value constraints for layer weights). The function must
take as input the unprojected Tensor representing the value of the
variable and return the Tensor for the projected value
(which must have the same shape). Constraints are not safe to
use when doing asynchronous distributed training.
synchronization: Indicates when a distributed a variable will be
aggregated. Accepted values are constants defined in the class
`tf.VariableSynchronization`. By default the synchronization is set to
`AUTO` and the current `DistributionStrategy` chooses
when to synchronize.
aggregation: Indicates how a distributed variable will be aggregated.
Accepted values are constants defined in the class
`tf.VariableAggregation`.
shape: (optional) The shape of this variable. If None, the shape of
`initial_value` will be used. When setting this argument to
`tf.TensorShape(None)` (representing an unspecified shape), the variable
can be assigned with values of different shapes.
Raises:
ValueError: If the initial value is not specified, or does not have a
shape and `validate_shape` is `True`.
RuntimeError: If lifted into the eager context.
"""
_ = expected_shape
if initial_value is None:
raise ValueError("initial_value must be specified.")
init_from_fn = callable(initial_value)
if collections is None:
collections = [ops.GraphKeys.GLOBAL_VARIABLES]
if not isinstance(collections, (list, tuple, set)):
raise ValueError(
"collections argument to Variable constructor must be a list, tuple, "
"or set. Got %s of type %s" % (collections, type(collections)))
if constraint is not None and not callable(constraint):
raise ValueError("The `constraint` argument must be a callable.")
# Store the graph key so optimizers know how to only retrieve variables from
# this graph.
self._graph_key = ops.get_default_graph()._graph_key # pylint: disable=protected-access
if isinstance(initial_value, trackable.CheckpointInitialValue):
self._maybe_initialize_trackable()
self._update_uid = initial_value.checkpoint_position.restore_uid
initial_value = initial_value.wrapped_value
synchronization, aggregation, trainable = (
validate_synchronization_aggregation_trainable(
synchronization, aggregation, trainable, name))
self._synchronization = synchronization
self._aggregation = aggregation
self._trainable = trainable
if trainable and ops.GraphKeys.TRAINABLE_VARIABLES not in collections:
collections = list(collections) + [ops.GraphKeys.TRAINABLE_VARIABLES]
with ops.init_scope():
# Ensure that we weren't lifted into the eager context.
if context.executing_eagerly():
raise RuntimeError(
"RefVariable not supported when eager execution is enabled. ")
with ops.name_scope(name, "Variable", [] if init_from_fn else
[initial_value]) as name:
if init_from_fn:
# Use attr_scope and device(None) to simulate the behavior of
# colocate_with when the variable we want to colocate with doesn't
# yet exist.
true_name = ops.name_from_scope_name(name) # pylint: disable=protected-access
attr = attr_value_pb2.AttrValue(
list=attr_value_pb2.AttrValue.ListValue(
s=[compat.as_bytes("loc:@%s" % true_name)]))
# pylint: disable=protected-access
with ops.get_default_graph()._attr_scope({"_class": attr}):
with ops.name_scope("Initializer"), ops.device(None):
self._initial_value = ops.convert_to_tensor(
initial_value(), name="initial_value", dtype=dtype)
if shape is None:
shape = (self._initial_value.get_shape()
if validate_shape else tensor_shape.unknown_shape())
self._variable = state_ops.variable_op_v2(
shape,
self._initial_value.dtype.base_dtype,
name=name)
# pylint: enable=protected-access
# Or get the initial value from a Tensor or Python object.
else:
self._initial_value = ops.convert_to_tensor(
initial_value, name="initial_value", dtype=dtype)
# pylint: disable=protected-access
if self._initial_value.op._get_control_flow_context() is not None:
raise ValueError(
"Initializer for variable %s is from inside a control-flow "
"construct, such as a loop or conditional. When creating a "
"variable inside a loop or conditional, use a lambda as the "
"initializer." % name)
if shape is None:
# pylint: enable=protected-access
shape = (self._initial_value.get_shape()
if validate_shape else tensor_shape.unknown_shape())
# In this case, the variable op can't be created until after the
# initial_value has been converted to a Tensor with a known type.
self._variable = state_ops.variable_op_v2(
shape,
self._initial_value.dtype.base_dtype,
name=name)
# Manually overrides the variable's shape with the initial value's.
if validate_shape:
initial_value_shape = self._initial_value.get_shape()
if not initial_value_shape.is_fully_defined():
raise ValueError("initial_value must have a shape specified: %s" %
self._initial_value)
# If 'initial_value' makes use of other variables, make sure we don't
# have an issue if these other variables aren't initialized first by
# using their initialized_value() method.
self._initializer_op = state_ops.assign(
self._variable,
_try_guard_against_uninitialized_dependencies(
name,
self._initial_value),
validate_shape=validate_shape).op
# TODO(vrv): Change this class to not take caching_device, but
# to take the op to colocate the snapshot with, so we can use
# colocation rather than devices.
if caching_device is not None:
with ops.device(caching_device):
self._snapshot = array_ops.identity(self._variable, name="read")
else:
with ops.colocate_with(self._variable.op):
self._snapshot = array_ops.identity(self._variable, name="read")
ops.add_to_collections(collections, self)
self._caching_device = caching_device
self._save_slice_info = None
self._constraint = constraint
def _init_from_proto(self, variable_def, import_scope=None):
"""Recreates the Variable object from a `VariableDef` protocol buffer.
Args:
variable_def: `VariableDef` protocol buffer, describing a variable
whose nodes already exists in the graph.
import_scope: Optional `string`. Name scope to add.
"""
assert isinstance(variable_def, variable_pb2.VariableDef)
# Create from variable_def.
g = ops.get_default_graph()
self._variable = g.as_graph_element(
ops.prepend_name_scope(variable_def.variable_name,
import_scope=import_scope))
self._initializer_op = g.as_graph_element(
ops.prepend_name_scope(variable_def.initializer_name,
import_scope=import_scope))
# Tests whether initial_value_name exists first for backwards compatibility.
if (hasattr(variable_def, "initial_value_name") and
variable_def.initial_value_name):
self._initial_value = g.as_graph_element(
ops.prepend_name_scope(variable_def.initial_value_name,
import_scope=import_scope))
else:
self._initial_value = None
synchronization, aggregation, trainable = (
validate_synchronization_aggregation_trainable(
variable_def.synchronization,
variable_def.aggregation,
variable_def.trainable,
variable_def.variable_name))
self._synchronization = synchronization
self._aggregation = aggregation
self._trainable = trainable
self._snapshot = g.as_graph_element(
ops.prepend_name_scope(variable_def.snapshot_name,
import_scope=import_scope))
if variable_def.HasField("save_slice_info_def"):
self._save_slice_info = Variable.SaveSliceInfo(
save_slice_info_def=variable_def.save_slice_info_def,
import_scope=import_scope)
else:
self._save_slice_info = None
self._caching_device = None
self._constraint = None
def _as_graph_element(self):
"""Conversion function for Graph.as_graph_element()."""
return self._variable
def value(self):
"""Returns the last snapshot of this variable.
You usually do not need to call this method as all ops that need the value
of the variable call it automatically through a `convert_to_tensor()` call.
Returns a `Tensor` which holds the value of the variable. You can not
assign a new value to this tensor as it is not a reference to the variable.
To avoid copies, if the consumer of the returned value is on the same device
as the variable, this actually returns the live value of the variable, not
a copy. Updates to the variable are seen by the consumer. If the consumer
is on a different device it will get a copy of the variable.
Returns:
A `Tensor` containing the value of the variable.
"""
return self._snapshot
def read_value(self):
"""Returns the value of this variable, read in the current context.
Can be different from value() if it's on another device, with control
dependencies, etc.
Returns:
A `Tensor` containing the value of the variable.
"""
return array_ops.identity(self._variable, name="read")
def _ref(self):
"""Returns a reference to this variable.
You usually do not need to call this method as all ops that need a reference
to the variable call it automatically.
Returns is a `Tensor` which holds a reference to the variable. You can
assign a new value to the variable by passing the tensor to an assign op.
See `tf.Variable.value` if you want to get the value of the
variable.
Returns:
A `Tensor` that is a reference to the variable.
"""
return self._variable
def set_shape(self, shape):
"""Overrides the shape for this variable.
Args:
shape: the `TensorShape` representing the overridden shape.
"""
self._ref().set_shape(shape)
self.value().set_shape(shape)
@property
def trainable(self):
return self._trainable
@property
def synchronization(self):
return self._synchronization
@property
def aggregation(self):
return self._aggregation
def eval(self, session=None):
"""In a session, computes and returns the value of this variable.
This is not a graph construction method, it does not add ops to the graph.
This convenience method requires a session where the graph
containing this variable has been launched. If no session is
passed, the default session is used. See `tf.compat.v1.Session` for more
information on launching a graph and on sessions.
```python
v = tf.Variable([1, 2])
init = tf.compat.v1.global_variables_initializer()
with tf.compat.v1.Session() as sess:
sess.run(init)
# Usage passing the session explicitly.
print(v.eval(sess))
# Usage with the default session. The 'with' block
# above makes 'sess' the default session.
print(v.eval())
```
Args:
session: The session to use to evaluate this variable. If
none, the default session is used.
Returns:
A numpy `ndarray` with a copy of the value of this variable.
"""
return self._variable.eval(session=session)
@property
def initial_value(self):
"""Returns the Tensor used as the initial value for the variable.
Note that this is different from `initialized_value()` which runs
the op that initializes the variable before returning its value.
This method returns the tensor that is used by the op that initializes
the variable.
Returns:
A `Tensor`.
"""
return self._initial_value
@property
def constraint(self):
"""Returns the constraint function associated with this variable.
Returns:
The constraint function that was passed to the variable constructor.
Can be `None` if no constraint was passed.
"""
return self._constraint
def assign(self, value, use_locking=False, name=None, read_value=True):
"""Assigns a new value to the variable.
This is essentially a shortcut for `assign(self, value)`.
Args:
value: A `Tensor`. The new value for this variable.
use_locking: If `True`, use locking during the assignment.
name: The name of the operation to be created
read_value: if True, will return something which evaluates to the
new value of the variable; if False will return the assign op.
Returns:
A `Tensor` that will hold the new value of this variable after
the assignment has completed.
"""
assign = state_ops.assign(self._variable, value, use_locking=use_locking,
name=name)
if read_value:
return assign
return assign.op
def assign_add(self, delta, use_locking=False, name=None, read_value=True):
"""Adds a value to this variable.
This is essentially a shortcut for `assign_add(self, delta)`.
Args:
delta: A `Tensor`. The value to add to this variable.
use_locking: If `True`, use locking during the operation.
name: The name of the operation to be created
read_value: if True, will return something which evaluates to the
new value of the variable; if False will return the assign op.
Returns:
A `Tensor` that will hold the new value of this variable after
the addition has completed.
"""
assign = state_ops.assign_add(
self._variable, delta, use_locking=use_locking, name=name)
if read_value:
return assign
return assign.op
def assign_sub(self, delta, use_locking=False, name=None, read_value=True):
"""Subtracts a value from this variable.
This is essentially a shortcut for `assign_sub(self, delta)`.
Args:
delta: A `Tensor`. The value to subtract from this variable.
use_locking: If `True`, use locking during the operation.
name: The name of the operation to be created
read_value: if True, will return something which evaluates to the
new value of the variable; if False will return the assign op.
Returns:
A `Tensor` that will hold the new value of this variable after
the subtraction has completed.
"""
assign = state_ops.assign_sub(
self._variable, delta, use_locking=use_locking, name=name)
if read_value:
return assign
return assign.op
def scatter_sub(self, sparse_delta, use_locking=False, name=None):
"""Subtracts `tf.IndexedSlices` from this variable.
Args:
sparse_delta: `tf.IndexedSlices` to be subtracted from this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered subtraction has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
if not isinstance(sparse_delta, ops.IndexedSlices):
raise TypeError("sparse_delta is not IndexedSlices: %s" % sparse_delta)
return gen_state_ops.scatter_sub(
self._variable,
sparse_delta.indices,
sparse_delta.values,
use_locking=use_locking,
name=name)
def scatter_add(self, sparse_delta, use_locking=False, name=None):
"""Adds `tf.IndexedSlices` to this variable.
Args:
sparse_delta: `tf.IndexedSlices` to be added to this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered addition has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
if not isinstance(sparse_delta, ops.IndexedSlices):
raise TypeError("sparse_delta is not IndexedSlices: %s" % sparse_delta)
return gen_state_ops.scatter_add(
self._variable,
sparse_delta.indices,
sparse_delta.values,
use_locking=use_locking,
name=name)
def scatter_max(self, sparse_delta, use_locking=False, name=None):
"""Updates this variable with the max of `tf.IndexedSlices` and itself.
Args:
sparse_delta: `tf.IndexedSlices` to use as an argument of max
with this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered maximization has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
if not isinstance(sparse_delta, ops.IndexedSlices):
raise TypeError("sparse_delta is not IndexedSlices: %s" % sparse_delta)
return gen_state_ops.scatter_max(
self._variable,
sparse_delta.indices,
sparse_delta.values,
use_locking=use_locking,
name=name)
def scatter_min(self, sparse_delta, use_locking=False, name=None):
"""Updates this variable with the min of `tf.IndexedSlices` and itself.
Args:
sparse_delta: `tf.IndexedSlices` to use as an argument of min
with this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered minimization has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
if not isinstance(sparse_delta, ops.IndexedSlices):
raise TypeError("sparse_delta is not IndexedSlices: %s" % sparse_delta)
return gen_state_ops.scatter_min(
self._variable,
sparse_delta.indices,
sparse_delta.values,
use_locking=use_locking,
name=name)
def scatter_mul(self, sparse_delta, use_locking=False, name=None):
"""Multiply this variable by `tf.IndexedSlices`.
Args:
sparse_delta: `tf.IndexedSlices` to multiply this variable by.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered multiplication has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
if not isinstance(sparse_delta, ops.IndexedSlices):
raise TypeError("sparse_delta is not IndexedSlices: %s" % sparse_delta)
return gen_state_ops.scatter_mul(
self._variable,
sparse_delta.indices,
sparse_delta.values,
use_locking=use_locking,
name=name)
def scatter_div(self, sparse_delta, use_locking=False, name=None):
"""Divide this variable by `tf.IndexedSlices`.
Args:
sparse_delta: `tf.IndexedSlices` to divide this variable by.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered division has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
if not isinstance(sparse_delta, ops.IndexedSlices):
raise TypeError("sparse_delta is not IndexedSlices: %s" % sparse_delta)
return gen_state_ops.scatter_div(
self._variable,
sparse_delta.indices,
sparse_delta.values,
use_locking=use_locking,
name=name)
def scatter_update(self, sparse_delta, use_locking=False, name=None):
"""Assigns `tf.IndexedSlices` to this variable.
Args:
sparse_delta: `tf.IndexedSlices` to be assigned to this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered assignment has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
if not isinstance(sparse_delta, ops.IndexedSlices):
raise TypeError("sparse_delta is not IndexedSlices: %s" % sparse_delta)
return gen_state_ops.scatter_update(
self._variable,
sparse_delta.indices,
sparse_delta.values,
use_locking=use_locking,
name=name)
def batch_scatter_update(self, sparse_delta, use_locking=False, name=None):
"""Assigns `tf.IndexedSlices` to this variable batch-wise.
Analogous to `batch_gather`. This assumes that this variable and the
sparse_delta IndexedSlices have a series of leading dimensions that are the
same for all of them, and the updates are performed on the last dimension of
indices. In other words, the dimensions should be the following:
`num_prefix_dims = sparse_delta.indices.ndims - 1`
`batch_dim = num_prefix_dims + 1`
`sparse_delta.updates.shape = sparse_delta.indices.shape + var.shape[
batch_dim:]`
where
`sparse_delta.updates.shape[:num_prefix_dims]`
`== sparse_delta.indices.shape[:num_prefix_dims]`
`== var.shape[:num_prefix_dims]`
And the operation performed can be expressed as:
`var[i_1, ..., i_n,
sparse_delta.indices[i_1, ..., i_n, j]] = sparse_delta.updates[
i_1, ..., i_n, j]`
When sparse_delta.indices is a 1D tensor, this operation is equivalent to
`scatter_update`.
To avoid this operation one can looping over the first `ndims` of the
variable and using `scatter_update` on the subtensors that result of slicing
the first dimension. This is a valid option for `ndims = 1`, but less
efficient than this implementation.
Args:
sparse_delta: `tf.IndexedSlices` to be assigned to this variable.
use_locking: If `True`, use locking during the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered assignment has completed.
Raises:
TypeError: if `sparse_delta` is not an `IndexedSlices`.
"""
return state_ops.batch_scatter_update(
self, sparse_delta.indices, sparse_delta.values,
use_locking=use_locking, name=name)
def scatter_nd_sub(self, indices, updates, name=None):
"""Applies sparse subtraction to individual values or slices in a Variable.
`ref` is a `Tensor` with rank `P` and `indices` is a `Tensor` of rank `Q`.
`indices` must be integer tensor, containing indices into `ref`.
It must be shape `[d_0, ..., d_{Q-2}, K]` where `0 < K <= P`.
The innermost dimension of `indices` (with length `K`) corresponds to
indices into elements (if `K = P`) or slices (if `K < P`) along the `K`th
dimension of `ref`.
`updates` is `Tensor` of rank `Q-1+P-K` with shape:
```
[d_0, ..., d_{Q-2}, ref.shape[K], ..., ref.shape[P-1]].
```
For example, say we want to add 4 scattered elements to a rank-1 tensor to
8 elements. In Python, that update would look like this:
```python
ref = tf.Variable([1, 2, 3, 4, 5, 6, 7, 8])
indices = tf.constant([[4], [3], [1] ,[7]])
updates = tf.constant([9, 10, 11, 12])
op = ref.scatter_nd_sub(indices, updates)
with tf.compat.v1.Session() as sess:
print sess.run(op)
```
The resulting update to ref would look like this:
[1, -9, 3, -6, -6, 6, 7, -4]
See `tf.scatter_nd` for more details about how to make updates to
slices.
Args:
indices: The indices to be used in the operation.
updates: The values to be used in the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered subtraction has completed.
"""
return gen_state_ops.scatter_nd_sub(
self._variable, indices, updates, use_locking=True, name=name)
def scatter_nd_add(self, indices, updates, name=None):
"""Applies sparse addition to individual values or slices in a Variable.
`ref` is a `Tensor` with rank `P` and `indices` is a `Tensor` of rank `Q`.
`indices` must be integer tensor, containing indices into `ref`.
It must be shape `[d_0, ..., d_{Q-2}, K]` where `0 < K <= P`.
The innermost dimension of `indices` (with length `K`) corresponds to
indices into elements (if `K = P`) or slices (if `K < P`) along the `K`th
dimension of `ref`.
`updates` is `Tensor` of rank `Q-1+P-K` with shape:
```
[d_0, ..., d_{Q-2}, ref.shape[K], ..., ref.shape[P-1]].
```
For example, say we want to add 4 scattered elements to a rank-1 tensor to
8 elements. In Python, that update would look like this:
```python
ref = tf.Variable([1, 2, 3, 4, 5, 6, 7, 8])
indices = tf.constant([[4], [3], [1] ,[7]])
updates = tf.constant([9, 10, 11, 12])
add = ref.scatter_nd_add(indices, updates)
with tf.compat.v1.Session() as sess:
print sess.run(add)
```
The resulting update to ref would look like this:
[1, 13, 3, 14, 14, 6, 7, 20]
See `tf.scatter_nd` for more details about how to make updates to
slices.
Args:
indices: The indices to be used in the operation.
updates: The values to be used in the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered addition has completed.
"""
return gen_state_ops.scatter_nd_add(
self._variable, indices, updates, use_locking=True, name=name)
def scatter_nd_update(self, indices, updates, name=None):
"""Applies sparse assignment to individual values or slices in a Variable.
`ref` is a `Tensor` with rank `P` and `indices` is a `Tensor` of rank `Q`.
`indices` must be integer tensor, containing indices into `ref`.
It must be shape `[d_0, ..., d_{Q-2}, K]` where `0 < K <= P`.
The innermost dimension of `indices` (with length `K`) corresponds to
indices into elements (if `K = P`) or slices (if `K < P`) along the `K`th
dimension of `ref`.
`updates` is `Tensor` of rank `Q-1+P-K` with shape:
```
[d_0, ..., d_{Q-2}, ref.shape[K], ..., ref.shape[P-1]].
```
For example, say we want to add 4 scattered elements to a rank-1 tensor to
8 elements. In Python, that update would look like this:
```python
ref = tf.Variable([1, 2, 3, 4, 5, 6, 7, 8])
indices = tf.constant([[4], [3], [1] ,[7]])
updates = tf.constant([9, 10, 11, 12])
op = ref.scatter_nd_update(indices, updates)
with tf.compat.v1.Session() as sess:
print sess.run(op)
```
The resulting update to ref would look like this:
[1, 11, 3, 10, 9, 6, 7, 12]
See `tf.scatter_nd` for more details about how to make updates to
slices.
Args:
indices: The indices to be used in the operation.
updates: The values to be used in the operation.
name: the name of the operation.
Returns:
A `Tensor` that will hold the new value of this variable after
the scattered assignment has completed.
"""
return gen_state_ops.scatter_nd_update(
self._variable, indices, updates, use_locking=True, name=name)
def _strided_slice_assign(self,
begin,
end,
strides,
value,
name,
begin_mask,
end_mask,
ellipsis_mask,
new_axis_mask,
shrink_axis_mask):
return gen_array_ops.strided_slice_assign(ref=self._ref(),
begin=begin,
end=end,
strides=strides,
value=value,
name=name,
begin_mask=begin_mask,
end_mask=end_mask,
ellipsis_mask=ellipsis_mask,
new_axis_mask=new_axis_mask,
shrink_axis_mask=shrink_axis_mask)
@deprecated(None, "Prefer Dataset.range instead.")
def count_up_to(self, limit):
"""Increments this variable until it reaches `limit`.
When that Op is run it tries to increment the variable by `1`. If
incrementing the variable would bring it above `limit` then the Op raises
the exception `OutOfRangeError`.
If no error is raised, the Op outputs the value of the variable before
the increment.
This is essentially a shortcut for `count_up_to(self, limit)`.
Args:
limit: value at which incrementing the variable raises an error.
Returns:
A `Tensor` that will hold the variable value before the increment. If no
other Op modifies this variable, the values produced will all be
distinct.
"""
return state_ops.count_up_to(self._variable, limit=limit)
# Conversion to tensor.
@staticmethod
def _TensorConversionFunction(v, dtype=None, name=None, as_ref=False): # pylint: disable=invalid-name
"""Utility function for converting a Variable to a Tensor."""
_ = name
if dtype and not dtype.is_compatible_with(v.dtype):
raise ValueError(
"Incompatible type conversion requested to type '%s' for variable "
"of type '%s'" % (dtype.name, v.dtype.name))
if as_ref:
return v._ref() # pylint: disable=protected-access
else:
return v.value()
# NOTE(mrry): This enables the Variable's overloaded "right" binary
# operators to run when the left operand is an ndarray, because it
# accords the Variable class higher priority than an ndarray, or a
# numpy matrix.
# TODO(mrry): Convert this to using numpy's __numpy_ufunc__
# mechanism, which allows more control over how Variables interact
# with ndarrays.
__array_priority__ = 100
@property
def name(self):
"""The name of this variable."""
return self._variable.name
@property
def initializer(self):
"""The initializer operation for this variable."""
return self._initializer_op
@property
def device(self):
"""The device of this variable."""
return self._variable.device
@property
def dtype(self):
"""The `DType` of this variable."""
return self._variable.dtype
@property
def op(self):
"""The `Operation` of this variable."""
return self._variable.op
@property
def graph(self):
"""The `Graph` of this variable."""
return self._variable.graph
@property
def _distribute_strategy(self):
"""The `tf.distribute.Strategy` that this variable was created under."""
return None # Ref variables are never created inside a strategy.
@property
def shape(self):
"""The `TensorShape` of this variable.
Returns:
A `TensorShape`.
"""
return self._variable.get_shape()
def to_proto(self, export_scope=None):
"""Converts a `Variable` to a `VariableDef` protocol buffer.
Args:
export_scope: Optional `string`. Name scope to remove.
Returns:
A `VariableDef` protocol buffer, or `None` if the `Variable` is not
in the specified name scope.
"""
if (export_scope is None or
self._variable.name.startswith(export_scope)):
var_def = variable_pb2.VariableDef()
var_def.variable_name = ops.strip_name_scope(
self._variable.name, export_scope)
if self._initial_value is not None:
# For backwards compatibility.
var_def.initial_value_name = ops.strip_name_scope(
self._initial_value.name, export_scope)
var_def.trainable = self.trainable
var_def.synchronization = self.synchronization.value
var_def.aggregation = self.aggregation.value
var_def.initializer_name = ops.strip_name_scope(
self.initializer.name, export_scope)
var_def.snapshot_name = ops.strip_name_scope(
self._snapshot.name, export_scope)
if self._save_slice_info:
var_def.save_slice_info_def.MergeFrom(self._save_slice_info.to_proto(
export_scope=export_scope))
return var_def
else:
return None
def __iadd__(self, other):
logging.log_first_n(
logging.WARN,
"Variable += will be deprecated. Use variable.assign_add"
" if you want assignment to the variable value or 'x = x + y'"
" if you want a new python Tensor object.", 1)
return self + other
def __isub__(self, other):
logging.log_first_n(
logging.WARN,
"Variable -= will be deprecated. Use variable.assign_sub"
" if you want assignment to the variable value or 'x = x - y'"
" if you want a new python Tensor object.", 1)
return self - other
def __imul__(self, other):
logging.log_first_n(
logging.WARN,
"Variable *= will be deprecated. Use `var.assign(var * other)`"
" if you want assignment to the variable value or `x = x * y`"
" if you want a new python Tensor object.", 1)
return self * other
def __idiv__(self, other):
logging.log_first_n(
logging.WARN,
"Variable /= will be deprecated. Use `var.assign(var / other)`"
" if you want assignment to the variable value or `x = x / y`"
" if you want a new python Tensor object.", 1)
return self / other
def __itruediv__(self, other):
logging.log_first_n(
logging.WARN,
"Variable /= will be deprecated. Use `var.assign(var / other)`"
" if you want assignment to the variable value or `x = x / y`"
" if you want a new python Tensor object.", 1)
return self / other
def __irealdiv__(self, other):
logging.log_first_n(
logging.WARN,
"Variable /= will be deprecated. Use `var.assign(var / other)`"
" if you want assignment to the variable value or `x = x / y`"
" if you want a new python Tensor object.", 1)
return self / other
def __ipow__(self, other):
logging.log_first_n(
logging.WARN,
"Variable **= will be deprecated. Use `var.assign(var ** other)`"
" if you want assignment to the variable value or `x = x ** y`"
" if you want a new python Tensor object.", 1)
return self ** other
def _try_guard_against_uninitialized_dependencies(name, initial_value):
"""Attempt to guard against dependencies on uninitialized variables.
Replace references to variables in `initial_value` with references to the
variable's initialized values. The initialized values are essentially
conditional TensorFlow graphs that return a variable's value if it is
initialized or its `initial_value` if it hasn't been initialized. This
replacement is done on a best effort basis:
- If the `initial_value` graph contains cycles, we don't do any
replacements for that graph.
- If the variables that `initial_value` depends on are not present in the
`GLOBAL_VARIABLES` or `LOCAL_VARIABLES` we don't replace them.
In these cases, it is up to the caller to ensure that the `initial_value`
graph uses initialized variables or that they guard access to variables
using their `initialized_value` method.
Args:
name: Variable name.
initial_value: `Tensor`. The initial value.
Returns:
A `Tensor` suitable to initialize a variable.
Raises:
TypeError: If `initial_value` is not a `Tensor`.
"""
if not isinstance(initial_value, ops.Tensor):
raise TypeError("initial_value needs to be a Tensor: %s" % initial_value)
# Don't modify initial_value if it contains any cyclic dependencies.
if _has_cycle(initial_value.op, state={}):
return initial_value
return _safe_initial_value_from_tensor(name, initial_value, op_cache={})
_UNKNOWN, _STARTED, _FINISHED = range(3)
def _has_cycle(op, state):
"""Detect cycles in the dependencies of `initial_value`."""
op_state = state.get(op.name, _UNKNOWN)
if op_state == _STARTED:
return True
elif op_state == _FINISHED:
return False
state[op.name] = _STARTED
for i in itertools.chain((i.op for i in op.inputs), op.control_inputs):
if _has_cycle(i, state):
return True
state[op.name] = _FINISHED
return False
def _safe_initial_value_from_tensor(name, tensor, op_cache):
"""Replace dependencies on variables with their initialized values.
Args:
name: Variable name.
tensor: A `Tensor`. The tensor to replace.
op_cache: A dict mapping operation names to `Operation`s. Used to memoize
the results so as to avoid creating redundant operations.
Returns:
A `Tensor` compatible with `tensor`. Any inputs that lead to variable
values will be replaced with a corresponding graph that uses the
variable's initialized values. This is done on a best-effort basis. If no
modifications need to be made then `tensor` will be returned unchanged.
"""
op = tensor.op
new_op = op_cache.get(op.name)
if new_op is None:
new_op = _safe_initial_value_from_op(name, op, op_cache)
op_cache[op.name] = new_op
return new_op.outputs[tensor.value_index]
def _safe_initial_value_from_op(name, op, op_cache):
"""Replace dependencies on variables with their initialized values.
Args:
name: Variable name.
op: An `Operation`. The operation to replace.
op_cache: A dict mapping operation names to `Operation`s. Used to memoize
the results so as to avoid creating redundant operations.
Returns:
An `Operation` compatible with `op`. Any inputs that lead to variable
values will be replaced with a corresponding graph that uses the
variable's initialized values. This is done on a best-effort basis. If no
modifications need to be made then `op` will be returned unchanged.
"""
op_type = op.node_def.op
if op_type in ("IsVariableInitialized", "VarIsInitializedOp",
"ReadVariableOp"):
return op
# Attempt to find the initialized_value of any variable reference / handles.
# TODO(b/70206927): Fix handling of ResourceVariables.
if op_type in ("Variable", "VariableV2", "VarHandleOp"):
initialized_value = _find_initialized_value_for_variable(op)
return op if initialized_value is None else initialized_value.op
# Recursively build initializer expressions for inputs.
modified = False
new_op_inputs = []
for op_input in op.inputs:
new_op_input = _safe_initial_value_from_tensor(name, op_input, op_cache)
new_op_inputs.append(new_op_input)
modified = modified or (new_op_input != op_input)
# If at least one input was modified, replace the op.
if modified:
new_op_type = op_type
if new_op_type == "RefSwitch":
new_op_type = "Switch"
new_op_name = op.node_def.name + "_" + name
new_op_name = new_op_name.replace(":", "_")
return op.graph.create_op(
new_op_type, new_op_inputs,
op._output_types, # pylint: disable=protected-access
name=new_op_name, attrs=op.node_def.attr)
return op
def _find_initialized_value_for_variable(variable_op):
"""Find the initialized value for a variable op.
To do so, lookup the variable op in the variables collection.
Args:
variable_op: A variable `Operation`.
Returns:
A `Tensor` representing the initialized value for the variable or `None`
if the initialized value could not be found.
"""
try:
var_names = [variable_op.node_def.name, variable_op.node_def.name + ":0"]
for collection_name in (ops.GraphKeys.GLOBAL_VARIABLES,
ops.GraphKeys.LOCAL_VARIABLES):
for var in variable_op.graph.get_collection(collection_name):
if var.name in var_names:
return var.initialized_value()
except AttributeError:
# Return None when an incomplete user-defined variable type was put in
# the collection.
return None
return None
class PartitionedVariable(object):
"""A container for partitioned `Variable` objects.
@compatibility(eager) `tf.PartitionedVariable` is not compatible with
eager execution. Use `tf.Variable` instead which is compatible
with both eager execution and graph construction. See [the
TensorFlow Eager Execution
guide](https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/eager/python/g3doc/guide.md#variables-and-optimizers)
for details on how variables work in eager execution.
@end_compatibility
"""
def __init__(self, name, shape, dtype, variable_list, partitions):
"""Creates a new partitioned variable wrapper.
Variables passed via the variable_list must contain a save_slice_info
field. Concatenation and iteration is in lexicographic order according
to the var_offset property of the save_slice_info.
Args:
name: String. Overall name of the variables.
shape: List of integers. Overall shape of the variables.
dtype: Type of the variables.
variable_list: List of `Variable` that comprise this partitioned variable.
partitions: List of integers. Number of partitions for each dimension.
Raises:
TypeError: If `variable_list` is not a list of `Variable` objects, or
`partitions` is not a list.
ValueError: If `variable_list` is empty, or the `Variable` shape
information does not match `shape`, or `partitions` has invalid values.
"""
if not isinstance(variable_list, (list, tuple)):
raise TypeError(
"variable_list is not a list or tuple: %s" % variable_list)
if not isinstance(partitions, (list, tuple)):
raise TypeError("partitions is not a list or tuple: %s" % partitions)
if not all(p >= 1 for p in partitions):
raise ValueError("partition values must be positive: %s" % partitions)
if not variable_list:
raise ValueError("variable_list may not be empty")
# pylint: disable=protected-access
for v in variable_list:
# Sort the variable_list lexicographically according to var offset value.
if not all(v._get_save_slice_info() is not None for v in variable_list):
raise ValueError(
"All variables must have a save_slice_info available: %s"
% [v.name for v in variable_list])
if len(shape) != len(partitions):
raise ValueError("len(shape) != len(partitions): %s vs. %s"
% (shape, partitions))
if v._get_save_slice_info().full_shape != shape:
raise ValueError(
"All variables' full shapes must match shape: %s; "
"but full shapes were: %s"
% (shape, str([v._get_save_slice_info().full_shape])))
self._variable_list = sorted(
variable_list, key=lambda v: v._get_save_slice_info().var_offset)
# pylint: enable=protected-access
self._name = name
self._shape = shape
self._dtype = dtype
self._partitions = partitions
self._as_tensor = None
def __iter__(self):
"""Return an iterable for accessing the underlying partition Variables."""
return iter(self._variable_list)
def __len__(self):
num_partition_axes = len(self._partition_axes())
if num_partition_axes > 1:
raise ValueError("Cannot get a length for %d > 1 partition axes"
% num_partition_axes)
return len(self._variable_list)
def _partition_axes(self):
if all(p == 1 for p in self._partitions):
return [0]
else:
return [i for i, p in enumerate(self._partitions) if p > 1]
def _concat(self):
"""Returns the overall concatenated value as a `Tensor`.
This is different from using the partitioned variable directly as a tensor
(through tensor conversion and `as_tensor`) in that it creates a new set of
operations that keeps the control dependencies from its scope.
Returns:
`Tensor` containing the concatenated value.
"""
if len(self._variable_list) == 1:
with ops.name_scope(None):
return array_ops.identity(self._variable_list[0], name=self._name)
partition_axes = self._partition_axes()
if len(partition_axes) > 1:
raise NotImplementedError(
"Cannot concatenate along more than one dimension: %s. "
"Multi-axis partition concat is not supported" % str(partition_axes))
partition_ix = partition_axes[0]
with ops.name_scope(self._name + "/ConcatPartitions/"):
concatenated = array_ops.concat(self._variable_list, partition_ix)
with ops.name_scope(None):
return array_ops.identity(concatenated, name=self._name)
def as_tensor(self):
"""Returns the overall concatenated value as a `Tensor`.
The returned tensor will not inherit the control dependencies from the scope
where the value is used, which is similar to getting the value of
`Variable`.
Returns:
`Tensor` containing the concatenated value.
"""
with ops.control_dependencies(None):
return self._concat()
@staticmethod
def _TensorConversionFunction(v, dtype=None, name=None, as_ref=False):
# pylint: disable=invalid-name
_ = name
if dtype is not None and not dtype.is_compatible_with(v.dtype):
raise ValueError(
"Incompatible type conversion requested to type '%s' for variable "
"of type '%s'" % (dtype.name, v.dtype.name))
if as_ref:
raise NotImplementedError(
"PartitionedVariable doesn't support being used as a reference.")
else:
return v.as_tensor()
@property
def name(self):
return self._name
@property
def dtype(self):
return self._dtype
@property
def shape(self):
return self.get_shape()
@property
def _distribute_strategy(self):
"""The `tf.distribute.Strategy` that this variable was created under."""
# NOTE(yuefengz): Today, no partitioned variables in a distribute strategy.
return None
def get_shape(self):
return self._shape
def _get_variable_list(self):
return self._variable_list
def _get_partitions(self):
return self._partitions
def _apply_assign_fn(self, assign_fn, value):
partition_axes = self._partition_axes()
if len(partition_axes) > 1:
raise NotImplementedError(
"Cannot do assign action along more than one dimension: %s. "
"Multi-axis partition assign action is not supported " %
str(partition_axes))
if isinstance(value, list):
assert len(value) == len(self._variable_list)
value_list = value
elif isinstance(value, PartitionedVariable):
value_list = [var_part for var_part in value]
else:
partition_ix = partition_axes[0]
size_splits_list = [
tensor_shape.dimension_value(var.shape[partition_ix])
for var in self._variable_list
]
value_list = array_ops.split(value, size_splits_list, axis=partition_ix)
op_list = [
assign_fn(var, value_list[idx])
for idx, var in enumerate(self._variable_list)
]
return op_list
def assign(self, value, use_locking=False, name=None, read_value=True):
assign_fn = lambda var, r_value: var.assign(
r_value, use_locking=use_locking,
name=name, read_value=read_value)
assign_list = self._apply_assign_fn(assign_fn, value)
if read_value:
return assign_list
return [assign.op for assign in assign_list]
def assign_add(self, value, use_locking=False, name=None, read_value=True):
assign_fn = lambda var, r_value: var.assign_add(
r_value, use_locking=use_locking,
name=name, read_value=read_value)
assign_list = self._apply_assign_fn(assign_fn, value)
if read_value:
return assign_list
return [assign.op for assign in assign_list]
def assign_sub(self, value, use_locking=False, name=None, read_value=True):
assign_fn = lambda var, r_value: var.assign_sub(
r_value, use_locking=use_locking,
name=name, read_value=read_value)
assign_list = self._apply_assign_fn(assign_fn, value)
if read_value:
return assign_list
return [assign.op for assign in assign_list]
# Register a conversion function which reads the value of the variable,
# allowing instances of the class to be used as tensors.
ops.register_tensor_conversion_function(
RefVariable,
RefVariable._TensorConversionFunction) # pylint: disable=protected-access
ops.register_dense_tensor_like_type(RefVariable)
@tf_export(v1=["global_variables"])
def global_variables(scope=None):
"""Returns global variables.
Global variables are variables that are shared across machines in a
distributed environment. The `Variable()` constructor or `get_variable()`
automatically adds new variables to the graph collection
`GraphKeys.GLOBAL_VARIABLES`.
This convenience function returns the contents of that collection.
An alternative to global variables are local variables. See
`tf.compat.v1.local_variables`
Args:
scope: (Optional.) A string. If supplied, the resulting list is filtered
to include only items whose `name` attribute matches `scope` using
`re.match`. Items without a `name` attribute are never returned if a
scope is supplied. The choice of `re.match` means that a `scope` without
special tokens filters by prefix.
Returns:
A list of `Variable` objects.
"""
return ops.get_collection(ops.GraphKeys.GLOBAL_VARIABLES, scope)
@tf_export(v1=["all_variables"])
@deprecated("2017-03-02", "Please use tf.global_variables instead.")
def all_variables():
"""Use `tf.compat.v1.global_variables` instead."""
return global_variables()
def _all_saveable_objects(scope=None):
"""Returns all variables and `SaveableObject`s that must be checkpointed.
Args:
scope: (Optional.) A string. If supplied, the resulting list is filtered
to include only items whose `name` attribute matches `scope` using
`re.match`. Items without a `name` attribute are never returned if a
scope is supplied. The choice of `re.match` means that a `scope` without
special tokens filters by prefix.
Returns:
A list of `Variable` and `SaveableObject` to be checkpointed
"""
# TODO(andreasst): make this function public once things are settled.
return (ops.get_collection(ops.GraphKeys.GLOBAL_VARIABLES, scope) +
ops.get_collection(ops.GraphKeys.SAVEABLE_OBJECTS, scope))
@tf_export(v1=["local_variables"])
def local_variables(scope=None):
"""Returns local variables.
Local variables - per process variables, usually not saved/restored to
checkpoint and used for temporary or intermediate values.
For example, they can be used as counters for metrics computation or
number of epochs this machine has read data.
The `tf.contrib.framework.local_variable()` function automatically adds the
new variable to `GraphKeys.LOCAL_VARIABLES`.
This convenience function returns the contents of that collection.
An alternative to local variables are global variables. See
`tf.compat.v1.global_variables`
Args:
scope: (Optional.) A string. If supplied, the resulting list is filtered
to include only items whose `name` attribute matches `scope` using
`re.match`. Items without a `name` attribute are never returned if a
scope is supplied. The choice of `re.match` means that a `scope` without
special tokens filters by prefix.
Returns:
A list of local `Variable` objects.
"""
return ops.get_collection(ops.GraphKeys.LOCAL_VARIABLES, scope)
@tf_export(v1=["model_variables"])
def model_variables(scope=None):
"""Returns all variables in the MODEL_VARIABLES collection.
Args:
scope: (Optional.) A string. If supplied, the resulting list is filtered
to include only items whose `name` attribute matches `scope` using
`re.match`. Items without a `name` attribute are never returned if a
scope is supplied. The choice of `re.match` means that a `scope` without
special tokens filters by prefix.
Returns:
A list of local Variable objects.
"""
return ops.get_collection(ops.GraphKeys.MODEL_VARIABLES, scope)
@tf_export(v1=["trainable_variables"])
def trainable_variables(scope=None):
"""Returns all variables created with `trainable=True`.
When passed `trainable=True`, the `Variable()` constructor automatically
adds new variables to the graph collection
`GraphKeys.TRAINABLE_VARIABLES`. This convenience function returns the
contents of that collection.
Args:
scope: (Optional.) A string. If supplied, the resulting list is filtered
to include only items whose `name` attribute matches `scope` using
`re.match`. Items without a `name` attribute are never returned if a
scope is supplied. The choice of `re.match` means that a `scope` without
special tokens filters by prefix.
Returns:
A list of Variable objects.
"""
return ops.get_collection(ops.GraphKeys.TRAINABLE_VARIABLES, scope)
@tf_export(v1=["moving_average_variables"])
def moving_average_variables(scope=None):
"""Returns all variables that maintain their moving averages.
If an `ExponentialMovingAverage` object is created and the `apply()`
method is called on a list of variables, these variables will
be added to the `GraphKeys.MOVING_AVERAGE_VARIABLES` collection.
This convenience function returns the contents of that collection.
Args:
scope: (Optional.) A string. If supplied, the resulting list is filtered
to include only items whose `name` attribute matches `scope` using
`re.match`. Items without a `name` attribute are never returned if a
scope is supplied. The choice of `re.match` means that a `scope` without
special tokens filters by prefix.
Returns:
A list of Variable objects.
"""
return ops.get_collection(ops.GraphKeys.MOVING_AVERAGE_VARIABLES, scope)
@tf_export(v1=["initializers.variables", "variables_initializer"])
def variables_initializer(var_list, name="init"):
"""Returns an Op that initializes a list of variables.
After you launch the graph in a session, you can run the returned Op to
initialize all the variables in `var_list`. This Op runs all the
initializers of the variables in `var_list` in parallel.
Calling `initialize_variables()` is equivalent to passing the list of
initializers to `Group()`.
If `var_list` is empty, however, the function still returns an Op that can
be run. That Op just has no effect.
Args:
var_list: List of `Variable` objects to initialize.
name: Optional name for the returned operation.
Returns:
An Op that run the initializers of all the specified variables.
"""
if var_list and not context.executing_eagerly():
return control_flow_ops.group(*[v.initializer for v in var_list], name=name)
return control_flow_ops.no_op(name=name)
@tf_export(v1=["initialize_variables"])
@tf_should_use.should_use_result
@deprecated("2017-03-02", "Use `tf.variables_initializer` instead.")
def initialize_variables(var_list, name="init"):
"""See `tf.compat.v1.variables_initializer`."""
return variables_initializer(var_list, name=name)
@tf_export(v1=["initializers.global_variables", "global_variables_initializer"])
def global_variables_initializer():
"""Returns an Op that initializes global variables.
This is just a shortcut for `variables_initializer(global_variables())`
Returns:
An Op that initializes global variables in the graph.
"""
if context.executing_eagerly():
return control_flow_ops.no_op(name="global_variables_initializer")
return variables_initializer(global_variables())
@tf_export(v1=["initialize_all_variables"])
@tf_should_use.should_use_result
@deprecated("2017-03-02", "Use `tf.global_variables_initializer` instead.")
def initialize_all_variables():
"""See `tf.compat.v1.global_variables_initializer`."""
return global_variables_initializer()
@tf_export(v1=["initializers.local_variables", "local_variables_initializer"])
def local_variables_initializer():
"""Returns an Op that initializes all local variables.
This is just a shortcut for `variables_initializer(local_variables())`
Returns:
An Op that initializes all local variables in the graph.
"""
if context.executing_eagerly():
return control_flow_ops.no_op(name="local_variables_initializer")
return variables_initializer(local_variables())
@tf_export(v1=["initialize_local_variables"])
@tf_should_use.should_use_result
@deprecated("2017-03-02", "Use `tf.local_variables_initializer` instead.")
def initialize_local_variables():
"""See `tf.compat.v1.local_variables_initializer`."""
return local_variables_initializer()
@tf_export(v1=["is_variable_initialized"])
@tf_should_use.should_use_result
def is_variable_initialized(variable):
"""Tests if a variable has been initialized.
Args:
variable: A `Variable`.
Returns:
Returns a scalar boolean Tensor, `True` if the variable has been
initialized, `False` otherwise.
"""
return state_ops.is_variable_initialized(variable)
@tf_export(v1=["assert_variables_initialized"])
@tf_should_use.should_use_result
def assert_variables_initialized(var_list=None):
"""Returns an Op to check if variables are initialized.
NOTE: This function is obsolete and will be removed in 6 months. Please
change your implementation to use `report_uninitialized_variables()`.
When run, the returned Op will raise the exception `FailedPreconditionError`
if any of the variables has not yet been initialized.
Note: This function is implemented by trying to fetch the values of the
variables. If one of the variables is not initialized a message may be
logged by the C++ runtime. This is expected.
Args:
var_list: List of `Variable` objects to check. Defaults to the
value of `global_variables().`
Returns:
An Op, or None if there are no variables.
"""
if var_list is None:
var_list = global_variables() + local_variables()
# Backwards compatibility for old-style variables. TODO(touts): remove.
if not var_list:
var_list = []
for op in ops.get_default_graph().get_operations():
if op.type in ["Variable", "VariableV2", "AutoReloadVariable"]:
var_list.append(op.outputs[0])
if not var_list:
return None
else:
ranks = []
for var in var_list:
with ops.colocate_with(var.op):
ranks.append(array_ops.rank_internal(var, optimize=False))
if len(ranks) == 1:
return ranks[0]
else:
return array_ops.stack(ranks)
@tf_export(v1=["report_uninitialized_variables"])
@tf_should_use.should_use_result
def report_uninitialized_variables(var_list=None,
name="report_uninitialized_variables"):
"""Adds ops to list the names of uninitialized variables.
When run, it returns a 1-D tensor containing the names of uninitialized
variables if there are any, or an empty array if there are none.
Args:
var_list: List of `Variable` objects to check. Defaults to the
value of `global_variables() + local_variables()`
name: Optional name of the `Operation`.
Returns:
A 1-D tensor containing names of the uninitialized variables, or an empty
1-D tensor if there are no variables or no uninitialized variables.
"""
if var_list is None:
var_list = global_variables() + local_variables()
# Backwards compatibility for old-style variables. TODO(touts): remove.
if not var_list:
var_list = []
for op in ops.get_default_graph().get_operations():
if op.type in ["Variable", "VariableV2", "AutoReloadVariable"]:
var_list.append(op.outputs[0])
with ops.name_scope(name):
# Run all operations on CPU
if var_list:
init_vars = [state_ops.is_variable_initialized(v) for v in var_list]
local_device = os.environ.get(
"TF_DEVICE_FOR_UNINITIALIZED_VARIABLE_REPORTING", "/cpu:0")
with ops.device(local_device):
if not var_list:
# Return an empty tensor so we only need to check for returned tensor
# size being 0 as an indication of model ready.
return array_ops.constant([], dtype=dtypes.string)
else:
# Get a 1-D boolean tensor listing whether each variable is initialized.
variables_mask = math_ops.logical_not(array_ops.stack(init_vars))
# Get a 1-D string tensor containing all the variable names.
variable_names_tensor = array_ops.constant(
[s.op.name for s in var_list])
# Return a 1-D tensor containing all the names of
# uninitialized variables.
return array_ops.boolean_mask(variable_names_tensor, variables_mask)
ops.register_tensor_conversion_function(
PartitionedVariable,
PartitionedVariable._TensorConversionFunction) # pylint: disable=protected-access
class AbstractVariableMetaclass(VariableMetaclass, abc.ABCMeta):
"""Metaclass combining `VariableMetaclass` and `abc.ABCMeta`."""
pass
@six.add_metaclass(AbstractVariableMetaclass)
class AbstractVariable(Variable):
"""`Variable`, but abstract."""
pass
|
tensorflow-master
|
tensorflow/python/ops/variables.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Wrappers for candidate sampling operations."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import random_seed
from tensorflow.python.ops import array_ops # pylint: disable=unused-import
from tensorflow.python.ops import gen_candidate_sampling_ops
from tensorflow.python.ops import math_ops # pylint: disable=unused-import
from tensorflow.python.util import deprecation
from tensorflow.python.util.tf_export import tf_export
@tf_export(
'random.uniform_candidate_sampler',
v1=['random.uniform_candidate_sampler', 'nn.uniform_candidate_sampler'])
@deprecation.deprecated_endpoints('nn.uniform_candidate_sampler')
def uniform_candidate_sampler(true_classes, num_true, num_sampled, unique,
range_max, seed=None, name=None):
"""Samples a set of classes using a uniform base distribution.
This operation randomly samples a tensor of sampled classes
(`sampled_candidates`) from the range of integers `[0, range_max)`.
The elements of `sampled_candidates` are drawn without replacement
(if `unique=True`) or with replacement (if `unique=False`) from
the base distribution.
The base distribution for this operation is the uniform distribution
over the range of integers `[0, range_max)`.
In addition, this operation returns tensors `true_expected_count`
and `sampled_expected_count` representing the number of times each
of the target classes (`true_classes`) and the sampled
classes (`sampled_candidates`) is expected to occur in an average
tensor of sampled classes. These values correspond to `Q(y|x)`
defined in [this
document](http://www.tensorflow.org/extras/candidate_sampling.pdf).
If `unique=True`, then these are post-rejection probabilities and we
compute them approximately.
Args:
true_classes: A `Tensor` of type `int64` and shape `[batch_size,
num_true]`. The target classes.
num_true: An `int`. The number of target classes per training example.
num_sampled: An `int`. The number of classes to randomly sample. The
`sampled_candidates` return value will have shape `[num_sampled]`. If
`unique=True`, `num_sampled` must be less than or equal to `range_max`.
unique: A `bool`. Determines whether all sampled classes in a batch are
unique.
range_max: An `int`. The number of possible classes.
seed: An `int`. An operation-specific seed. Default is 0.
name: A name for the operation (optional).
Returns:
sampled_candidates: A tensor of type `int64` and shape `[num_sampled]`. The
sampled classes, either with possible duplicates (`unique=False`) or all
unique (`unique=True`). In either case, `sampled_candidates` is
independent of the true classes.
true_expected_count: A tensor of type `float`. Same shape as
`true_classes`. The expected counts under the sampling distribution
of each of `true_classes`.
sampled_expected_count: A tensor of type `float`. Same shape as
`sampled_candidates`. The expected counts under the sampling distribution
of each of `sampled_candidates`.
"""
seed1, seed2 = random_seed.get_seed(seed)
return gen_candidate_sampling_ops.uniform_candidate_sampler(
true_classes, num_true, num_sampled, unique, range_max, seed=seed1,
seed2=seed2, name=name)
@tf_export(
'random.log_uniform_candidate_sampler',
v1=[
'random.log_uniform_candidate_sampler',
'nn.log_uniform_candidate_sampler'
])
@deprecation.deprecated_endpoints('nn.log_uniform_candidate_sampler')
def log_uniform_candidate_sampler(true_classes, num_true, num_sampled, unique,
range_max, seed=None, name=None):
"""Samples a set of classes using a log-uniform (Zipfian) base distribution.
This operation randomly samples a tensor of sampled classes
(`sampled_candidates`) from the range of integers `[0, range_max)`.
The elements of `sampled_candidates` are drawn without replacement
(if `unique=True`) or with replacement (if `unique=False`) from
the base distribution.
The base distribution for this operation is an approximately log-uniform
or Zipfian distribution:
`P(class) = (log(class + 2) - log(class + 1)) / log(range_max + 1)`
This sampler is useful when the target classes approximately follow such
a distribution - for example, if the classes represent words in a lexicon
sorted in decreasing order of frequency. If your classes are not ordered by
decreasing frequency, do not use this op.
In addition, this operation returns tensors `true_expected_count`
and `sampled_expected_count` representing the number of times each
of the target classes (`true_classes`) and the sampled
classes (`sampled_candidates`) is expected to occur in an average
tensor of sampled classes. These values correspond to `Q(y|x)`
defined in [this
document](http://www.tensorflow.org/extras/candidate_sampling.pdf).
If `unique=True`, then these are post-rejection probabilities and we
compute them approximately.
Args:
true_classes: A `Tensor` of type `int64` and shape `[batch_size,
num_true]`. The target classes.
num_true: An `int`. The number of target classes per training example.
num_sampled: An `int`. The number of classes to randomly sample.
unique: A `bool`. Determines whether all sampled classes in a batch are
unique.
range_max: An `int`. The number of possible classes.
seed: An `int`. An operation-specific seed. Default is 0.
name: A name for the operation (optional).
Returns:
sampled_candidates: A tensor of type `int64` and shape `[num_sampled]`.
The sampled classes.
true_expected_count: A tensor of type `float`. Same shape as
`true_classes`. The expected counts under the sampling distribution
of each of `true_classes`.
sampled_expected_count: A tensor of type `float`. Same shape as
`sampled_candidates`. The expected counts under the sampling distribution
of each of `sampled_candidates`.
"""
seed1, seed2 = random_seed.get_seed(seed)
return gen_candidate_sampling_ops.log_uniform_candidate_sampler(
true_classes, num_true, num_sampled, unique, range_max, seed=seed1,
seed2=seed2, name=name)
@tf_export(
'random.learned_unigram_candidate_sampler',
'nn.learned_unigram_candidate_sampler')
@deprecation.deprecated_endpoints(['nn.learned_unigram_candidate_sampler'])
def learned_unigram_candidate_sampler(true_classes, num_true, num_sampled,
unique, range_max, seed=None, name=None):
"""Samples a set of classes from a distribution learned during training.
This operation randomly samples a tensor of sampled classes
(`sampled_candidates`) from the range of integers `[0, range_max)`.
The elements of `sampled_candidates` are drawn without replacement
(if `unique=True`) or with replacement (if `unique=False`) from
the base distribution.
The base distribution for this operation is constructed on the fly
during training. It is a unigram distribution over the target
classes seen so far during training. Every integer in `[0, range_max)`
begins with a weight of 1, and is incremented by 1 each time it is
seen as a target class. The base distribution is not saved to checkpoints,
so it is reset when the model is reloaded.
In addition, this operation returns tensors `true_expected_count`
and `sampled_expected_count` representing the number of times each
of the target classes (`true_classes`) and the sampled
classes (`sampled_candidates`) is expected to occur in an average
tensor of sampled classes. These values correspond to `Q(y|x)`
defined in [this
document](http://www.tensorflow.org/extras/candidate_sampling.pdf).
If `unique=True`, then these are post-rejection probabilities and we
compute them approximately.
Args:
true_classes: A `Tensor` of type `int64` and shape `[batch_size,
num_true]`. The target classes.
num_true: An `int`. The number of target classes per training example.
num_sampled: An `int`. The number of classes to randomly sample.
unique: A `bool`. Determines whether all sampled classes in a batch are
unique.
range_max: An `int`. The number of possible classes.
seed: An `int`. An operation-specific seed. Default is 0.
name: A name for the operation (optional).
Returns:
sampled_candidates: A tensor of type `int64` and shape `[num_sampled]`.
The sampled classes.
true_expected_count: A tensor of type `float`. Same shape as
`true_classes`. The expected counts under the sampling distribution
of each of `true_classes`.
sampled_expected_count: A tensor of type `float`. Same shape as
`sampled_candidates`. The expected counts under the sampling distribution
of each of `sampled_candidates`.
"""
seed1, seed2 = random_seed.get_seed(seed)
return gen_candidate_sampling_ops.learned_unigram_candidate_sampler(
true_classes, num_true, num_sampled, unique, range_max, seed=seed1,
seed2=seed2, name=name)
@tf_export('random.fixed_unigram_candidate_sampler',
'nn.fixed_unigram_candidate_sampler')
def fixed_unigram_candidate_sampler(true_classes,
num_true,
num_sampled,
unique,
range_max,
vocab_file='',
distortion=1.0,
num_reserved_ids=0,
num_shards=1,
shard=0,
unigrams=(),
seed=None,
name=None):
"""Samples a set of classes using the provided (fixed) base distribution.
This operation randomly samples a tensor of sampled classes
(`sampled_candidates`) from the range of integers `[0, range_max)`.
The elements of `sampled_candidates` are drawn without replacement
(if `unique=True`) or with replacement (if `unique=False`) from
the base distribution.
The base distribution is read from a file or passed in as an
in-memory array. There is also an option to skew the distribution by
applying a distortion power to the weights.
In addition, this operation returns tensors `true_expected_count`
and `sampled_expected_count` representing the number of times each
of the target classes (`true_classes`) and the sampled
classes (`sampled_candidates`) is expected to occur in an average
tensor of sampled classes. These values correspond to `Q(y|x)`
defined in [this
document](http://www.tensorflow.org/extras/candidate_sampling.pdf).
If `unique=True`, then these are post-rejection probabilities and we
compute them approximately.
Args:
true_classes: A `Tensor` of type `int64` and shape `[batch_size,
num_true]`. The target classes.
num_true: An `int`. The number of target classes per training example.
num_sampled: An `int`. The number of classes to randomly sample.
unique: A `bool`. Determines whether all sampled classes in a batch are
unique.
range_max: An `int`. The number of possible classes.
vocab_file: Each valid line in this file (which should have a CSV-like
format) corresponds to a valid word ID. IDs are in sequential order,
starting from num_reserved_ids. The last entry in each line is expected
to be a value corresponding to the count or relative probability. Exactly
one of `vocab_file` and `unigrams` needs to be passed to this operation.
distortion: The distortion is used to skew the unigram probability
distribution. Each weight is first raised to the distortion's power
before adding to the internal unigram distribution. As a result,
`distortion = 1.0` gives regular unigram sampling (as defined by the vocab
file), and `distortion = 0.0` gives a uniform distribution.
num_reserved_ids: Optionally some reserved IDs can be added in the range
`[0, num_reserved_ids)` by the users. One use case is that a special
unknown word token is used as ID 0. These IDs will have a sampling
probability of 0.
num_shards: A sampler can be used to sample from a subset of the original
range in order to speed up the whole computation through parallelism. This
parameter (together with `shard`) indicates the number of partitions that
are being used in the overall computation.
shard: A sampler can be used to sample from a subset of the original range
in order to speed up the whole computation through parallelism. This
parameter (together with `num_shards`) indicates the particular partition
number of the operation, when partitioning is being used.
unigrams: A list of unigram counts or probabilities, one per ID in
sequential order. Exactly one of `vocab_file` and `unigrams` should be
passed to this operation.
seed: An `int`. An operation-specific seed. Default is 0.
name: A name for the operation (optional).
Returns:
sampled_candidates: A tensor of type `int64` and shape `[num_sampled]`.
The sampled classes.
true_expected_count: A tensor of type `float`. Same shape as
`true_classes`. The expected counts under the sampling distribution
of each of `true_classes`.
sampled_expected_count: A tensor of type `float`. Same shape as
`sampled_candidates`. The expected counts under the sampling distribution
of each of `sampled_candidates`.
"""
seed1, seed2 = random_seed.get_seed(seed)
return gen_candidate_sampling_ops.fixed_unigram_candidate_sampler(
true_classes, num_true, num_sampled, unique, range_max,
vocab_file=vocab_file, distortion=distortion,
num_reserved_ids=num_reserved_ids, num_shards=num_shards, shard=shard,
unigrams=unigrams, seed=seed1, seed2=seed2, name=name)
@tf_export('random.all_candidate_sampler', 'nn.all_candidate_sampler')
def all_candidate_sampler(true_classes, num_true, num_sampled, unique,
seed=None, name=None):
"""Generate the set of all classes.
Deterministically generates and returns the set of all possible classes.
For testing purposes. There is no need to use this, since you might as
well use full softmax or full logistic regression.
Args:
true_classes: A `Tensor` of type `int64` and shape `[batch_size,
num_true]`. The target classes.
num_true: An `int`. The number of target classes per training example.
num_sampled: An `int`. The number of possible classes.
unique: A `bool`. Ignored.
unique.
seed: An `int`. An operation-specific seed. Default is 0.
name: A name for the operation (optional).
Returns:
sampled_candidates: A tensor of type `int64` and shape `[num_sampled]`.
This operation deterministically returns the entire range
`[0, num_sampled]`.
true_expected_count: A tensor of type `float`. Same shape as
`true_classes`. The expected counts under the sampling distribution
of each of `true_classes`. All returned values are 1.0.
sampled_expected_count: A tensor of type `float`. Same shape as
`sampled_candidates`. The expected counts under the sampling distribution
of each of `sampled_candidates`. All returned values are 1.0.
"""
seed1, seed2 = random_seed.get_seed(seed)
return gen_candidate_sampling_ops.all_candidate_sampler(
true_classes, num_true, num_sampled, unique, seed=seed1, seed2=seed2,
name=name)
@tf_export('nn.compute_accidental_hits')
def compute_accidental_hits(true_classes, sampled_candidates, num_true,
seed=None, name=None):
"""Compute the position ids in `sampled_candidates` matching `true_classes`.
In Candidate Sampling, this operation facilitates virtually removing
sampled classes which happen to match target classes. This is done
in Sampled Softmax and Sampled Logistic.
See our [Candidate Sampling Algorithms
Reference](http://www.tensorflow.org/extras/candidate_sampling.pdf).
We presuppose that the `sampled_candidates` are unique.
We call it an 'accidental hit' when one of the target classes
matches one of the sampled classes. This operation reports
accidental hits as triples `(index, id, weight)`, where `index`
represents the row number in `true_classes`, `id` represents the
position in `sampled_candidates`, and weight is `-FLOAT_MAX`.
The result of this op should be passed through a `sparse_to_dense`
operation, then added to the logits of the sampled classes. This
removes the contradictory effect of accidentally sampling the true
target classes as noise classes for the same example.
Args:
true_classes: A `Tensor` of type `int64` and shape `[batch_size,
num_true]`. The target classes.
sampled_candidates: A tensor of type `int64` and shape `[num_sampled]`.
The sampled_candidates output of CandidateSampler.
num_true: An `int`. The number of target classes per training example.
seed: An `int`. An operation-specific seed. Default is 0.
name: A name for the operation (optional).
Returns:
indices: A `Tensor` of type `int32` and shape `[num_accidental_hits]`.
Values indicate rows in `true_classes`.
ids: A `Tensor` of type `int64` and shape `[num_accidental_hits]`.
Values indicate positions in `sampled_candidates`.
weights: A `Tensor` of type `float` and shape `[num_accidental_hits]`.
Each value is `-FLOAT_MAX`.
"""
seed1, seed2 = random_seed.get_seed(seed)
return gen_candidate_sampling_ops.compute_accidental_hits(
true_classes, sampled_candidates, num_true, seed=seed1, seed2=seed2,
name=name)
|
tensorflow-master
|
tensorflow/python/ops/candidate_sampling_ops.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# =============================================================================
"""Functional operations."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.eager import context
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import ops
from tensorflow.python.framework import sparse_tensor
from tensorflow.python.framework import tensor_shape
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import tensor_array_ops
from tensorflow.python.ops import variable_scope as vs
from tensorflow.python.platform import tf_logging as logging
from tensorflow.python.util import nest
from tensorflow.python.util.tf_export import tf_export
@tf_export("map_fn")
def map_fn(fn, elems, dtype=None, parallel_iterations=None, back_prop=True,
swap_memory=False, infer_shape=True, name=None):
"""map on the list of tensors unpacked from `elems` on dimension 0.
The simplest version of `map_fn` repeatedly applies the callable `fn` to a
sequence of elements from first to last. The elements are made of the
tensors unpacked from `elems`. `dtype` is the data type of the return
value of `fn`. Users must provide `dtype` if it is different from
the data type of `elems`.
Suppose that `elems` is unpacked into `values`, a list of tensors. The shape
of the result tensor is `[values.shape[0]] + fn(values[0]).shape`.
This method also allows multi-arity `elems` and output of `fn`. If `elems`
is a (possibly nested) list or tuple of tensors, then each of these tensors
must have a matching first (unpack) dimension. The signature of `fn` may
match the structure of `elems`. That is, if `elems` is
`(t1, [t2, t3, [t4, t5]])`, then an appropriate signature for `fn` is:
`fn = lambda (t1, [t2, t3, [t4, t5]]):`.
Furthermore, `fn` may emit a different structure than its input. For example,
`fn` may look like: `fn = lambda t1: return (t1 + 1, t1 - 1)`. In this case,
the `dtype` parameter is not optional: `dtype` must be a type or (possibly
nested) tuple of types matching the output of `fn`.
To apply a functional operation to the nonzero elements of a SparseTensor
one of the following methods is recommended. First, if the function is
expressible as TensorFlow ops, use
```python
result = SparseTensor(input.indices, fn(input.values), input.dense_shape)
```
If, however, the function is not expressible as a TensorFlow op, then use
```python
result = SparseTensor(
input.indices, map_fn(fn, input.values), input.dense_shape)
```
instead.
When executing eagerly, map_fn does not execute in parallel even if
`parallel_iterations` is set to a value > 1. You can still get the
performance benefits of running a function in parallel by using the
`tf.contrib.eager.defun` decorator,
```python
# Assume the function being used in map_fn is fn.
# To ensure map_fn calls fn in parallel, use the defun decorator.
@tf.contrib.eager.defun
def func(tensor):
return tf.map_fn(fn, tensor)
```
Note that if you use the defun decorator, any non-TensorFlow Python code
that you may have written in your function won't get executed. See
`tf.contrib.eager.defun` for more details. The recommendation would be to
debug without defun but switch to defun to get performance benefits of
running map_fn in parallel.
Args:
fn: The callable to be performed. It accepts one argument, which will
have the same (possibly nested) structure as `elems`. Its output
must have the same structure as `dtype` if one is provided, otherwise
it must have the same structure as `elems`.
elems: A tensor or (possibly nested) sequence of tensors, each of which
will be unpacked along their first dimension. The nested sequence
of the resulting slices will be applied to `fn`.
dtype: (optional) The output type(s) of `fn`. If `fn` returns a structure
of Tensors differing from the structure of `elems`, then `dtype` is not
optional and must have the same structure as the output of `fn`.
parallel_iterations: (optional) The number of iterations allowed to run
in parallel. When graph building, the default value is 10. While executing
eagerly, the default value is set to 1.
back_prop: (optional) True enables support for back propagation.
swap_memory: (optional) True enables GPU-CPU memory swapping.
infer_shape: (optional) False disables tests for consistent output shapes.
name: (optional) Name prefix for the returned tensors.
Returns:
A tensor or (possibly nested) sequence of tensors. Each tensor packs the
results of applying `fn` to tensors unpacked from `elems` along the first
dimension, from first to last.
Raises:
TypeError: if `fn` is not callable or the structure of the output of
`fn` and `dtype` do not match, or if elems is a SparseTensor.
ValueError: if the lengths of the output of `fn` and `dtype` do not match.
Examples:
```python
elems = np.array([1, 2, 3, 4, 5, 6])
squares = map_fn(lambda x: x * x, elems)
# squares == [1, 4, 9, 16, 25, 36]
```
```python
elems = (np.array([1, 2, 3]), np.array([-1, 1, -1]))
alternate = map_fn(lambda x: x[0] * x[1], elems, dtype=tf.int64)
# alternate == [-1, 2, -3]
```
```python
elems = np.array([1, 2, 3])
alternates = map_fn(lambda x: (x, -x), elems, dtype=(tf.int64, tf.int64))
# alternates[0] == [1, 2, 3]
# alternates[1] == [-1, -2, -3]
```
"""
if not callable(fn):
raise TypeError("fn must be callable.")
if isinstance(elems, sparse_tensor.SparseTensor):
raise TypeError(
"To perform a map on the values of a sparse tensor use either "
" SparseTensor(input.indices, fn(input.values), input.dense_shape) or "
" SparseTensor(input.indices, map_fn(fn, input.values), "
"input.dense_shape)")
in_graph_mode = not context.executing_eagerly()
# Set the default number of parallel_iterations depending on graph/eager mode.
if in_graph_mode and not parallel_iterations:
parallel_iterations = 10
elif not in_graph_mode and not parallel_iterations:
parallel_iterations = 1
if not in_graph_mode and parallel_iterations > 1:
logging.log_first_n(logging.WARN, "Setting parallel_iterations > 1 has no "
"effect when executing eagerly. Consider calling map_fn"
" with tf.contrib.eager.defun to execute fn in "
"parallel.", 1)
parallel_iterations = 1
input_is_sequence = nest.is_sequence(elems)
input_flatten = lambda x: nest.flatten(x) if input_is_sequence else [x]
def input_pack(x):
return nest.pack_sequence_as(elems, x) if input_is_sequence else x[0]
if dtype is None:
output_is_sequence = input_is_sequence
output_flatten = input_flatten
output_pack = input_pack
else:
output_is_sequence = nest.is_sequence(dtype)
output_flatten = lambda x: nest.flatten(x) if output_is_sequence else [x]
def output_pack(x):
return (nest.pack_sequence_as(dtype, x)
if output_is_sequence else x[0])
elems_flat = input_flatten(elems)
with ops.name_scope(name, "map", elems_flat):
# TODO(akshayka): Remove the in_graph_mode check once caching devices are
# supported in Eager
if in_graph_mode:
# Any get_variable calls in fn will cache the first call locally
# and not issue repeated network I/O requests for each iteration.
varscope = vs.get_variable_scope()
varscope_caching_device_was_none = False
if varscope.caching_device is None:
# TODO(ebrevdo): Change to using colocate_with here and in other
# methods.
varscope.set_caching_device(lambda op: op.device)
varscope_caching_device_was_none = True
elems_flat = [
ops.convert_to_tensor(elem, name="elem") for elem in elems_flat]
dtype = dtype or input_pack([elem.dtype for elem in elems_flat])
dtype_flat = output_flatten(dtype)
# Convert elems to tensor array. n may be known statically.
static_shape = elems_flat[0].shape
if static_shape.ndims is not None and static_shape.ndims < 1:
if len(elems_flat) == 1:
raise ValueError("elems must be a 1+ dimensional Tensor, not a scalar")
else:
raise ValueError(
"elements in elems must be 1+ dimensional Tensors, not scalars"
)
n = (tensor_shape.dimension_value(static_shape[0])
or array_ops.shape(elems_flat[0])[0])
# TensorArrays are always flat
elems_ta = [
tensor_array_ops.TensorArray(dtype=elem.dtype,
size=n,
dynamic_size=False,
infer_shape=True)
for elem in elems_flat]
# Unpack elements
elems_ta = [
elem_ta.unstack(elem) for elem_ta, elem in zip(elems_ta, elems_flat)]
i = constant_op.constant(0)
accs_ta = [
tensor_array_ops.TensorArray(dtype=dt,
size=n,
dynamic_size=False,
infer_shape=infer_shape)
for dt in dtype_flat]
def compute(i, tas):
"""The loop body of map_fn.
Args:
i: the loop counter
tas: the flat TensorArray accumulator list
Returns:
(i + 1, tas): the updated counter + updated TensorArrays
Raises:
TypeError: if dtype and packed_fn_values structure do not match
ValueType: if dtype and packed_fn_values lengths do not match
"""
packed_values = input_pack([elem_ta.read(i) for elem_ta in elems_ta])
packed_fn_values = fn(packed_values)
nest.assert_same_structure(dtype or elems, packed_fn_values)
flat_fn_values = output_flatten(packed_fn_values)
tas = [ta.write(i, value) for (ta, value) in zip(tas, flat_fn_values)]
return (i + 1, tas)
_, r_a = control_flow_ops.while_loop(
lambda i, _: i < n, compute, (i, accs_ta),
parallel_iterations=parallel_iterations,
back_prop=back_prop,
swap_memory=swap_memory,
maximum_iterations=n)
results_flat = [r.stack() for r in r_a]
n_static = tensor_shape.Dimension(tensor_shape.dimension_value(
elems_flat[0].get_shape().with_rank_at_least(1)[0]))
for elem in elems_flat[1:]:
n_static.merge_with(tensor_shape.Dimension(tensor_shape.dimension_value(
elem.get_shape().with_rank_at_least(1)[0])))
for r in results_flat:
r.set_shape(tensor_shape.TensorShape(n_static).concatenate(
r.get_shape()[1:]))
# TODO(akshayka): Remove the in_graph_mode check once caching devices are
# supported in Eager
if in_graph_mode and varscope_caching_device_was_none:
varscope.set_caching_device(None)
return output_pack(results_flat)
|
tensorflow-master
|
tensorflow/python/ops/map_fn.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for Collective Operations."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.core.protobuf import config_pb2
from tensorflow.core.protobuf import rewriter_config_pb2
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import ops
from tensorflow.python.framework import test_util
from tensorflow.python.ops import collective_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.platform import test
class CollectiveOpTest(test.TestCase):
def _testCollectiveReduce(self, t0, t1, expected, set_graph_key):
group_key = 1
instance_key = 1
with self.session(
config=config_pb2.ConfigProto(device_count={'CPU': 2})) as sess:
with ops.device('/CPU:0'):
in0 = constant_op.constant(t0)
colred0 = collective_ops.all_reduce(in0, 2, group_key, instance_key,
'Add', 'Div')
with ops.device('/CPU:1'):
in1 = constant_op.constant(t1)
colred1 = collective_ops.all_reduce(in1, 2, group_key, instance_key,
'Add', 'Div')
run_options = config_pb2.RunOptions()
if set_graph_key:
run_options.experimental.collective_graph_key = 1
results = sess.run([colred0, colred1], options=run_options)
self.assertAllClose(results[0], expected, rtol=1e-5, atol=1e-5)
self.assertAllClose(results[1], expected, rtol=1e-5, atol=1e-5)
def _testMultipleConcurrentCollectiveReduce(self, t0, t1, expected):
group_key = 1
group_size = 2
num_instances = 2
all_reduces = []
config = config_pb2.ConfigProto(device_count={'CPU': group_size})
config.experimental.collective_deterministic_sequential_execution = True
with self.session(config=config) as sess:
for cpu in range(group_size):
with ops.device('/CPU:%d' % cpu):
in_tensor = constant_op.constant(t0 if cpu == 0 else t1)
for instance in range(num_instances):
all_reduces.append(collective_ops.all_reduce(
in_tensor, group_size, group_key, instance, 'Add', 'Div'))
results = sess.run(all_reduces)
for i in range(group_size * num_instances):
self.assertAllClose(results[i], expected, rtol=1e-5, atol=1e-5)
@test_util.run_deprecated_v1
def testCollectiveReduce(self):
self._testCollectiveReduce([0.1, 1.1, 2.1, 3.1, 4.1, 5.1, 6.1, 7.1],
[0.3, 1.3, 2.3, 3.3, 4.3, 5.3, 6.3, 7.3],
[0.2, 1.2, 2.2, 3.2, 4.2, 5.2, 6.2, 7.2], True)
@test_util.run_deprecated_v1
def testCollectiveAutoGraphKey(self):
self._testCollectiveReduce([0.1, 1.1, 2.1, 3.1, 4.1, 5.1, 6.1, 7.1],
[0.3, 1.3, 2.3, 3.3, 4.3, 5.3, 6.3, 7.3],
[0.2, 1.2, 2.2, 3.2, 4.2, 5.2, 6.2, 7.2], False)
@test_util.run_deprecated_v1
def testCollectiveMultipleConcurrentReduce(self):
self._testMultipleConcurrentCollectiveReduce(
[0.1, 1.1, 2.1, 3.1, 4.1, 5.1, 6.1, 7.1],
[0.3, 1.3, 2.3, 3.3, 4.3, 5.3, 6.3, 7.3],
[0.2, 1.2, 2.2, 3.2, 4.2, 5.2, 6.2, 7.2])
@test_util.run_deprecated_v1
def testWhileWithScopedAllocator(self):
group_size = 2
group_key = 1
instance_key0 = 1
instance_key1 = 2
config = config_pb2.ConfigProto(device_count={'CPU': group_size})
rewrite_options = config.graph_options.rewrite_options
rewrite_options.scoped_allocator_optimization = (
rewriter_config_pb2.RewriterConfig.ON)
del rewrite_options.scoped_allocator_opts.enable_op[:]
rewrite_options.scoped_allocator_opts.enable_op.append('CollectiveReduce')
with self.session(config=config) as sess:
run_ops = []
for i in range(group_size):
with ops.device('CPU:%d' % i):
constant = constant_op.constant(0.)
cond = lambda i: math_ops.less(i, 10.)
body = lambda i: math_ops.add(i, 1.)
input0 = control_flow_ops.while_loop(cond, body, [constant])
input1 = math_ops.add(constant, 5)
colred0 = collective_ops.all_reduce(input0, group_size, group_key,
instance_key0, 'Add', 'Id')
colred1 = collective_ops.all_reduce(input1, group_size, group_key,
instance_key1, 'Add', 'Id')
run_ops.append(math_ops.add_n([colred0, colred1]))
results = sess.run(run_ops)
self.assertEqual(results, [30., 30.])
@test_util.run_deprecated_v1
def testCollectiveReduceScalar(self):
self._testCollectiveReduce(0.1, 0.3, 0.2, True)
def _testCollectiveBroadcast(self, t0):
group_key = 1
instance_key = 1
with self.session(
config=config_pb2.ConfigProto(device_count={'CPU': 2})) as sess:
with ops.device('/CPU:0'):
in0 = constant_op.constant(t0)
out0 = collective_ops.broadcast_send(in0, in0.shape, in0.dtype,
2, group_key, instance_key)
with ops.device('/CPU:1'):
c1 = constant_op.constant(t0)
out1 = collective_ops.broadcast_recv(c1.shape, c1.dtype,
2, group_key, instance_key)
run_options = config_pb2.RunOptions()
run_options.experimental.collective_graph_key = 1
results = sess.run([out0, out1], options=run_options)
self.assertAllClose(results[0], t0, rtol=1e-5, atol=1e-5)
self.assertAllClose(results[1], t0, rtol=1e-5, atol=1e-5)
@test_util.run_deprecated_v1
def testCollectiveBroadcast(self):
self._testCollectiveBroadcast([0.1, 1.1, 2.1, 3.1, 4.1, 5.1, 6.1, 7.1])
def _testCollectiveGather(self, t0, t1, expected, set_graph_key):
group_key = 1
instance_key = 1
with self.session(
config=config_pb2.ConfigProto(device_count={'CPU': 2})) as sess:
with ops.device('/CPU:0'):
in0 = constant_op.constant(t0)
colred0 = collective_ops.all_gather(in0, 2, group_key, instance_key)
with ops.device('/CPU:1'):
in1 = constant_op.constant(t1)
colred1 = collective_ops.all_gather(in1, 2, group_key, instance_key)
run_options = config_pb2.RunOptions()
if set_graph_key:
run_options.experimental.collective_graph_key = 1
results = sess.run([colred0, colred1], options=run_options)
self.assertAllClose(results[0], expected, rtol=1e-5, atol=1e-5)
self.assertAllClose(results[1], expected, rtol=1e-5, atol=1e-5)
@test_util.run_deprecated_v1
def testCollectiveGather(self):
self._testCollectiveGather([0, 1, 2, 3, 4, 5, 6, 7],
[10, 11, 12, 13, 14, 15, 16, 17],
[0, 1, 2, 3, 4, 5, 6, 7,
10, 11, 12, 13, 14, 15, 16, 17],
True)
self._testCollectiveGather([[0, 1, 2, 3], [4, 5, 6, 7]],
[[10, 11, 12, 13], [14, 15, 16, 17]],
[[0, 1, 2, 3], [4, 5, 6, 7],
[10, 11, 12, 13], [14, 15, 16, 17]],
True)
self._testCollectiveGather([[[0, 1], [2, 3]], [[4, 5], [6, 7]]],
[[[10, 11], [12, 13]], [[14, 15], [16, 17]]],
[[[0, 1], [2, 3]], [[4, 5], [6, 7]],
[[10, 11], [12, 13]], [[14, 15], [16, 17]]],
True)
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/ops/collective_ops_test.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
# pylint: disable=g-short-docstring-punctuation
"""Sparse Tensor Representation.
See also `tf.SparseTensor`.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import numbers
import numpy as np
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import sparse_tensor
from tensorflow.python.framework import tensor_shape
from tensorflow.python.framework import tensor_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import check_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import gen_sparse_ops
from tensorflow.python.ops import math_ops
# go/tf-wildcard-import
# pylint: disable=wildcard-import
from tensorflow.python.ops.gen_sparse_ops import *
# pylint: enable=wildcard-import
from tensorflow.python.util import compat
from tensorflow.python.util import deprecation
from tensorflow.python.util import dispatch
from tensorflow.python.util import tf_inspect
from tensorflow.python.util.tf_export import get_canonical_name_for_symbol
from tensorflow.python.util.tf_export import tf_export
def _convert_to_sparse_tensor(sp_input):
"""Convert `sp_input` to `SparseTensor` and return it.
Args:
sp_input: `SparseTensor` or `SparseTensorValue`.
Returns:
`sp_input` converted to `SparseTensor`.
Raises:
ValueError: if `sp_input` is neither `SparseTensor` nor `SparseTensorValue`.
"""
if isinstance(sp_input, sparse_tensor.SparseTensorValue):
return sparse_tensor.SparseTensor.from_value(sp_input)
if not isinstance(sp_input, sparse_tensor.SparseTensor):
raise TypeError("Input must be a SparseTensor.")
return sp_input
def _convert_to_sparse_tensors(sp_inputs):
"""Convert `sp_inputs` to `SparseTensor` objects and return them.
Args:
sp_inputs: `list` or `tuple` of `SparseTensor` or `SparseTensorValue`
objects.
Returns:
`sp_inputs` converted to `SparseTensor` objects.
Raises:
ValueError: if any item in `sp_inputs` is neither `SparseTensor` nor
`SparseTensorValue`.
"""
if isinstance(sp_inputs, list):
return [_convert_to_sparse_tensor(sp_input) for sp_input in sp_inputs]
if isinstance(sp_inputs, tuple):
return (_convert_to_sparse_tensor(sp_input) for sp_input in sp_inputs)
raise TypeError("Inputs must be a list or tuple.")
def _make_int64_tensor(value, name):
if isinstance(value, compat.integral_types):
return ops.convert_to_tensor(value, name=name, dtype=dtypes.int64)
if not isinstance(value, ops.Tensor):
raise TypeError("{} must be an integer value".format(name))
if value.dtype == dtypes.int64:
return value
return math_ops.cast(value, dtypes.int64)
@tf_export("sparse.expand_dims")
def sparse_expand_dims(sp_input, axis=None, name=None):
"""Inserts a dimension of 1 into a tensor's shape.
Given a tensor `sp_input`, this operation inserts a dimension of 1 at the
dimension index `axis` of `sp_input`'s shape. The dimension index `axis`
starts at zero; if you specify a negative number for `axis` it is counted
backwards from the end.
Args:
sp_input: A `SparseTensor`.
axis: 0-D (scalar). Specifies the dimension index at which to expand the
shape of `input`. Must be in the range `[-rank(sp_input) - 1,
rank(sp_input)]`.
name: The name of the output `SparseTensor`.
Returns:
A `SparseTensor` with the same data as `sp_input`, but its shape has an
additional dimension of size 1 added.
"""
rank = sp_input.dense_shape.get_shape()[0]
axis = -1 if axis is None else axis
with ops.name_scope(name, default_name="expand_dims", values=[sp_input]):
if isinstance(axis, compat.integral_types):
axis = ops.convert_to_tensor(axis, name="axis", dtype=dtypes.int32)
elif not isinstance(axis, ops.Tensor):
raise TypeError("axis must be an integer value in range [-rank(sp_input)"
" - 1, rank(sp_input)]")
# Convert axis to a positive value if it is negative.
axis = array_ops.where(axis >= 0, axis, axis + rank + 1)
# Create the new column of indices for the sparse tensor by slicing
# the indices and inserting a new column of indices for the new dimension.
column_size = array_ops.shape(sp_input.indices)[0]
new_index = array_ops.zeros([column_size, 1], dtype=dtypes.int64)
indices_before = array_ops.slice(sp_input.indices, [0, 0], [-1, axis])
indices_after = array_ops.slice(sp_input.indices, [0, axis], [-1, -1])
indices = array_ops.concat(
[indices_before, new_index, indices_after], axis=1)
# Create the new dense shape by splicing the tensor [1] in the correct
# dimension of the existing shape.
shape_before = array_ops.slice(sp_input.dense_shape, [0], [axis])
shape_after = array_ops.slice(sp_input.dense_shape, [axis], [-1])
new_shape = ops.convert_to_tensor([1], name="new_shape", dtype=dtypes.int64)
shape = array_ops.concat([shape_before, new_shape, shape_after], axis=0)
# Create the output sparse tensor.
return sparse_tensor.SparseTensor(
indices=indices, values=sp_input.values, dense_shape=shape)
@tf_export("sparse.eye")
def sparse_eye(num_rows,
num_columns=None,
dtype=dtypes.float32,
name=None):
"""Creates a two-dimensional sparse tensor with ones along the diagonal.
Args:
num_rows: Non-negative integer or `int32` scalar `tensor` giving the number
of rows in the resulting matrix.
num_columns: Optional non-negative integer or `int32` scalar `tensor` giving
the number of columns in the resulting matrix. Defaults to `num_rows`.
dtype: The type of element in the resulting `Tensor`.
name: A name for this `Op`. Defaults to "eye".
Returns:
A `SparseTensor` of shape [num_rows, num_columns] with ones along the
diagonal.
"""
with ops.name_scope(name, default_name="eye", values=[num_rows, num_columns]):
num_rows = _make_int64_tensor(num_rows, "num_rows")
num_columns = num_rows if num_columns is None else _make_int64_tensor(
num_columns, "num_columns")
# Create the sparse tensor.
diag_size = math_ops.minimum(num_rows, num_columns)
diag_range = math_ops.range(diag_size, dtype=dtypes.int64)
return sparse_tensor.SparseTensor(
indices=array_ops.stack([diag_range, diag_range], axis=1),
values=array_ops.ones(diag_size, dtype=dtype),
dense_shape=[num_rows, num_columns])
# pylint: disable=protected-access
@tf_export(v1=["sparse.concat", "sparse_concat"])
@deprecation.deprecated_endpoints("sparse_concat")
@deprecation.deprecated_args(
None, "concat_dim is deprecated, use axis instead", "concat_dim")
def sparse_concat(axis,
sp_inputs,
name=None,
expand_nonconcat_dim=False,
concat_dim=None,
expand_nonconcat_dims=None):
"""Concatenates a list of `SparseTensor` along the specified dimension.
Concatenation is with respect to the dense versions of each sparse input.
It is assumed that each inputs is a `SparseTensor` whose elements are ordered
along increasing dimension number.
If expand_nonconcat_dim is False, all inputs' shapes must match, except for
the concat dimension. If expand_nonconcat_dim is True, then inputs' shapes are
allowed to vary among all inputs.
The `indices`, `values`, and `shapes` lists must have the same length.
If expand_nonconcat_dim is False, then the output shape is identical to the
inputs', except along the concat dimension, where it is the sum of the inputs'
sizes along that dimension.
If expand_nonconcat_dim is True, then the output shape along the non-concat
dimensions will be expand to be the largest among all inputs, and it is the
sum of the inputs sizes along the concat dimension.
The output elements will be resorted to preserve the sort order along
increasing dimension number.
This op runs in `O(M log M)` time, where `M` is the total number of non-empty
values across all inputs. This is due to the need for an internal sort in
order to concatenate efficiently across an arbitrary dimension.
For example, if `axis = 1` and the inputs are
sp_inputs[0]: shape = [2, 3]
[0, 2]: "a"
[1, 0]: "b"
[1, 1]: "c"
sp_inputs[1]: shape = [2, 4]
[0, 1]: "d"
[0, 2]: "e"
then the output will be
shape = [2, 7]
[0, 2]: "a"
[0, 4]: "d"
[0, 5]: "e"
[1, 0]: "b"
[1, 1]: "c"
Graphically this is equivalent to doing
[ a] concat [ d e ] = [ a d e ]
[b c ] [ ] [b c ]
Another example, if 'axis = 1' and the inputs are
sp_inputs[0]: shape = [3, 3]
[0, 2]: "a"
[1, 0]: "b"
[2, 1]: "c"
sp_inputs[1]: shape = [2, 4]
[0, 1]: "d"
[0, 2]: "e"
if expand_nonconcat_dim = False, this will result in an error. But if
expand_nonconcat_dim = True, this will result in:
shape = [3, 7]
[0, 2]: "a"
[0, 4]: "d"
[0, 5]: "e"
[1, 0]: "b"
[2, 1]: "c"
Graphically this is equivalent to doing
[ a] concat [ d e ] = [ a d e ]
[b ] [ ] [b ]
[ c ] [ c ]
Args:
axis: Dimension to concatenate along. Must be in range [-rank, rank),
where rank is the number of dimensions in each input `SparseTensor`.
sp_inputs: List of `SparseTensor` to concatenate.
name: A name prefix for the returned tensors (optional).
expand_nonconcat_dim: Whether to allow the expansion in the non-concat
dimensions. Defaulted to False.
concat_dim: The old (deprecated) name for axis.
expand_nonconcat_dims: alias for expand_nonconcat_dim
Returns:
A `SparseTensor` with the concatenated output.
Raises:
TypeError: If `sp_inputs` is not a list of `SparseTensor`.
"""
expand_nonconcat_dim = deprecation.deprecated_argument_lookup(
"expand_nonconcat_dims", expand_nonconcat_dims,
"expand_nonconcat_dim", expand_nonconcat_dim)
if expand_nonconcat_dims is not None:
expand_nonconcat_dim = expand_nonconcat_dims
axis = deprecation.deprecated_argument_lookup("axis", axis, "concat_dim",
concat_dim)
return sparse_concat_v2(axis, sp_inputs, expand_nonconcat_dim, name)
@tf_export("sparse.concat", v1=[])
def sparse_concat_v2(axis, sp_inputs, expand_nonconcat_dims=False, name=None): # pylint: disable=missing-docstring
sp_inputs = _convert_to_sparse_tensors(sp_inputs)
if len(sp_inputs) == 1: # Degenerate case of one tensor.
return sp_inputs[0]
inds = [sp_input.indices for sp_input in sp_inputs]
vals = [sp_input.values for sp_input in sp_inputs]
shapes = [sp_input.dense_shape for sp_input in sp_inputs]
if expand_nonconcat_dims:
max_shape = math_ops.reduce_max(
array_ops.concat(
[array_ops.reshape(shape, [1, -1]) for shape in shapes], 0), 0)
shapes = [
array_ops.concat([
max_shape[:axis], shape[-1:]
if axis == -1 else shape[axis:axis + 1], []
if axis == -1 else max_shape[axis + 1:]
], 0) for shape in shapes
]
output_ind, output_val, output_shape = (
gen_sparse_ops.sparse_concat(inds, vals, shapes, axis, name=name))
shapes_value = [tensor_util.constant_value(shape) for shape in shapes]
if shapes_value and all(shape is not None for shape in shapes_value):
dim = sum(shape[axis] for shape in shapes_value)
output_shape = shapes_value[0]
output_shape[axis] = dim
output_shape = ops.convert_to_tensor(output_shape)
return sparse_tensor.SparseTensor(output_ind, output_val, output_shape)
sparse_concat_v2.__doc__ = sparse_concat.__doc__.replace(
" concat_dim: The old (deprecated) name for axis.\n", "")
@tf_export(v1=["sparse.add", "sparse_add"])
@deprecation.deprecated_endpoints("sparse_add")
@deprecation.deprecated_args(
None, "thresh is deprecated, use threshold instead", "thresh")
def sparse_add(a, b, threshold=None, thresh=None):
"""Adds two tensors, at least one of each is a `SparseTensor`.
If one `SparseTensor` and one `Tensor` are passed in, returns a `Tensor`. If
both arguments are `SparseTensor`s, this returns a `SparseTensor`. The order
of arguments does not matter. Use vanilla `tf.add()` for adding two dense
`Tensor`s.
The shapes of the two operands must match: broadcasting is not supported.
The indices of any input `SparseTensor` are assumed ordered in standard
lexicographic order. If this is not the case, before this step run
`SparseReorder` to restore index ordering.
If both arguments are sparse, we perform "clipping" as follows. By default,
if two values sum to zero at some index, the output `SparseTensor` would still
include that particular location in its index, storing a zero in the
corresponding value slot. To override this, callers can specify `thresh`,
indicating that if the sum has a magnitude strictly smaller than `thresh`, its
corresponding value and index would then not be included. In particular,
`thresh == 0.0` (default) means everything is kept and actual thresholding
happens only for a positive value.
For example, suppose the logical sum of two sparse operands is (densified):
[ 2]
[.1 0]
[ 6 -.2]
Then,
* `thresh == 0` (the default): all 5 index/value pairs will be returned.
* `thresh == 0.11`: only .1 and 0 will vanish, and the remaining three
index/value pairs will be returned.
* `thresh == 0.21`: .1, 0, and -.2 will vanish.
Args:
a: The first operand; `SparseTensor` or `Tensor`.
b: The second operand; `SparseTensor` or `Tensor`. At least one operand
must be sparse.
threshold: An optional 0-D `Tensor` (defaults to `0`). The magnitude
threshold that determines if an output value/index pair takes space. Its
dtype should match that of the values if they are real; if the latter are
complex64/complex128, then the dtype should be float32/float64,
correspondingly.
thresh: Deprecated alias for `threshold`.
Returns:
A `SparseTensor` or a `Tensor`, representing the sum.
Raises:
TypeError: If both `a` and `b` are `Tensor`s. Use `tf.add()` instead.
"""
threshold = deprecation.deprecated_argument_lookup("threshold", threshold,
"thresh", thresh)
if threshold is None:
threshold = 0
return sparse_add_v2(a, b, threshold)
@tf_export("sparse.add", v1=[])
def sparse_add_v2(a, b, threshold=0):
"""Adds two tensors, at least one of each is a `SparseTensor`.
If one `SparseTensor` and one `Tensor` are passed in, returns a `Tensor`. If
both arguments are `SparseTensor`s, this returns a `SparseTensor`. The order
of arguments does not matter. Use vanilla `tf.add()` for adding two dense
`Tensor`s.
The shapes of the two operands must match: broadcasting is not supported.
The indices of any input `SparseTensor` are assumed ordered in standard
lexicographic order. If this is not the case, before this step run
`SparseReorder` to restore index ordering.
If both arguments are sparse, we perform "clipping" as follows. By default,
if two values sum to zero at some index, the output `SparseTensor` would still
include that particular location in its index, storing a zero in the
corresponding value slot. To override this, callers can specify `threshold`,
indicating that if the sum has a magnitude strictly smaller than `threshold`,
its corresponding value and index would then not be included. In particular,
`threshold == 0.0` (default) means everything is kept and actual thresholding
happens only for a positive value.
For example, suppose the logical sum of two sparse operands is (densified):
[ 2]
[.1 0]
[ 6 -.2]
Then,
* `threshold == 0` (the default): all 5 index/value pairs will be
returned.
* `threshold == 0.11`: only .1 and 0 will vanish, and the remaining three
index/value pairs will be returned.
* `threshold == 0.21`: .1, 0, and -.2 will vanish.
Args:
a: The first operand; `SparseTensor` or `Tensor`.
b: The second operand; `SparseTensor` or `Tensor`. At least one operand
must be sparse.
threshold: A 0-D `Tensor`. The magnitude threshold that determines if an
output value/index pair takes space. Its dtype should match that of the
values if they are real; if the latter are complex64/complex128, then the
dtype should be float32/float64, correspondingly.
Returns:
A `SparseTensor` or a `Tensor`, representing the sum.
Raises:
TypeError: If both `a` and `b` are `Tensor`s. Use `tf.add()` instead.
"""
sparse_classes = (sparse_tensor.SparseTensor, sparse_tensor.SparseTensorValue)
if not any(isinstance(inp, sparse_classes) for inp in [a, b]):
raise TypeError("At least one input should be SparseTensor; do you mean to"
" use tf.add()?")
if all(isinstance(inp, sparse_classes) for inp in [a, b]):
a = _convert_to_sparse_tensor(a)
b = _convert_to_sparse_tensor(b)
threshold = ops.convert_to_tensor(
threshold, dtype=a.values.dtype.real_dtype.base_dtype, name="threshold")
output_ind, output_val, output_shape = (
gen_sparse_ops.sparse_add(a.indices, a.values, a.dense_shape,
b.indices, b.values, b.dense_shape,
threshold))
# Attempt to get output_shape statically.
a.get_shape().assert_is_compatible_with(b.get_shape())
static_shape = array_ops.broadcast_static_shape(a.get_shape(),
b.get_shape())
if static_shape.is_fully_defined():
output_shape = static_shape.as_list()
return sparse_tensor.SparseTensor(output_ind, output_val, output_shape)
else:
# swap to make `a` the SparseTensor.
if isinstance(b, sparse_classes):
a, b = b, a
return gen_sparse_ops.sparse_tensor_dense_add(a.indices, a.values,
a.dense_shape, b)
@tf_export("sparse.cross")
def sparse_cross(inputs, name=None):
"""Generates sparse cross from a list of sparse and dense tensors.
For example, if the inputs are
* inputs[0]: SparseTensor with shape = [2, 2]
[0, 0]: "a"
[1, 0]: "b"
[1, 1]: "c"
* inputs[1]: SparseTensor with shape = [2, 1]
[0, 0]: "d"
[1, 0]: "e"
* inputs[2]: Tensor [["f"], ["g"]]
then the output will be:
shape = [2, 2]
[0, 0]: "a_X_d_X_f"
[1, 0]: "b_X_e_X_g"
[1, 1]: "c_X_e_X_g"
Args:
inputs: An iterable of `Tensor` or `SparseTensor`.
name: Optional name for the op.
Returns:
A `SparseTensor` of type `string`.
"""
return _sparse_cross_internal(inputs=inputs, hashed_output=False, name=name)
_sparse_cross = sparse_cross
@tf_export("sparse.cross_hashed")
def sparse_cross_hashed(inputs, num_buckets=0, hash_key=None, name=None):
"""Generates hashed sparse cross from a list of sparse and dense tensors.
For example, if the inputs are
* inputs[0]: SparseTensor with shape = [2, 2]
[0, 0]: "a"
[1, 0]: "b"
[1, 1]: "c"
* inputs[1]: SparseTensor with shape = [2, 1]
[0, 0]: "d"
[1, 0]: "e"
* inputs[2]: Tensor [["f"], ["g"]]
then the output will be:
shape = [2, 2]
[0, 0]: FingerprintCat64(
Fingerprint64("f"), FingerprintCat64(
Fingerprint64("d"), Fingerprint64("a")))
[1, 0]: FingerprintCat64(
Fingerprint64("g"), FingerprintCat64(
Fingerprint64("e"), Fingerprint64("b")))
[1, 1]: FingerprintCat64(
Fingerprint64("g"), FingerprintCat64(
Fingerprint64("e"), Fingerprint64("c")))
Args:
inputs: An iterable of `Tensor` or `SparseTensor`.
num_buckets: An `int` that is `>= 0`.
output = hashed_value%num_buckets if num_buckets > 0 else hashed_value.
hash_key: Integer hash_key that will be used by the `FingerprintCat64`
function. If not given, will use a default key.
name: Optional name for the op.
Returns:
A `SparseTensor` of type `int64`.
"""
return _sparse_cross_internal(
inputs=inputs,
hashed_output=True,
num_buckets=num_buckets,
hash_key=hash_key,
name=name)
_sparse_cross_hashed = sparse_cross_hashed
_DEFAULT_HASH_KEY = 0xDECAFCAFFE
def _sparse_cross_internal(inputs,
hashed_output=False,
num_buckets=0,
hash_key=None,
name=None):
"""See gen_sparse_ops.sparse_cross."""
if not isinstance(inputs, list):
raise TypeError("Inputs must be a list")
if not all(
isinstance(i, sparse_tensor.SparseTensor) or isinstance(i, ops.Tensor)
for i in inputs):
raise TypeError("All inputs must be SparseTensors")
sparse_inputs = [
i for i in inputs if isinstance(i, sparse_tensor.SparseTensor)
]
dense_inputs = [
i for i in inputs if not isinstance(i, sparse_tensor.SparseTensor)
]
indices = [sp_input.indices for sp_input in sparse_inputs]
values = [sp_input.values for sp_input in sparse_inputs]
shapes = [sp_input.dense_shape for sp_input in sparse_inputs]
out_type = dtypes.int64 if hashed_output else dtypes.string
internal_type = dtypes.string
for i in range(len(values)):
if values[i].dtype != dtypes.string:
values[i] = math_ops.cast(values[i], dtypes.int64)
internal_type = dtypes.int64
for i in range(len(dense_inputs)):
if dense_inputs[i].dtype != dtypes.string:
dense_inputs[i] = math_ops.cast(dense_inputs[i], dtypes.int64)
internal_type = dtypes.int64
indices_out, values_out, shape_out = gen_sparse_ops.sparse_cross(
indices=indices,
values=values,
shapes=shapes,
dense_inputs=dense_inputs,
hashed_output=hashed_output,
num_buckets=num_buckets,
hash_key=hash_key or _DEFAULT_HASH_KEY,
out_type=out_type,
internal_type=internal_type,
name=name)
return sparse_tensor.SparseTensor(indices_out, values_out, shape_out)
def sparse_dense_cwise_add(sp_t, dense_t):
"""Adds up a SparseTensor and a dense Tensor, using these special rules:
(1) Broadcasts the dense side to have the same shape as the sparse side, if
eligible;
(2) Then, only the dense values pointed to by the indices of the SparseTensor
participate in the cwise addition.
By the rules, the result is a logical SparseTensor with exactly the same
indices and shape, but possibly with different non-zero values. The output of
this Op is the resultant non-zero values.
Args:
sp_t: the SparseTensor operand.
dense_t: the dense Tensor operand; must have the same dtype and a
broadcast-compatible shape as `sp_t`.
Returns:
output: the SparseTensor output.
"""
result = gen_sparse_ops.sparse_dense_cwise_add(sp_t.indices, sp_t.values,
sp_t.dense_shape, dense_t)
return sparse_tensor.SparseTensor(sp_t.indices, result, sp_t.dense_shape)
@tf_export("sparse.reorder", v1=["sparse.reorder", "sparse_reorder"])
@deprecation.deprecated_endpoints("sparse_reorder")
def sparse_reorder(sp_input, name=None):
"""Reorders a `SparseTensor` into the canonical, row-major ordering.
Note that by convention, all sparse ops preserve the canonical ordering
along increasing dimension number. The only time ordering can be violated
is during manual manipulation of the indices and values to add entries.
Reordering does not affect the shape of the `SparseTensor`.
For example, if `sp_input` has shape `[4, 5]` and `indices` / `values`:
[0, 3]: b
[0, 1]: a
[3, 1]: d
[2, 0]: c
then the output will be a `SparseTensor` of shape `[4, 5]` and
`indices` / `values`:
[0, 1]: a
[0, 3]: b
[2, 0]: c
[3, 1]: d
Args:
sp_input: The input `SparseTensor`.
name: A name prefix for the returned tensors (optional)
Returns:
A `SparseTensor` with the same shape and non-empty values, but in
canonical ordering.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
reordered_ind, reordered_val = (
gen_sparse_ops.sparse_reorder(
sp_input.indices, sp_input.values, sp_input.dense_shape, name=name))
if sp_input.get_shape().is_fully_defined():
dense_shape = sp_input.get_shape().as_list()
else:
dense_shape = array_ops.identity(sp_input.dense_shape)
return sparse_tensor.SparseTensor(reordered_ind, reordered_val, dense_shape)
@tf_export("sparse.reshape", v1=["sparse.reshape", "sparse_reshape"])
@deprecation.deprecated_endpoints("sparse_reshape")
def sparse_reshape(sp_input, shape, name=None):
"""Reshapes a `SparseTensor` to represent values in a new dense shape.
This operation has the same semantics as `reshape` on the represented dense
tensor. The indices of non-empty values in `sp_input` are recomputed based
on the new dense shape, and a new `SparseTensor` is returned containing the
new indices and new shape. The order of non-empty values in `sp_input` is
unchanged.
If one component of `shape` is the special value -1, the size of that
dimension is computed so that the total dense size remains constant. At
most one component of `shape` can be -1. The number of dense elements
implied by `shape` must be the same as the number of dense elements
originally represented by `sp_input`.
For example, if `sp_input` has shape `[2, 3, 6]` and `indices` / `values`:
[0, 0, 0]: a
[0, 0, 1]: b
[0, 1, 0]: c
[1, 0, 0]: d
[1, 2, 3]: e
and `shape` is `[9, -1]`, then the output will be a `SparseTensor` of
shape `[9, 4]` and `indices` / `values`:
[0, 0]: a
[0, 1]: b
[1, 2]: c
[4, 2]: d
[8, 1]: e
Args:
sp_input: The input `SparseTensor`.
shape: A 1-D (vector) int64 `Tensor` specifying the new dense shape of the
represented `SparseTensor`.
name: A name prefix for the returned tensors (optional)
Returns:
A `SparseTensor` with the same non-empty values but with indices calculated
by the new dense shape.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
ValueError: If argument `shape` requests a `SparseTensor` with a different
number of elements than `sp_input`.
ValueError: If `shape` has more than one inferred (== -1) dimension.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
shape = math_ops.cast(shape, dtype=dtypes.int64)
with ops.name_scope(name, "SparseReshape", [sp_input]) as name:
reshaped_ind, reshaped_shape = gen_sparse_ops.sparse_reshape(
sp_input.indices, sp_input.dense_shape, shape, name=name)
reshaped_shape_const = tensor_util.constant_value(shape)
if (reshaped_shape_const is not None and
sp_input.get_shape().is_fully_defined()):
num_implied = sum((dim == -1) for dim in reshaped_shape_const)
if num_implied > 1:
raise ValueError("At most one dimension can be inferred (-1). Found: %s"
% reshaped_shape_const)
original_reshaped_shape = list(reshaped_shape_const) # Copy.
in_shape_size = np.prod(sp_input.get_shape().as_list())
if num_implied:
implied_idx = original_reshaped_shape.index(-1)
non_implied_idx = (
original_reshaped_shape[:implied_idx] +
original_reshaped_shape[implied_idx + 1:])
reshaped_shape_const[implied_idx] = (
in_shape_size // np.prod(non_implied_idx))
reshaped_size = np.prod(reshaped_shape_const)
if reshaped_size != in_shape_size:
raise ValueError("Cannot reshape a tensor with %d elements to shape %s "
"(%d elements)." %
(in_shape_size, original_reshaped_shape,
reshaped_size))
reshaped_shape = reshaped_shape_const
return sparse_tensor.SparseTensor(reshaped_ind,
array_ops.identity(sp_input.values),
reshaped_shape)
# TODO(aselle): Remove keyword required once for 1.0 final
class KeywordRequired(object):
def __repr__(self):
# This is needed to make documentation without fully qualified module paths
return "KeywordRequired()"
@tf_export(v1=["sparse.split", "sparse_split"])
@deprecation.deprecated_endpoints("sparse_split")
@deprecation.deprecated_args(
None, "split_dim is deprecated, use axis instead", "split_dim")
def sparse_split(keyword_required=KeywordRequired(),
sp_input=None,
num_split=None,
axis=None,
name=None,
split_dim=None):
"""Split a `SparseTensor` into `num_split` tensors along `axis`.
If the `sp_input.dense_shape[axis]` is not an integer multiple of `num_split`
each slice starting from 0:`shape[axis] % num_split` gets extra one
dimension. For example, if `axis = 1` and `num_split = 2` and the
input is:
input_tensor = shape = [2, 7]
[ a d e ]
[b c ]
Graphically the output tensors are:
output_tensor[0] =
[ a ]
[b c ]
output_tensor[1] =
[ d e ]
[ ]
Args:
keyword_required: Python 2 standin for * (temporary for argument reorder)
sp_input: The `SparseTensor` to split.
num_split: A Python integer. The number of ways to split.
axis: A 0-D `int32` `Tensor`. The dimension along which to split.
name: A name for the operation (optional).
split_dim: Deprecated old name for axis.
Returns:
`num_split` `SparseTensor` objects resulting from splitting `value`.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
ValueError: If the deprecated `split_dim` and `axis` are both non None.
"""
if not isinstance(keyword_required, KeywordRequired):
raise ValueError("Keyword arguments are required for this function.")
if sp_input is None:
raise ValueError("sp_input is required")
if num_split is None:
raise ValueError("num_split is required")
if axis is None:
raise ValueError("axis is required")
axis = deprecation.deprecated_argument_lookup("axis", axis, "split_dim",
split_dim)
sp_input = _convert_to_sparse_tensor(sp_input)
output_inds, output_vals, output_shapes = (
gen_sparse_ops.sparse_split(
axis,
sp_input.indices,
sp_input.values,
sp_input.dense_shape,
num_split,
name=name))
sparse_tensors = []
for i in range(0, num_split):
sparse_tensors.append(
sparse_tensor.SparseTensor(output_inds[i], output_vals[i],
output_shapes[i]))
return sparse_tensors
@tf_export("sparse.split", v1=[])
def sparse_split_v2(sp_input=None,
num_split=None,
axis=None,
name=None):
"""Split a `SparseTensor` into `num_split` tensors along `axis`.
If the `sp_input.dense_shape[axis]` is not an integer multiple of `num_split`
each slice starting from 0:`shape[axis] % num_split` gets extra one
dimension. For example, if `axis = 1` and `num_split = 2` and the
input is:
input_tensor = shape = [2, 7]
[ a d e ]
[b c ]
Graphically the output tensors are:
output_tensor[0] =
[ a ]
[b c ]
output_tensor[1] =
[ d e ]
[ ]
Args:
sp_input: The `SparseTensor` to split.
num_split: A Python integer. The number of ways to split.
axis: A 0-D `int32` `Tensor`. The dimension along which to split.
name: A name for the operation (optional).
Returns:
`num_split` `SparseTensor` objects resulting from splitting `value`.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
return sparse_split(sp_input=sp_input,
num_split=num_split,
axis=axis,
name=name,
split_dim=None)
@tf_export("sparse.slice", v1=["sparse.slice", "sparse_slice"])
@deprecation.deprecated_endpoints("sparse_slice")
def sparse_slice(sp_input, start, size, name=None):
"""Slice a `SparseTensor` based on the `start` and `size.
For example, if the input is
input_tensor = shape = [2, 7]
[ a d e ]
[b c ]
Graphically the output tensors are:
sparse.slice([0, 0], [2, 4]) = shape = [2, 4]
[ a ]
[b c ]
sparse.slice([0, 4], [2, 3]) = shape = [2, 3]
[ d e ]
[ ]
Args:
sp_input: The `SparseTensor` to split.
start: 1-D. tensor represents the start of the slice.
size: 1-D. tensor represents the size of the slice.
name: A name for the operation (optional).
Returns:
A `SparseTensor` objects resulting from splicing.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
start = ops.convert_to_tensor(start, dtypes.int64)
size = ops.convert_to_tensor(size, dtypes.int64)
with ops.name_scope(name, "SparseSlice", [sp_input]) as name:
output_indices, output_values, output_shape = gen_sparse_ops.sparse_slice(
sp_input.indices,
sp_input.values,
sp_input.dense_shape,
start,
size,
name=name)
return sparse_tensor.SparseTensor(output_indices, output_values,
output_shape)
@tf_export(v1=["sparse_to_dense"])
@deprecation.deprecated(
None,
"Create a `tf.sparse.SparseTensor` and use `tf.sparse.to_dense` instead.")
def sparse_to_dense(sparse_indices,
output_shape,
sparse_values,
default_value=0,
validate_indices=True,
name=None):
"""Converts a sparse representation into a dense tensor.
Builds an array `dense` with shape `output_shape` such that
```python
# If sparse_indices is scalar
dense[i] = (i == sparse_indices ? sparse_values : default_value)
# If sparse_indices is a vector, then for each i
dense[sparse_indices[i]] = sparse_values[i]
# If sparse_indices is an n by d matrix, then for each i in [0, n)
dense[sparse_indices[i][0], ..., sparse_indices[i][d-1]] = sparse_values[i]
```
All other values in `dense` are set to `default_value`. If `sparse_values`
is a scalar, all sparse indices are set to this single value.
Indices should be sorted in lexicographic order, and indices must not
contain any repeats. If `validate_indices` is True, these properties
are checked during execution.
Args:
sparse_indices: A 0-D, 1-D, or 2-D `Tensor` of type `int32` or `int64`.
`sparse_indices[i]` contains the complete index where `sparse_values[i]`
will be placed.
output_shape: A 1-D `Tensor` of the same type as `sparse_indices`. Shape
of the dense output tensor.
sparse_values: A 0-D or 1-D `Tensor`. Values corresponding to each row of
`sparse_indices`, or a scalar value to be used for all sparse indices.
default_value: A 0-D `Tensor` of the same type as `sparse_values`. Value
to set for indices not specified in `sparse_indices`. Defaults to zero.
validate_indices: A boolean value. If True, indices are checked to make
sure they are sorted in lexicographic order and that there are no repeats.
name: A name for the operation (optional).
Returns:
Dense `Tensor` of shape `output_shape`. Has the same type as
`sparse_values`.
"""
return gen_sparse_ops.sparse_to_dense(
sparse_indices,
output_shape,
sparse_values,
default_value=default_value,
validate_indices=validate_indices,
name=name)
@tf_export("sparse.reduce_max", v1=[])
def sparse_reduce_max_v2(
sp_input, axis=None, keepdims=None, output_is_sparse=False, name=None):
"""Computes the max of elements across dimensions of a SparseTensor.
This Op takes a SparseTensor and is the sparse counterpart to
`tf.reduce_max()`. In particular, this Op also returns a dense `Tensor`
if `output_is_sparse` is `False`, or a `SparseTensor` if `output_is_sparse`
is `True`.
Note: A gradient is not defined for this function, so it can't be used
in training models that need gradient descent.
Reduces `sp_input` along the dimensions given in `axis`. Unless
`keepdims` is true, the rank of the tensor is reduced by 1 for each entry in
`axis`. If `keepdims` is true, the reduced dimensions are retained
with length 1.
If `axis` has no entries, all dimensions are reduced, and a tensor
with a single element is returned. Additionally, the axes can be negative,
similar to the indexing rules in Python.
The values not defined in `sp_input` don't participate in the reduce max,
as opposed to be implicitly assumed 0 -- hence it can return negative values
for sparse `axis`. But, in case there are no values in
`axis`, it will reduce to 0. See second example below.
For example:
```python
# 'x' represents [[1, ?, 2]
# [?, 3, ?]]
# where ? is implicitly-zero.
tf.sparse.reduce_max(x) ==> 3
tf.sparse.reduce_max(x, 0) ==> [1, 3, 2]
tf.sparse.reduce_max(x, 1) ==> [2, 3] # Can also use -1 as the axis.
tf.sparse.reduce_max(x, 1, keepdims=True) ==> [[2], [3]]
tf.sparse.reduce_max(x, [0, 1]) ==> 3
# 'y' represents [[-7, ?]
# [ 4, 3]
# [ ?, ?]
tf.sparse.reduce_max(x, 1) ==> [-7, 4, 0]
```
Args:
sp_input: The SparseTensor to reduce. Should have numeric type.
axis: The dimensions to reduce; list or scalar. If `None` (the
default), reduces all dimensions.
keepdims: If true, retain reduced dimensions with length 1.
output_is_sparse: If true, returns a `SparseTensor` instead of a dense
`Tensor` (the default).
name: A name for the operation (optional).
Returns:
The reduced Tensor or the reduced SparseTensor if `output_is_sparse` is
True.
"""
if keepdims is None:
keepdims = False
# reduction_axes is the deprecated name for axis.
reduction_axes = None
if output_is_sparse:
output_ind, output_val, output_shape = (
gen_sparse_ops.sparse_reduce_max_sparse(
sp_input.indices, sp_input.values, sp_input.dense_shape,
math_ops._ReductionDims(sp_input, axis, reduction_axes), keepdims,
name=name))
return sparse_tensor.SparseTensor(output_ind, output_val, output_shape)
return gen_sparse_ops.sparse_reduce_max(
sp_input.indices, sp_input.values, sp_input.dense_shape,
math_ops._ReductionDims(sp_input, axis, reduction_axes), keepdims,
name=name)
@tf_export(v1=["sparse.reduce_max", "sparse_reduce_max"])
@deprecation.deprecated_endpoints("sparse_reduce_max")
@deprecation.deprecated_args(
None, "keep_dims is deprecated, use keepdims instead", "keep_dims")
@deprecation.deprecated_args(
None, "reduction_axes is deprecated, use axis instead",
"reduction_axes")
def sparse_reduce_max(sp_input, axis=None, keepdims=None,
reduction_axes=None, keep_dims=None):
"""Computes the max of elements across dimensions of a SparseTensor.
This Op takes a SparseTensor and is the sparse counterpart to
`tf.reduce_max()`. In particular, this Op also returns a dense `Tensor`
instead of a sparse one.
Note: A gradient is not defined for this function, so it can't be used
in training models that need gradient descent.
Reduces `sp_input` along the dimensions given in `reduction_axes`. Unless
`keepdims` is true, the rank of the tensor is reduced by 1 for each entry in
`reduction_axes`. If `keepdims` is true, the reduced dimensions are retained
with length 1.
If `reduction_axes` has no entries, all dimensions are reduced, and a tensor
with a single element is returned. Additionally, the axes can be negative,
similar to the indexing rules in Python.
The values not defined in `sp_input` don't participate in the reduce max,
as opposed to be implicitly assumed 0 -- hence it can return negative values
for sparse `reduction_axes`. But, in case there are no values in
`reduction_axes`, it will reduce to 0. See second example below.
For example:
```python
# 'x' represents [[1, ?, 2]
# [?, 3, ?]]
# where ? is implicitly-zero.
tf.sparse.reduce_max(x) ==> 3
tf.sparse.reduce_max(x, 0) ==> [1, 3, 2]
tf.sparse.reduce_max(x, 1) ==> [2, 3] # Can also use -1 as the axis.
tf.sparse.reduce_max(x, 1, keepdims=True) ==> [[2], [3]]
tf.sparse.reduce_max(x, [0, 1]) ==> 3
# 'y' represents [[-7, ?]
# [ 4, 3]
# [ ?, ?]
tf.sparse.reduce_max(x, 1) ==> [-7, 4, 0]
```
Args:
sp_input: The SparseTensor to reduce. Should have numeric type.
axis: The dimensions to reduce; list or scalar. If `None` (the
default), reduces all dimensions.
keepdims: If true, retain reduced dimensions with length 1.
reduction_axes: Deprecated name of `axis`.
keep_dims: Deprecated alias for `keepdims`.
Returns:
The reduced Tensor.
"""
keepdims = deprecation.deprecated_argument_lookup("keepdims", keepdims,
"keep_dims", keep_dims)
if keepdims is None:
keepdims = False
return gen_sparse_ops.sparse_reduce_max(
sp_input.indices, sp_input.values, sp_input.dense_shape,
math_ops._ReductionDims(sp_input, axis, reduction_axes), keepdims)
@tf_export(v1=["sparse.reduce_max_sparse", "sparse_reduce_max_sparse"])
@deprecation.deprecated_endpoints("sparse_reduce_max_sparse")
@deprecation.deprecated_args(
None, "keep_dims is deprecated, use keepdims instead", "keep_dims")
def sparse_reduce_max_sparse(sp_input,
axis=None,
keepdims=None,
reduction_axes=None,
keep_dims=None):
"""Computes the max of elements across dimensions of a SparseTensor.
This Op takes a SparseTensor and is the sparse counterpart to
`tf.reduce_max()`. In contrast to SparseReduceSum, this Op returns a
SparseTensor.
Note: A gradient is not defined for this function, so it can't be used
in training models that need gradient descent.
Reduces `sp_input` along the dimensions given in `reduction_axes`. Unless
`keepdims` is true, the rank of the tensor is reduced by 1 for each entry in
`reduction_axes`. If `keepdims` is true, the reduced dimensions are retained
with length 1.
If `reduction_axes` has no entries, all dimensions are reduced, and a tensor
with a single element is returned. Additionally, the axes can be negative,
which are interpreted according to the indexing rules in Python.
Args:
sp_input: The SparseTensor to reduce. Should have numeric type.
axis: The dimensions to reduce; list or scalar. If `None` (the
default), reduces all dimensions.
keepdims: If true, retain reduced dimensions with length 1.
reduction_axes: Deprecated name of axis.
keep_dims: Deprecated alias for `keepdims`.
Returns:
The reduced SparseTensor.
"""
keepdims = deprecation.deprecated_argument_lookup("keepdims", keepdims,
"keep_dims", keep_dims)
if keepdims is None:
keepdims = False
output_ind, output_val, output_shape = (
gen_sparse_ops.sparse_reduce_max_sparse(
sp_input.indices, sp_input.values, sp_input.dense_shape,
math_ops._ReductionDims(sp_input, axis, reduction_axes), keepdims))
return sparse_tensor.SparseTensor(output_ind, output_val, output_shape)
@tf_export("sparse.reduce_sum", v1=[])
def sparse_reduce_sum_v2(
sp_input, axis=None, keepdims=None, output_is_sparse=False, name=None):
"""Computes the sum of elements across dimensions of a SparseTensor.
This Op takes a SparseTensor and is the sparse counterpart to
`tf.reduce_sum()`. In particular, this Op also returns a dense `Tensor`
if `output_is_sparse` is `False`, or a `SparseTensor` if `output_is_sparse`
is `True`.
Note: if `output_is_sparse` is True, a gradient is not defined for this
function, so it can't be used in training models that need gradient descent.
Reduces `sp_input` along the dimensions given in `axis`. Unless `keepdims` is
true, the rank of the tensor is reduced by 1 for each entry in `axis`. If
`keepdims` is true, the reduced dimensions are retained with length 1.
If `axis` has no entries, all dimensions are reduced, and a tensor
with a single element is returned. Additionally, the axes can be negative,
similar to the indexing rules in Python.
For example:
```python
# 'x' represents [[1, ?, 1]
# [?, 1, ?]]
# where ? is implicitly-zero.
tf.sparse.reduce_sum(x) ==> 3
tf.sparse.reduce_sum(x, 0) ==> [1, 1, 1]
tf.sparse.reduce_sum(x, 1) ==> [2, 1] # Can also use -1 as the axis.
tf.sparse.reduce_sum(x, 1, keepdims=True) ==> [[2], [1]]
tf.sparse.reduce_sum(x, [0, 1]) ==> 3
```
Args:
sp_input: The SparseTensor to reduce. Should have numeric type.
axis: The dimensions to reduce; list or scalar. If `None` (the
default), reduces all dimensions.
keepdims: If true, retain reduced dimensions with length 1.
output_is_sparse: If true, returns a `SparseTensor` instead of a dense
`Tensor` (the default).
name: A name for the operation (optional).
Returns:
The reduced Tensor or the reduced SparseTensor if `output_is_sparse` is
True.
"""
if keepdims is None:
keepdims = False
# reduction_axes is the deprecated name for axis.
reduction_axes = None
if output_is_sparse:
output_ind, output_val, output_shape = (
gen_sparse_ops.sparse_reduce_sum_sparse(
sp_input.indices, sp_input.values, sp_input.dense_shape,
math_ops._ReductionDims(sp_input, axis, reduction_axes), keepdims,
name=name))
return sparse_tensor.SparseTensor(output_ind, output_val, output_shape)
return gen_sparse_ops.sparse_reduce_sum(
sp_input.indices, sp_input.values, sp_input.dense_shape,
math_ops._ReductionDims(sp_input, axis, reduction_axes), keepdims,
name=name)
@tf_export(v1=["sparse.reduce_sum", "sparse_reduce_sum"])
@deprecation.deprecated_endpoints("sparse_reduce_sum")
@deprecation.deprecated_args(
None, "keep_dims is deprecated, use keepdims instead", "keep_dims")
@deprecation.deprecated_args(
None, "reduction_axes is deprecated, use axis instead",
"reduction_axes")
def sparse_reduce_sum(sp_input, axis=None, keepdims=None,
reduction_axes=None, keep_dims=None):
"""Computes the sum of elements across dimensions of a SparseTensor.
This Op takes a SparseTensor and is the sparse counterpart to
`tf.reduce_sum()`. In particular, this Op also returns a dense `Tensor`
instead of a sparse one.
Reduces `sp_input` along the dimensions given in `reduction_axes`. Unless
`keepdims` is true, the rank of the tensor is reduced by 1 for each entry in
`reduction_axes`. If `keepdims` is true, the reduced dimensions are retained
with length 1.
If `reduction_axes` has no entries, all dimensions are reduced, and a tensor
with a single element is returned. Additionally, the axes can be negative,
similar to the indexing rules in Python.
For example:
```python
# 'x' represents [[1, ?, 1]
# [?, 1, ?]]
# where ? is implicitly-zero.
tf.sparse.reduce_sum(x) ==> 3
tf.sparse.reduce_sum(x, 0) ==> [1, 1, 1]
tf.sparse.reduce_sum(x, 1) ==> [2, 1] # Can also use -1 as the axis.
tf.sparse.reduce_sum(x, 1, keepdims=True) ==> [[2], [1]]
tf.sparse.reduce_sum(x, [0, 1]) ==> 3
```
Args:
sp_input: The SparseTensor to reduce. Should have numeric type.
axis: The dimensions to reduce; list or scalar. If `None` (the
default), reduces all dimensions.
keepdims: If true, retain reduced dimensions with length 1.
reduction_axes: Deprecated name of `axis`.
keep_dims: Deprecated alias for `keepdims`.
Returns:
The reduced Tensor.
"""
keepdims = deprecation.deprecated_argument_lookup("keepdims", keepdims,
"keep_dims", keep_dims)
if keepdims is None:
keepdims = False
return gen_sparse_ops.sparse_reduce_sum(
sp_input.indices, sp_input.values, sp_input.dense_shape,
math_ops._ReductionDims(sp_input, axis, reduction_axes), keepdims)
@tf_export(v1=["sparse.reduce_sum_sparse", "sparse_reduce_sum_sparse"])
@deprecation.deprecated_endpoints("sparse_reduce_sum_sparse")
@deprecation.deprecated_args(
None, "keep_dims is deprecated, use keepdims instead", "keep_dims")
def sparse_reduce_sum_sparse(sp_input,
axis=None,
keepdims=None,
reduction_axes=None,
keep_dims=None):
"""Computes the sum of elements across dimensions of a SparseTensor.
This Op takes a SparseTensor and is the sparse counterpart to
`tf.reduce_sum()`. In contrast to SparseReduceSum, this Op returns a
SparseTensor.
Note: A gradient is not defined for this function, so it can't be used
in training models that need gradient descent.
Reduces `sp_input` along the dimensions given in `reduction_axes`. Unless
`keepdims` is true, the rank of the tensor is reduced by 1 for each entry in
`reduction_axes`. If `keepdims` is true, the reduced dimensions are retained
with length 1.
If `reduction_axes` has no entries, all dimensions are reduced, and a tensor
with a single element is returned. Additionally, the axes can be negative,
which are interpreted according to the indexing rules in Python.
Args:
sp_input: The SparseTensor to reduce. Should have numeric type.
axis: The dimensions to reduce; list or scalar. If `None` (the
default), reduces all dimensions.
keepdims: If true, retain reduced dimensions with length 1.
reduction_axes: Deprecated name of axis.
keep_dims: Deprecated alias for `keepdims`.
Returns:
The reduced SparseTensor.
"""
keepdims = deprecation.deprecated_argument_lookup("keepdims", keepdims,
"keep_dims", keep_dims)
if keepdims is None:
keepdims = False
output_ind, output_val, output_shape = (
gen_sparse_ops.sparse_reduce_sum_sparse(
sp_input.indices, sp_input.values, sp_input.dense_shape,
math_ops._ReductionDims(sp_input, axis, reduction_axes), keepdims))
return sparse_tensor.SparseTensor(output_ind, output_val, output_shape)
@tf_export("sparse.to_dense", v1=["sparse.to_dense", "sparse_tensor_to_dense"])
@deprecation.deprecated_endpoints("sparse_tensor_to_dense")
def sparse_tensor_to_dense(sp_input,
default_value=0,
validate_indices=True,
name=None):
"""Converts a `SparseTensor` into a dense tensor.
This op is a convenience wrapper around `sparse_to_dense` for `SparseTensor`s.
For example, if `sp_input` has shape `[3, 5]` and non-empty string values:
[0, 1]: a
[0, 3]: b
[2, 0]: c
and `default_value` is `x`, then the output will be a dense `[3, 5]`
string tensor with values:
[[x a x b x]
[x x x x x]
[c x x x x]]
Indices must be without repeats. This is only
tested if `validate_indices` is `True`.
Args:
sp_input: The input `SparseTensor`.
default_value: Scalar value to set for indices not specified in
`sp_input`. Defaults to zero.
validate_indices: A boolean value. If `True`, indices are checked to make
sure they are sorted in lexicographic order and that there are no repeats.
name: A name prefix for the returned tensors (optional).
Returns:
A dense tensor with shape `sp_input.dense_shape` and values specified by
the non-empty values in `sp_input`. Indices not in `sp_input` are assigned
`default_value`.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
return gen_sparse_ops.sparse_to_dense(
sp_input.indices,
sp_input.dense_shape,
sp_input.values,
default_value=default_value,
validate_indices=validate_indices,
name=name)
@tf_export(
"sparse.to_indicator", v1=["sparse.to_indicator", "sparse_to_indicator"])
@deprecation.deprecated_endpoints("sparse_to_indicator")
def sparse_to_indicator(sp_input, vocab_size, name=None):
"""Converts a `SparseTensor` of ids into a dense bool indicator tensor.
The last dimension of `sp_input.indices` is discarded and replaced with
the values of `sp_input`. If `sp_input.dense_shape = [D0, D1, ..., Dn, K]`,
then `output.shape = [D0, D1, ..., Dn, vocab_size]`, where
output[d_0, d_1, ..., d_n, sp_input[d_0, d_1, ..., d_n, k]] = True
and False elsewhere in `output`.
For example, if `sp_input.dense_shape = [2, 3, 4]` with non-empty values:
[0, 0, 0]: 0
[0, 1, 0]: 10
[1, 0, 3]: 103
[1, 1, 2]: 150
[1, 1, 3]: 149
[1, 1, 4]: 150
[1, 2, 1]: 121
and `vocab_size = 200`, then the output will be a `[2, 3, 200]` dense bool
tensor with False everywhere except at positions
(0, 0, 0), (0, 1, 10), (1, 0, 103), (1, 1, 149), (1, 1, 150),
(1, 2, 121).
Note that repeats are allowed in the input SparseTensor.
This op is useful for converting `SparseTensor`s into dense formats for
compatibility with ops that expect dense tensors.
The input `SparseTensor` must be in row-major order.
Args:
sp_input: A `SparseTensor` with `values` property of type `int32` or
`int64`.
vocab_size: A scalar int64 Tensor (or Python int) containing the new size
of the last dimension, `all(0 <= sp_input.values < vocab_size)`.
name: A name prefix for the returned tensors (optional)
Returns:
A dense bool indicator tensor representing the indices with specified value.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
with ops.name_scope(name, "SparseToIndicator", [sp_input]) as name:
num_entries = array_ops.shape(sp_input.indices)[0]
new_values = array_ops.fill(array_ops.expand_dims(num_entries, 0), True)
sp_values = sparse_tensor.SparseTensor(sp_input.indices, new_values,
sp_input.dense_shape)
sp_new = sparse_merge_impl(sp_input, sp_values, vocab_size, name)
# validate_indices may be False because we allow duplicates in new_indices:
# repeated indices are allowed when creating an indicator matrix.
return sparse_tensor_to_dense(
sp_new, default_value=False, validate_indices=False, name=name)
@tf_export(v1=["sparse.merge", "sparse_merge"])
@deprecation.deprecated(None, "No similar op available at this time.")
def sparse_merge(sp_ids, sp_values, vocab_size, name=None,
already_sorted=False):
"""Combines a batch of feature ids and values into a single `SparseTensor`.
The most common use case for this function occurs when feature ids and
their corresponding values are stored in `Example` protos on disk.
`parse_example` will return a batch of ids and a batch of values, and this
function joins them into a single logical `SparseTensor` for use in
functions such as `sparse_tensor_dense_matmul`, `sparse_to_dense`, etc.
The `SparseTensor` returned by this function has the following properties:
- `indices` is equivalent to `sp_ids.indices` with the last
dimension discarded and replaced with `sp_ids.values`.
- `values` is simply `sp_values.values`.
- If `sp_ids.dense_shape = [D0, D1, ..., Dn, K]`, then
`output.shape = [D0, D1, ..., Dn, vocab_size]`.
For example, consider the following feature vectors:
```python
vector1 = [-3, 0, 0, 0, 0, 0]
vector2 = [ 0, 1, 0, 4, 1, 0]
vector3 = [ 5, 0, 0, 9, 0, 0]
```
These might be stored sparsely in the following Example protos by storing
only the feature ids (column number if the vectors are treated as a matrix)
of the non-zero elements and the corresponding values:
```python
examples = [Example(features={
"ids": Feature(int64_list=Int64List(value=[0])),
"values": Feature(float_list=FloatList(value=[-3]))}),
Example(features={
"ids": Feature(int64_list=Int64List(value=[1, 4, 3])),
"values": Feature(float_list=FloatList(value=[1, 1, 4]))}),
Example(features={
"ids": Feature(int64_list=Int64List(value=[0, 3])),
"values": Feature(float_list=FloatList(value=[5, 9]))})]
```
The result of calling parse_example on these examples will produce a
dictionary with entries for "ids" and "values". Passing those two objects
to this function along with vocab_size=6, will produce a `SparseTensor` that
sparsely represents all three instances. Namely, the `indices` property will
contain the coordinates of the non-zero entries in the feature matrix (the
first dimension is the row number in the matrix, i.e., the index within the
batch, and the second dimension is the column number, i.e., the feature id);
`values` will contain the actual values. `shape` will be the shape of the
original matrix, i.e., (3, 6). For our example above, the output will be
equal to:
```python
SparseTensor(indices=[[0, 0], [1, 1], [1, 3], [1, 4], [2, 0], [2, 3]],
values=[-3, 1, 4, 1, 5, 9],
dense_shape=[3, 6])
```
This method generalizes to higher-dimensions by simply providing a list for
both the sp_ids as well as the vocab_size.
In this case the resulting `SparseTensor` has the following properties:
- `indices` is equivalent to `sp_ids[0].indices` with the last
dimension discarded and concatenated with
`sp_ids[0].values, sp_ids[1].values, ...`.
- `values` is simply `sp_values.values`.
- If `sp_ids.dense_shape = [D0, D1, ..., Dn, K]`, then
`output.shape = [D0, D1, ..., Dn] + vocab_size`.
Args:
sp_ids: A single `SparseTensor` with `values` property of type `int32`
or `int64` or a Python list of such `SparseTensor`s or a list thereof.
sp_values: A `SparseTensor` of any type.
vocab_size: A scalar `int64` Tensor (or Python int) containing the new size
of the last dimension, `all(0 <= sp_ids.values < vocab_size)`.
Or a list thereof with `all(0 <= sp_ids[i].values < vocab_size[i])` for
all `i`.
name: A name prefix for the returned tensors (optional)
already_sorted: A boolean to specify whether the per-batch values in
`sp_values` are already sorted. If so skip sorting, False by default
(optional).
Returns:
A `SparseTensor` compactly representing a batch of feature ids and values,
useful for passing to functions that expect such a `SparseTensor`.
Raises:
TypeError: If `sp_values` is not a `SparseTensor`. Or if `sp_ids` is neither
a `SparseTensor` nor a list thereof. Or if `vocab_size` is not a
`Tensor` or a Python int and `sp_ids` is a `SparseTensor`. Or if
`vocab_size` is not a or list thereof and `sp_ids` is a list.
ValueError: If `sp_ids` and `vocab_size` are lists of different lengths.
"""
return sparse_merge_impl(sp_ids, sp_values, vocab_size, name, already_sorted)
def sparse_merge_impl(sp_ids,
sp_values,
vocab_size,
name=None,
already_sorted=False):
"""Internal implementation for sparse_merge to avoid deprecation warnings."""
if isinstance(sp_ids, sparse_tensor.SparseTensorValue) or isinstance(
sp_ids, sparse_tensor.SparseTensor):
sp_ids = [sp_ids]
if not (isinstance(vocab_size, ops.Tensor) or
isinstance(vocab_size, numbers.Integral)):
raise TypeError("vocab_size has to be a Tensor or Python int. Found %s" %
type(vocab_size))
vocab_size = [vocab_size]
else:
if not isinstance(sp_ids, collections.Iterable):
raise TypeError("sp_ids has to be a SparseTensor or list thereof. "
"Found %s" % type(sp_ids))
if not isinstance(vocab_size, collections.Iterable):
raise TypeError("vocab_size has to be a list of Tensors or Python ints. "
"Found %s" % type(vocab_size))
for dim in vocab_size:
if not (isinstance(dim, ops.Tensor) or isinstance(dim, numbers.Integral)):
raise TypeError(
"vocab_size has to be a list of Tensors or Python ints. Found %s" %
type(dim))
if len(sp_ids) != len(vocab_size):
raise ValueError("sp_ids and vocab_size have to have equal lengths.")
with ops.name_scope(name, "SparseMerge", [sp_ids, sp_values]):
sp_ids = [_convert_to_sparse_tensor(sp_ids_dim) for sp_ids_dim in sp_ids]
sp_values = _convert_to_sparse_tensor(sp_values)
ids = []
for sp_ids_dim in sp_ids:
ids_dim = sp_ids_dim.values
if sp_ids_dim.dtype != dtypes.int64:
ids_dim = math_ops.cast(ids_dim, dtypes.int64)
ids += [array_ops.expand_dims(ids_dim, axis=1)]
vocab_size = [math_ops.cast(x, dtypes.int64) for x in vocab_size]
# Slice off the last dimension of indices, then tack on the ids
indices_columns_to_preserve = sp_ids[0].indices[:, :-1]
new_indices = array_ops.concat([indices_columns_to_preserve] + ids, 1)
new_values = sp_values.values
new_shape = array_ops.concat([sp_ids[0].dense_shape[:-1], vocab_size], 0)
result = sparse_tensor.SparseTensor(new_indices, new_values, new_shape)
if already_sorted:
return result
sorted_result = sparse_reorder(result)
return sparse_tensor.SparseTensor(
sorted_result.indices, sorted_result.values, new_shape)
@tf_export("sparse.retain", v1=["sparse.retain", "sparse_retain"])
@deprecation.deprecated_endpoints("sparse_retain")
def sparse_retain(sp_input, to_retain):
"""Retains specified non-empty values within a `SparseTensor`.
For example, if `sp_input` has shape `[4, 5]` and 4 non-empty string values:
[0, 1]: a
[0, 3]: b
[2, 0]: c
[3, 1]: d
and `to_retain = [True, False, False, True]`, then the output will
be a `SparseTensor` of shape `[4, 5]` with 2 non-empty values:
[0, 1]: a
[3, 1]: d
Args:
sp_input: The input `SparseTensor` with `N` non-empty elements.
to_retain: A bool vector of length `N` with `M` true values.
Returns:
A `SparseTensor` with the same shape as the input and `M` non-empty
elements corresponding to the true positions in `to_retain`.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
to_retain = ops.convert_to_tensor(to_retain)
# Shape checking, if shape is known at graph construction time
retain_shape = to_retain.get_shape()
retain_shape.assert_has_rank(1)
if sp_input.values.get_shape().dims is not None:
sp_input.values.get_shape().dims[0].merge_with(
tensor_shape.dimension_at_index(retain_shape, 0))
where_true = array_ops.reshape(array_ops.where(to_retain), [-1])
new_indices = array_ops.gather(sp_input.indices, where_true)
new_values = array_ops.gather(sp_input.values, where_true)
return sparse_tensor.SparseTensor(new_indices, new_values,
array_ops.identity(sp_input.dense_shape))
@tf_export(
"sparse.reset_shape", v1=["sparse.reset_shape", "sparse_reset_shape"])
@deprecation.deprecated_endpoints("sparse_reset_shape")
def sparse_reset_shape(sp_input, new_shape=None):
"""Resets the shape of a `SparseTensor` with indices and values unchanged.
If `new_shape` is None, returns a copy of `sp_input` with its shape reset
to the tight bounding box of `sp_input`. This will be a shape consisting of
all zeros if sp_input has no values.
If `new_shape` is provided, then it must be larger or equal in all dimensions
compared to the shape of `sp_input`. When this condition is met, the returned
SparseTensor will have its shape reset to `new_shape` and its indices and
values unchanged from that of `sp_input.`
For example:
Consider a `sp_input` with shape [2, 3, 5]:
[0, 0, 1]: a
[0, 1, 0]: b
[0, 2, 2]: c
[1, 0, 3]: d
- It is an error to set `new_shape` as [3, 7] since this represents a
rank-2 tensor while `sp_input` is rank-3. This is either a ValueError
during graph construction (if both shapes are known) or an OpError during
run time.
- Setting `new_shape` as [2, 3, 6] will be fine as this shape is larger or
equal in every dimension compared to the original shape [2, 3, 5].
- On the other hand, setting new_shape as [2, 3, 4] is also an error: The
third dimension is smaller than the original shape [2, 3, 5] (and an
`InvalidArgumentError` will be raised).
- If `new_shape` is None, the returned SparseTensor will have a shape
[2, 3, 4], which is the tight bounding box of `sp_input`.
Args:
sp_input: The input `SparseTensor`.
new_shape: None or a vector representing the new shape for the returned
`SparseTensor`.
Returns:
A `SparseTensor` indices and values unchanged from `input_sp`. Its shape is
`new_shape` if that is set. Otherwise it is the tight bounding box of
`input_sp`
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
ValueError: If `new_shape` represents a tensor with a different rank from
that of `sp_input` (if shapes are known when graph is constructed).
ValueError: If `new_shape` is determined during graph build to have
dimension sizes that are too small.
OpError:
- If `new_shape` has dimension sizes that are too small.
- If shapes are not known during graph construction time, and during run
time it is found out that the ranks do not match.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
in_indices = array_ops.identity(sp_input.indices)
in_values = array_ops.identity(sp_input.values)
in_shape = array_ops.identity(sp_input.dense_shape)
if new_shape is None:
dim_low_bound = math_ops.reduce_max(in_indices, axis=0)
output_shape_tensor = math_ops.maximum(
array_ops.constant(0, dtype=dtypes.int64),
math_ops.add(dim_low_bound, array_ops.ones_like(in_shape)))
else:
output_shape_tensor = ops.convert_to_tensor(new_shape)
output_shape_tensor.get_shape().assert_has_rank(1)
output_shape_tensor = math_ops.cast(output_shape_tensor, dtypes.int64)
# For cases when shape is known during graph construction, this catches the
# error before the sparse_tensor.SparseTensor catches it.
if output_shape_tensor.get_shape().rank is not None:
output_shape_tensor.get_shape().dims[0].merge_with(
in_shape.get_shape().dims[0])
output_shape_tensor_const = tensor_util.constant_value(output_shape_tensor)
# For cases where all shapes are known during graph construction
if (output_shape_tensor_const is not None and
sp_input.get_shape().is_fully_defined()):
in_shape_const = np.array(sp_input.get_shape().as_list())
if not np.all(in_shape_const <= output_shape_tensor_const):
raise ValueError(
"Requested new_shape should have dimension sizes >= sp_input.shape."
" Found new_shape (%s), sp_input.shape (%s)." %
(in_shape_const, output_shape_tensor_const))
output_shape_tensor = output_shape_tensor_const
else:
# For cases where shape is not known during graph construction.
output_shape_tensor = control_flow_ops.with_dependencies([
check_ops.assert_equal(
array_ops.shape(in_shape), array_ops.shape(output_shape_tensor))
], output_shape_tensor)
output_shape_tensor = control_flow_ops.with_dependencies(
[check_ops.assert_less_equal(in_shape, output_shape_tensor)],
output_shape_tensor)
return sparse_tensor.SparseTensor(in_indices, in_values, output_shape_tensor)
@tf_export(
"sparse.fill_empty_rows",
v1=["sparse.fill_empty_rows", "sparse_fill_empty_rows"])
@deprecation.deprecated_endpoints("sparse_fill_empty_rows")
def sparse_fill_empty_rows(sp_input, default_value, name=None):
"""Fills empty rows in the input 2-D `SparseTensor` with a default value.
This op adds entries with the specified `default_value` at index
`[row, 0]` for any row in the input that does not already have a value.
For example, suppose `sp_input` has shape `[5, 6]` and non-empty values:
[0, 1]: a
[0, 3]: b
[2, 0]: c
[3, 1]: d
Rows 1 and 4 are empty, so the output will be of shape `[5, 6]` with values:
[0, 1]: a
[0, 3]: b
[1, 0]: default_value
[2, 0]: c
[3, 1]: d
[4, 0]: default_value
Note that the input may have empty columns at the end, with no effect on
this op.
The output `SparseTensor` will be in row-major order and will have the
same shape as the input.
This op also returns an indicator vector such that
empty_row_indicator[i] = True iff row i was an empty row.
Args:
sp_input: A `SparseTensor` with shape `[N, M]`.
default_value: The value to fill for empty rows, with the same type as
`sp_input.`
name: A name prefix for the returned tensors (optional)
Returns:
sp_ordered_output: A `SparseTensor` with shape `[N, M]`, and with all empty
rows filled in with `default_value`.
empty_row_indicator: A bool vector of length `N` indicating whether each
input row was empty.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
with ops.name_scope(name, "SparseFillEmptyRows", [sp_input]):
default_value = ops.convert_to_tensor(
default_value, dtype=sp_input.values.dtype)
(output_indices, output_values, empty_row_indicator,
unused_reverse_index_map) = gen_sparse_ops.sparse_fill_empty_rows(
indices=sp_input.indices,
values=sp_input.values,
dense_shape=sp_input.dense_shape,
default_value=default_value)
return (sparse_tensor.SparseTensor(
indices=output_indices,
values=output_values,
dense_shape=sp_input.dense_shape), empty_row_indicator)
@tf_export(v1=["io.serialize_sparse", "serialize_sparse"])
@deprecation.deprecated_endpoints("serialize_sparse")
def serialize_sparse(sp_input, name=None, out_type=dtypes.string):
"""Serialize a `SparseTensor` into a 3-vector (1-D `Tensor`) object.
Args:
sp_input: The input `SparseTensor`.
name: A name prefix for the returned tensors (optional).
out_type: The `dtype` to use for serialization.
Returns:
A 3-vector (1-D `Tensor`), with each column representing the serialized
`SparseTensor`'s indices, values, and shape (respectively).
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
return serialize_sparse_v2(sp_input, out_type, name)
@tf_export("io.serialize_sparse", v1=[])
def serialize_sparse_v2(sp_input, out_type=dtypes.string, name=None):
"""Serialize a `SparseTensor` into a 3-vector (1-D `Tensor`) object.
Args:
sp_input: The input `SparseTensor`.
out_type: The `dtype` to use for serialization.
name: A name prefix for the returned tensors (optional).
Returns:
A 3-vector (1-D `Tensor`), with each column representing the serialized
`SparseTensor`'s indices, values, and shape (respectively).
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
return gen_sparse_ops.serialize_sparse(
sp_input.indices,
sp_input.values,
sp_input.dense_shape,
name=name,
out_type=out_type)
@tf_export(v1=["io.serialize_many_sparse", "serialize_many_sparse"])
@deprecation.deprecated_endpoints("serialize_many_sparse")
def serialize_many_sparse(sp_input, name=None, out_type=dtypes.string):
"""Serialize `N`-minibatch `SparseTensor` into an `[N, 3]` `Tensor`.
The `SparseTensor` must have rank `R` greater than 1, and the first dimension
is treated as the minibatch dimension. Elements of the `SparseTensor`
must be sorted in increasing order of this first dimension. The serialized
`SparseTensor` objects going into each row of the output `Tensor` will have
rank `R-1`.
The minibatch size `N` is extracted from `sparse_shape[0]`.
Args:
sp_input: The input rank `R` `SparseTensor`.
name: A name prefix for the returned tensors (optional).
out_type: The `dtype` to use for serialization.
Returns:
A matrix (2-D `Tensor`) with `N` rows and `3` columns. Each column
represents serialized `SparseTensor`'s indices, values, and shape
(respectively).
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
return serialize_many_sparse_v2(sp_input, out_type, name)
@tf_export("io.serialize_many_sparse", v1=[])
def serialize_many_sparse_v2(sp_input, out_type=dtypes.string, name=None):
"""Serialize `N`-minibatch `SparseTensor` into an `[N, 3]` `Tensor`.
The `SparseTensor` must have rank `R` greater than 1, and the first dimension
is treated as the minibatch dimension. Elements of the `SparseTensor`
must be sorted in increasing order of this first dimension. The serialized
`SparseTensor` objects going into each row of the output `Tensor` will have
rank `R-1`.
The minibatch size `N` is extracted from `sparse_shape[0]`.
Args:
sp_input: The input rank `R` `SparseTensor`.
out_type: The `dtype` to use for serialization.
name: A name prefix for the returned tensors (optional).
Returns:
A matrix (2-D `Tensor`) with `N` rows and `3` columns. Each column
represents serialized `SparseTensor`'s indices, values, and shape
(respectively).
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
return gen_sparse_ops.serialize_many_sparse(
sp_input.indices,
sp_input.values,
sp_input.dense_shape,
name=name,
out_type=out_type)
def deserialize_sparse(serialized_sparse, dtype, rank=None, name=None):
"""Deserialize `SparseTensor` objects.
The input `serialized_sparse` must have the shape `[?, ?, ..., ?, 3]` where
the last dimension stores serialized `SparseTensor` objects and the other N
dimensions (N >= 0) correspond to a batch. The ranks of the original
`SparseTensor` objects must all match. When the final `SparseTensor` is
created, its rank is the rank of the incoming `SparseTensor` objects plus N;
the sparse tensors have been concatenated along new dimensions, one for each
batch.
The output `SparseTensor` object's shape values for the original dimensions
are the max across the input `SparseTensor` objects' shape values for the
corresponding dimensions. The new dimensions match the size of the batch.
The input `SparseTensor` objects' indices are assumed ordered in
standard lexicographic order. If this is not the case, after this
step run `SparseReorder` to restore index ordering.
For example, if the serialized input is a `[2 x 3]` matrix representing two
original `SparseTensor` objects:
index = [ 0]
[10]
[20]
values = [1, 2, 3]
shape = [50]
and
index = [ 2]
[10]
values = [4, 5]
shape = [30]
then the final deserialized `SparseTensor` will be:
index = [0 0]
[0 10]
[0 20]
[1 2]
[1 10]
values = [1, 2, 3, 4, 5]
shape = [2 50]
Args:
serialized_sparse: The serialized `SparseTensor` objects.
The last dimension must have 3 columns.
dtype: The `dtype` of the serialized `SparseTensor` objects.
rank: (optional) Python int, the rank of the `SparseTensor` objects.
name: A name prefix for the returned tensors (optional).
Returns:
A `SparseTensor` representing the deserialized `SparseTensor` objects.
"""
output_indices, output_values, output_shape = (
gen_sparse_ops.deserialize_sparse(serialized_sparse, dtype, name=name))
# Feed rank data back in, if available
output_indices.set_shape([None, rank])
output_shape.set_shape([rank])
return sparse_tensor.SparseTensor(output_indices, output_values, output_shape)
@tf_export(
"io.deserialize_many_sparse",
v1=["io.deserialize_many_sparse", "deserialize_many_sparse"])
@deprecation.deprecated_endpoints("deserialize_many_sparse")
def deserialize_many_sparse(serialized_sparse, dtype, rank=None, name=None):
"""Deserialize and concatenate `SparseTensors` from a serialized minibatch.
The input `serialized_sparse` must be a string matrix of shape `[N x 3]` where
`N` is the minibatch size and the rows correspond to packed outputs of
`serialize_sparse`. The ranks of the original `SparseTensor` objects
must all match. When the final `SparseTensor` is created, it has rank one
higher than the ranks of the incoming `SparseTensor` objects (they have been
concatenated along a new row dimension).
The output `SparseTensor` object's shape values for all dimensions but the
first are the max across the input `SparseTensor` objects' shape values
for the corresponding dimensions. Its first shape value is `N`, the minibatch
size.
The input `SparseTensor` objects' indices are assumed ordered in
standard lexicographic order. If this is not the case, after this
step run `sparse.reorder` to restore index ordering.
For example, if the serialized input is a `[2, 3]` matrix representing two
original `SparseTensor` objects:
index = [ 0]
[10]
[20]
values = [1, 2, 3]
shape = [50]
and
index = [ 2]
[10]
values = [4, 5]
shape = [30]
then the final deserialized `SparseTensor` will be:
index = [0 0]
[0 10]
[0 20]
[1 2]
[1 10]
values = [1, 2, 3, 4, 5]
shape = [2 50]
Args:
serialized_sparse: 2-D `Tensor` of type `string` of shape `[N, 3]`.
The serialized and packed `SparseTensor` objects.
dtype: The `dtype` of the serialized `SparseTensor` objects.
rank: (optional) Python int, the rank of the `SparseTensor` objects.
name: A name prefix for the returned tensors (optional)
Returns:
A `SparseTensor` representing the deserialized `SparseTensor`s,
concatenated along the `SparseTensor`s' first dimension.
All of the serialized `SparseTensor`s must have had the same rank and type.
"""
output_indices, output_values, output_shape = (
gen_sparse_ops.deserialize_many_sparse(
serialized_sparse, dtype, name=name))
# Feed rank data back in, if available
output_indices.set_shape([None, rank])
output_shape.set_shape([rank])
return sparse_tensor.SparseTensor(output_indices, output_values, output_shape)
@tf_export("sparse.sparse_dense_matmul",
v1=["sparse.sparse_dense_matmul", "sparse.matmul",
"sparse_tensor_dense_matmul"])
@deprecation.deprecated_endpoints("sparse_tensor_dense_matmul")
def sparse_tensor_dense_matmul(sp_a,
b,
adjoint_a=False,
adjoint_b=False,
name=None):
# pylint: disable=line-too-long
"""Multiply SparseTensor (of rank 2) "A" by dense matrix "B".
No validity checking is performed on the indices of `A`. However, the
following input format is recommended for optimal behavior:
* If `adjoint_a == false`: `A` should be sorted in lexicographically
increasing order. Use `sparse.reorder` if you're not sure.
* If `adjoint_a == true`: `A` should be sorted in order of increasing
dimension 1 (i.e., "column major" order instead of "row major" order).
Using `tf.nn.embedding_lookup_sparse` for sparse multiplication:
It's not obvious but you can consider `embedding_lookup_sparse` as another
sparse and dense multiplication. In some situations, you may prefer to use
`embedding_lookup_sparse` even though you're not dealing with embeddings.
There are two questions to ask in the decision process: Do you need gradients
computed as sparse too? Is your sparse data represented as two
`SparseTensor`s: ids and values? There is more explanation about data format
below. If you answer any of these questions as yes, consider using
`tf.nn.embedding_lookup_sparse`.
Following explains differences between the expected SparseTensors:
For example if dense form of your sparse data has shape `[3, 5]` and values:
[[ a ]
[b c]
[ d ]]
`SparseTensor` format expected by `sparse_tensor_dense_matmul`:
`sp_a` (indices, values):
[0, 1]: a
[1, 0]: b
[1, 4]: c
[2, 2]: d
`SparseTensor` format expected by `embedding_lookup_sparse`:
`sp_ids` `sp_weights`
[0, 0]: 1 [0, 0]: a
[1, 0]: 0 [1, 0]: b
[1, 1]: 4 [1, 1]: c
[2, 0]: 2 [2, 0]: d
Deciding when to use `sparse_tensor_dense_matmul` vs.
`matmul`(a_is_sparse=True):
There are a number of questions to ask in the decision process, including:
* Will the SparseTensor `A` fit in memory if densified?
* Is the column count of the product large (>> 1)?
* Is the density of `A` larger than approximately 15%?
If the answer to several of these questions is yes, consider
converting the `SparseTensor` to a dense one and using `tf.matmul` with
`a_is_sparse=True`.
This operation tends to perform well when `A` is more sparse, if the column
size of the product is small (e.g. matrix-vector multiplication), if
`sp_a.dense_shape` takes on large values.
Below is a rough speed comparison between `sparse_tensor_dense_matmul`,
labeled 'sparse', and `matmul`(a_is_sparse=True), labeled 'dense'. For
purposes of the comparison, the time spent converting from a `SparseTensor` to
a dense `Tensor` is not included, so it is overly conservative with respect to
the time ratio.
Benchmark system:
CPU: Intel Ivybridge with HyperThreading (6 cores) dL1:32KB dL2:256KB dL3:12MB
GPU: NVidia Tesla k40c
Compiled with:
`-c opt --config=cuda --copt=-mavx`
```
tensorflow/python/sparse_tensor_dense_matmul_op_test --benchmarks
A sparse [m, k] with % nonzero values between 1% and 80%
B dense [k, n]
% nnz n gpu m k dt(dense) dt(sparse) dt(sparse)/dt(dense)
0.01 1 True 100 100 0.000221166 0.00010154 0.459112
0.01 1 True 100 1000 0.00033858 0.000109275 0.322745
0.01 1 True 1000 100 0.000310557 9.85661e-05 0.317385
0.01 1 True 1000 1000 0.0008721 0.000100875 0.115669
0.01 1 False 100 100 0.000208085 0.000107603 0.51711
0.01 1 False 100 1000 0.000327112 9.51118e-05 0.290762
0.01 1 False 1000 100 0.000308222 0.00010345 0.335635
0.01 1 False 1000 1000 0.000865721 0.000101397 0.117124
0.01 10 True 100 100 0.000218522 0.000105537 0.482958
0.01 10 True 100 1000 0.000340882 0.000111641 0.327506
0.01 10 True 1000 100 0.000315472 0.000117376 0.372064
0.01 10 True 1000 1000 0.000905493 0.000123263 0.136128
0.01 10 False 100 100 0.000221529 9.82571e-05 0.44354
0.01 10 False 100 1000 0.000330552 0.000112615 0.340687
0.01 10 False 1000 100 0.000341277 0.000114097 0.334324
0.01 10 False 1000 1000 0.000819944 0.000120982 0.147549
0.01 25 True 100 100 0.000207806 0.000105977 0.509981
0.01 25 True 100 1000 0.000322879 0.00012921 0.400181
0.01 25 True 1000 100 0.00038262 0.00014158 0.370035
0.01 25 True 1000 1000 0.000865438 0.000202083 0.233504
0.01 25 False 100 100 0.000209401 0.000104696 0.499979
0.01 25 False 100 1000 0.000321161 0.000130737 0.407076
0.01 25 False 1000 100 0.000377012 0.000136801 0.362856
0.01 25 False 1000 1000 0.000861125 0.00020272 0.235413
0.2 1 True 100 100 0.000206952 9.69219e-05 0.46833
0.2 1 True 100 1000 0.000348674 0.000147475 0.422959
0.2 1 True 1000 100 0.000336908 0.00010122 0.300439
0.2 1 True 1000 1000 0.001022 0.000203274 0.198898
0.2 1 False 100 100 0.000207532 9.5412e-05 0.459746
0.2 1 False 100 1000 0.000356127 0.000146824 0.41228
0.2 1 False 1000 100 0.000322664 0.000100918 0.312764
0.2 1 False 1000 1000 0.000998987 0.000203442 0.203648
0.2 10 True 100 100 0.000211692 0.000109903 0.519165
0.2 10 True 100 1000 0.000372819 0.000164321 0.440753
0.2 10 True 1000 100 0.000338651 0.000144806 0.427596
0.2 10 True 1000 1000 0.00108312 0.000758876 0.70064
0.2 10 False 100 100 0.000215727 0.000110502 0.512231
0.2 10 False 100 1000 0.000375419 0.0001613 0.429653
0.2 10 False 1000 100 0.000336999 0.000145628 0.432132
0.2 10 False 1000 1000 0.00110502 0.000762043 0.689618
0.2 25 True 100 100 0.000218705 0.000129913 0.594009
0.2 25 True 100 1000 0.000394794 0.00029428 0.745402
0.2 25 True 1000 100 0.000404483 0.0002693 0.665788
0.2 25 True 1000 1000 0.0012002 0.00194494 1.62052
0.2 25 False 100 100 0.000221494 0.0001306 0.589632
0.2 25 False 100 1000 0.000396436 0.000297204 0.74969
0.2 25 False 1000 100 0.000409346 0.000270068 0.659754
0.2 25 False 1000 1000 0.00121051 0.00193737 1.60046
0.5 1 True 100 100 0.000214981 9.82111e-05 0.456836
0.5 1 True 100 1000 0.000415328 0.000223073 0.537101
0.5 1 True 1000 100 0.000358324 0.00011269 0.314492
0.5 1 True 1000 1000 0.00137612 0.000437401 0.317851
0.5 1 False 100 100 0.000224196 0.000101423 0.452386
0.5 1 False 100 1000 0.000400987 0.000223286 0.556841
0.5 1 False 1000 100 0.000368825 0.00011224 0.304318
0.5 1 False 1000 1000 0.00136036 0.000429369 0.31563
0.5 10 True 100 100 0.000222125 0.000112308 0.505608
0.5 10 True 100 1000 0.000461088 0.00032357 0.701753
0.5 10 True 1000 100 0.000394624 0.000225497 0.571422
0.5 10 True 1000 1000 0.00158027 0.00190898 1.20801
0.5 10 False 100 100 0.000232083 0.000114978 0.495418
0.5 10 False 100 1000 0.000454574 0.000324632 0.714146
0.5 10 False 1000 100 0.000379097 0.000227768 0.600817
0.5 10 False 1000 1000 0.00160292 0.00190168 1.18638
0.5 25 True 100 100 0.00023429 0.000151703 0.647501
0.5 25 True 100 1000 0.000497462 0.000598873 1.20386
0.5 25 True 1000 100 0.000460778 0.000557038 1.20891
0.5 25 True 1000 1000 0.00170036 0.00467336 2.74845
0.5 25 False 100 100 0.000228981 0.000155334 0.678371
0.5 25 False 100 1000 0.000496139 0.000620789 1.25124
0.5 25 False 1000 100 0.00045473 0.000551528 1.21287
0.5 25 False 1000 1000 0.00171793 0.00467152 2.71927
0.8 1 True 100 100 0.000222037 0.000105301 0.47425
0.8 1 True 100 1000 0.000410804 0.000329327 0.801664
0.8 1 True 1000 100 0.000349735 0.000131225 0.375212
0.8 1 True 1000 1000 0.00139219 0.000677065 0.48633
0.8 1 False 100 100 0.000214079 0.000107486 0.502085
0.8 1 False 100 1000 0.000413746 0.000323244 0.781261
0.8 1 False 1000 100 0.000348983 0.000131983 0.378193
0.8 1 False 1000 1000 0.00136296 0.000685325 0.50282
0.8 10 True 100 100 0.000229159 0.00011825 0.516017
0.8 10 True 100 1000 0.000498845 0.000532618 1.0677
0.8 10 True 1000 100 0.000383126 0.00029935 0.781336
0.8 10 True 1000 1000 0.00162866 0.00307312 1.88689
0.8 10 False 100 100 0.000230783 0.000124958 0.541452
0.8 10 False 100 1000 0.000493393 0.000550654 1.11606
0.8 10 False 1000 100 0.000377167 0.000298581 0.791642
0.8 10 False 1000 1000 0.00165795 0.00305103 1.84024
0.8 25 True 100 100 0.000233496 0.000175241 0.75051
0.8 25 True 100 1000 0.00055654 0.00102658 1.84458
0.8 25 True 1000 100 0.000463814 0.000783267 1.68875
0.8 25 True 1000 1000 0.00186905 0.00755344 4.04132
0.8 25 False 100 100 0.000240243 0.000175047 0.728625
0.8 25 False 100 1000 0.000578102 0.00104499 1.80763
0.8 25 False 1000 100 0.000485113 0.000776849 1.60138
0.8 25 False 1000 1000 0.00211448 0.00752736 3.55992
```
Args:
sp_a: SparseTensor A, of rank 2.
b: A dense Matrix with the same dtype as sp_a.
adjoint_a: Use the adjoint of A in the matrix multiply. If A is complex,
this is transpose(conj(A)). Otherwise it's transpose(A).
adjoint_b: Use the adjoint of B in the matrix multiply. If B is complex,
this is transpose(conj(B)). Otherwise it's transpose(B).
name: A name prefix for the returned tensors (optional)
Returns:
A dense matrix (pseudo-code in dense np.matrix notation):
`A = A.H if adjoint_a else A`
`B = B.H if adjoint_b else B`
`return A*B`
"""
# pylint: enable=line-too-long
sp_a = _convert_to_sparse_tensor(sp_a)
with ops.name_scope(name, "SparseTensorDenseMatMul",
[sp_a.indices, sp_a.values, b]) as name:
b = ops.convert_to_tensor(b, name="b")
return gen_sparse_ops.sparse_tensor_dense_mat_mul(
a_indices=sp_a.indices,
a_values=sp_a.values,
a_shape=sp_a.dense_shape,
b=b,
adjoint_a=adjoint_a,
adjoint_b=adjoint_b)
@tf_export("sparse.softmax", v1=["sparse.softmax", "sparse_softmax"])
@deprecation.deprecated_endpoints("sparse_softmax")
def sparse_softmax(sp_input, name=None):
"""Applies softmax to a batched N-D `SparseTensor`.
The inputs represent an N-D SparseTensor with logical shape `[..., B, C]`
(where `N >= 2`), and with indices sorted in the canonical lexicographic
order.
This op is equivalent to applying the normal `tf.nn.softmax()` to each
innermost logical submatrix with shape `[B, C]`, but with the catch that *the
implicitly zero elements do not participate*. Specifically, the algorithm is
equivalent to:
(1) Applies `tf.nn.softmax()` to a densified view of each innermost
submatrix with shape `[B, C]`, along the size-C dimension;
(2) Masks out the original implicitly-zero locations;
(3) Renormalizes the remaining elements.
Hence, the `SparseTensor` result has exactly the same non-zero indices and
shape.
Example:
```python
# First batch:
# [? e.]
# [1. ? ]
# Second batch:
# [e ? ]
# [e e ]
shape = [2, 2, 2] # 3-D SparseTensor
values = np.asarray([[[0., np.e], [1., 0.]], [[np.e, 0.], [np.e, np.e]]])
indices = np.vstack(np.where(values)).astype(np.int64).T
result = tf.sparse.softmax(tf.SparseTensor(indices, values, shape))
# ...returning a 3-D SparseTensor, equivalent to:
# [? 1.] [1 ?]
# [1. ? ] and [.5 .5]
# where ? means implicitly zero.
```
Args:
sp_input: N-D `SparseTensor`, where `N >= 2`.
name: optional name of the operation.
Returns:
output: N-D `SparseTensor` representing the results.
"""
with ops.name_scope(name, "SparseSoftmax",
[sp_input.indices, sp_input.values]) as name:
out_vals = gen_sparse_ops.sparse_softmax(sp_input.indices, sp_input.values,
sp_input.dense_shape)
return sparse_tensor.SparseTensor(sp_input.indices, out_vals,
sp_input.dense_shape)
@tf_export("sparse.maximum", v1=["sparse.maximum", "sparse_maximum"])
@deprecation.deprecated_endpoints("sparse_maximum")
def sparse_maximum(sp_a, sp_b, name=None):
"""Returns the element-wise max of two SparseTensors.
Assumes the two SparseTensors have the same shape, i.e., no broadcasting.
Example:
```python
sp_zero = sparse_tensor.SparseTensor([[0]], [0], [7])
sp_one = sparse_tensor.SparseTensor([[1]], [1], [7])
res = tf.sparse.maximum(sp_zero, sp_one).eval()
# "res" should be equal to SparseTensor([[0], [1]], [0, 1], [7]).
```
Args:
sp_a: a `SparseTensor` operand whose dtype is real, and indices
lexicographically ordered.
sp_b: the other `SparseTensor` operand with the same requirements (and the
same shape).
name: optional name of the operation.
Returns:
output: the output SparseTensor.
"""
with ops.name_scope(
name, "SparseSparseMaximum",
[sp_a.indices, sp_a.values, sp_b.indices, sp_b.values]) as name:
out_indices, out_values = gen_sparse_ops.sparse_sparse_maximum(
sp_a.indices,
sp_a.values,
sp_a.dense_shape,
sp_b.indices,
sp_b.values,
sp_b.dense_shape,
name=name)
return sparse_tensor.SparseTensor(out_indices, out_values, sp_a.dense_shape)
@tf_export("sparse.minimum", v1=["sparse.minimum", "sparse_minimum"])
@deprecation.deprecated_endpoints("sparse_minimum")
def sparse_minimum(sp_a, sp_b, name=None):
"""Returns the element-wise min of two SparseTensors.
Assumes the two SparseTensors have the same shape, i.e., no broadcasting.
Example:
```python
sp_zero = sparse_tensor.SparseTensor([[0]], [0], [7])
sp_one = sparse_tensor.SparseTensor([[1]], [1], [7])
res = tf.sparse.minimum(sp_zero, sp_one).eval()
# "res" should be equal to SparseTensor([[0], [1]], [0, 0], [7]).
```
Args:
sp_a: a `SparseTensor` operand whose dtype is real, and indices
lexicographically ordered.
sp_b: the other `SparseTensor` operand with the same requirements (and the
same shape).
name: optional name of the operation.
Returns:
output: the output SparseTensor.
"""
with ops.name_scope(
name, "SparseSparseMinimum",
[sp_a.indices, sp_a.values, sp_b.indices, sp_b.values]) as name:
out_indices, out_values = gen_sparse_ops.sparse_sparse_minimum(
sp_a.indices,
sp_a.values,
sp_a.dense_shape,
sp_b.indices,
sp_b.values,
sp_b.dense_shape,
name=name)
return sparse_tensor.SparseTensor(out_indices, out_values, sp_a.dense_shape)
@tf_export("sparse.transpose", v1=["sparse.transpose", "sparse_transpose"])
@deprecation.deprecated_endpoints("sparse_transpose")
def sparse_transpose(sp_input, perm=None, name=None):
"""Transposes a `SparseTensor`
The returned tensor's dimension i will correspond to the input dimension
`perm[i]`. If `perm` is not given, it is set to (n-1...0), where n is
the rank of the input tensor. Hence by default, this operation performs a
regular matrix transpose on 2-D input Tensors.
For example, if `sp_input` has shape `[4, 5]` and `indices` / `values`:
[0, 3]: b
[0, 1]: a
[3, 1]: d
[2, 0]: c
then the output will be a `SparseTensor` of shape `[5, 4]` and
`indices` / `values`:
[0, 2]: c
[1, 0]: a
[1, 3]: d
[3, 0]: b
Args:
sp_input: The input `SparseTensor`.
perm: A permutation of the dimensions of `sp_input`.
name: A name prefix for the returned tensors (optional)
Returns:
A transposed `SparseTensor`.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
with ops.name_scope(name, "SparseTranspose", [sp_input]) as name:
if perm is None:
rank = array_ops.rank(sp_input)
perm = (rank - 1) - math_ops.range(0, rank, 1)
indices = sp_input.indices
transposed_indices = array_ops.transpose(
array_ops.gather(array_ops.transpose(indices), perm))
perm_ = tensor_util.constant_value(ops.convert_to_tensor(perm))
if perm_ is not None and sp_input.get_shape().is_fully_defined():
old_shape_ = sp_input.get_shape().as_list()
transposed_dense_shape = list(old_shape_) # Copy.
for i, p in enumerate(perm_):
transposed_dense_shape[i] = old_shape_[p]
else:
dense_shape = sp_input.dense_shape
transposed_dense_shape = array_ops.gather(dense_shape, perm)
transposed_st = sparse_tensor.SparseTensor(
transposed_indices, sp_input.values, transposed_dense_shape)
transposed_st = sparse_reorder(transposed_st)
return transposed_st
def _add_sparse_to_tensors_map(sp_input,
container=None,
shared_name=None,
name=None):
"""Add a `SparseTensor` to a `SparseTensorsMap` and return its handle.
Args:
sp_input: The input `SparseTensor`.
container: The container for the underlying `SparseTensorsMap` (optional).
shared_name: The shared name for the underlying `SparseTensorsMap`
(optional, defaults to the name of the newly created op).
name: A name prefix for the returned tensors (optional).
Returns:
A string 1-vector (1D `Tensor`), with the single element representing the
a unique handle to a `SparseTensor` stored by the `SparseTensorMap`
underlying this op.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
return gen_sparse_ops.add_sparse_to_tensors_map(
sp_input.indices,
sp_input.values,
sp_input.dense_shape,
container=container,
shared_name=shared_name,
name=name)
def _add_many_sparse_to_tensors_map(sp_input,
container=None,
shared_name=None,
name=None):
"""Add a minibatch `SparseTensor` to a `SparseTensorsMap`, return `N` handles.
The `SparseTensor` must have rank `R` greater than 1, and the first dimension
is treated as the minibatch dimension. Elements of the `SparseTensor`
must be sorted in increasing order of this first dimension. The serialized
`SparseTensor` objects going into each row of the output `Tensor` will have
rank `R-1`.
The minibatch size `N` is extracted from `sparse_shape[0]`.
Args:
sp_input: The input rank `R` `SparseTensor`.
container: The container for the underlying `SparseTensorsMap` (optional).
shared_name: The shared name for the underlying `SparseTensorsMap`
(optional, defaults to the name of the newly created op).
name: A name prefix for the returned tensors (optional).
Returns:
A string matrix (2-D `Tensor`) with `N` rows and `1` column.
Each row represents a unique handle to a `SparseTensor` stored by
the `SparseTensorMap` underlying this op.
Raises:
TypeError: If `sp_input` is not a `SparseTensor`.
"""
sp_input = _convert_to_sparse_tensor(sp_input)
return gen_sparse_ops.add_many_sparse_to_tensors_map(
sp_input.indices,
sp_input.values,
sp_input.dense_shape,
container=container,
shared_name=shared_name,
name=name)
def _take_many_sparse_from_tensors_map(sparse_map_op,
sparse_handles,
rank=None,
name=None):
"""Read `SparseTensors` from a `SparseTensorsMap` and concatenate them.
The input `sparse_handles` must be a string matrix of shape `[N, 1]` where
`N` is the minibatch size and the rows correspond to packed outputs of
`add_sparse_to_tensors_map`. The ranks of the original `SparseTensor` objects
must all match. When the final `SparseTensor` is created, it has rank one
higher than the ranks of the incoming `SparseTensor` objects (they have been
concatenated along a new row dimension).
The output `SparseTensor` object's shape values for all dimensions but the
first are the max across the input `SparseTensor` objects' shape values
for the corresponding dimensions. Its first shape value is `N`, the minibatch
size.
The input `SparseTensor` objects' indices are assumed ordered in
standard lexicographic order. If this is not the case, after this
step run `sparse.reorder` to restore index ordering.
For example, if the serialized input is a `[2, 3]` matrix representing two
original `SparseTensor` objects:
index = [ 0]
[10]
[20]
values = [1, 2, 3]
shape = [50]
and
index = [ 2]
[10]
values = [4, 5]
shape = [30]
then the final deserialized `SparseTensor` will be:
index = [0 0]
[0 10]
[0 20]
[1 2]
[1 10]
values = [1, 2, 3, 4, 5]
shape = [2 50]
Args:
sparse_map_op: The `Operation` that created the original handles.
Usually this is, e.g., `add_sparse_to_tensors_map(...).op`.
sparse_handles: 2-D `Tensor` of type `string` of shape `[N, 1]`.
The serialized and packed `SparseTensor` objects.
rank: (optional) Python int, the rank of the `SparseTensor` objects.
name: A name prefix for the returned tensors (optional)
Returns:
A `SparseTensor` representing the deserialized `SparseTensor`s,
concatenated along the `SparseTensor`s' first dimension.
All of the serialized `SparseTensor`s must have had the same rank and type.
"""
if not isinstance(sparse_map_op, ops.Operation):
raise TypeError("sparse_map_op be an Operation")
if sparse_map_op.type not in ("AddSparseToTensorsMap",
"AddManySparseToTensorsMap"):
raise TypeError(
"sparse_map_op must be one of AddSparseToTensorsMap or "
"AddSparseToTensorsMap. Instead, found `%s`." % sparse_map_op.type)
with ops.colocate_with(sparse_map_op):
shared_name = sparse_map_op.get_attr("shared_name") or sparse_map_op.name
output_indices, output_values, output_shape = (
gen_sparse_ops.take_many_sparse_from_tensors_map(
sparse_handles,
dtype=sparse_map_op.get_attr("T"),
container=sparse_map_op.get_attr("container"),
shared_name=shared_name,
name=name))
# Feed rank data back in, if available
output_indices.set_shape([None, rank])
output_shape.set_shape([rank])
return sparse_tensor.SparseTensor(output_indices, output_values, output_shape)
class _UnaryMapValueDispatcher(dispatch.OpDispatcher):
"""OpDispatcher for unary ops that maps base function across sparse values."""
def __init__(self, original_func):
self._original_func = original_func
func_name = get_canonical_name_for_symbol(original_func)
arg_names = tf_inspect.getfullargspec(original_func)[0]
self._x = arg_names[0]
original_func.__doc__ = (
original_func.__doc__.rstrip() + "\n\n" +
(" If `{x}` is a `SparseTensor`, returns\n"
" `SparseTensor({x}.indices, tf.{func}({x}.values, ...), "
"{x}.dense_shape)`").format(x=self._x, func=func_name))
def handle(self, args, kwargs):
if args:
x, args = args[0], args[1:]
else:
kwargs = kwargs.copy()
x = kwargs.pop(self._x, None)
if isinstance(x, sparse_tensor.SparseTensor):
return sparse_tensor.SparseTensor(
indices=x.indices,
values=self._original_func(x.values, *args, **kwargs),
dense_shape=x.dense_shape)
else:
return self.NOT_SUPPORTED
_UNARY_OPS = [
# TODO(b/120307967) Add dispatchers for additional TensorFlow ops.
math_ops.abs,
math_ops.negative,
math_ops.sign,
math_ops.square,
math_ops.sqrt,
math_ops.erf,
math_ops.tanh,
math_ops.bessel_i0e,
math_ops.bessel_i1e,
]
for unary_op in _UNARY_OPS:
_UnaryMapValueDispatcher(unary_op).register(unary_op)
|
tensorflow-master
|
tensorflow/python/ops/sparse_ops.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Gradients for operators defined in math_ops.py."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.compat import compat
from tensorflow.python.eager import context
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import tensor_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gen_array_ops
from tensorflow.python.ops import gen_math_ops
from tensorflow.python.ops import math_ops
def _safe_shape_div(x, y):
"""Divides `x / y` assuming `x, y >= 0`, treating `0 / 0 = 0`."""
return x // math_ops.maximum(y, 1)
@ops.RegisterGradient("ArgMax")
def _ArgMaxGrad(op, grad):
del op, grad
return [None, None]
@ops.RegisterGradient("ArgMin")
def _ArgMinGrad(op, grad):
del op, grad
return [None, None]
# TODO(rmlarsen): Implement gradient.
ops.NotDifferentiable("EuclideanNorm")
_empty_tuple = ()
def _IsScalar(x):
return x._shape_tuple() is _empty_tuple # pylint: disable=protected-access
@ops.RegisterGradient("Sum")
def _SumGrad(op, grad):
"""Gradient for Sum."""
# Fast path for when reducing to a scalar and ndims is known: adds only
# Reshape and Tile ops (and possibly a Shape).
input_0_shape = op.inputs[0]._shape_tuple() # pylint: disable=protected-access
if input_0_shape is not None:
axes = tensor_util.constant_value(op.inputs[1])
if axes is not None:
rank = len(input_0_shape)
if np.array_equal(axes, np.arange(rank)): # Reduce all dims.
if context.executing_eagerly():
ctx = context.context()
new_shape = ctx.ones_rank_cache().get(rank)
if new_shape is None:
new_shape = constant_op.constant([1] * rank, dtype=dtypes.int32)
ctx.ones_rank_cache().put(rank, new_shape)
else:
new_shape = [1] * rank
grad = array_ops.reshape(grad, new_shape)
# If shape is not fully defined (but rank is), we use Shape.
if None not in input_0_shape:
input_shape = constant_op.constant(input_0_shape, dtype=dtypes.int32)
else:
input_shape = array_ops.shape(op.inputs[0])
return [array_ops.tile(grad, input_shape), None]
input_shape = array_ops.shape(op.inputs[0])
# TODO(apassos) remove this once device placement for eager ops makes more
# sense.
with ops.colocate_with(input_shape):
output_shape_kept_dims = math_ops.reduced_shape(input_shape, op.inputs[1])
tile_scaling = _safe_shape_div(input_shape, output_shape_kept_dims)
grad = array_ops.reshape(grad, output_shape_kept_dims)
return [array_ops.tile(grad, tile_scaling), None]
def _MinOrMaxGrad(op, grad):
"""Gradient for Min or Max. Amazingly it's precisely the same code."""
input_shape = array_ops.shape(op.inputs[0])
output_shape_kept_dims = math_ops.reduced_shape(input_shape, op.inputs[1])
y = op.outputs[0]
y = array_ops.reshape(y, output_shape_kept_dims)
grad = array_ops.reshape(grad, output_shape_kept_dims)
# Compute the number of selected (maximum or minimum) elements in each
# reduction dimension. If there are multiple minimum or maximum elements
# then the gradient will be divided between them.
indicators = math_ops.cast(math_ops.equal(y, op.inputs[0]), grad.dtype)
num_selected = array_ops.reshape(
math_ops.reduce_sum(indicators, op.inputs[1]), output_shape_kept_dims)
return [math_ops.divide(indicators, num_selected) * grad, None]
@ops.RegisterGradient("Max")
def _MaxGrad(op, grad):
"""Gradient for Max."""
return _MinOrMaxGrad(op, grad)
@ops.RegisterGradient("Min")
def _MinGrad(op, grad):
return _MinOrMaxGrad(op, grad)
@ops.RegisterGradient("Mean")
def _MeanGrad(op, grad):
"""Gradient for Mean."""
sum_grad = _SumGrad(op, grad)[0]
input_shape = op.inputs[0]._shape_tuple() # pylint: disable=protected-access
output_shape = op.outputs[0]._shape_tuple() # pylint: disable=protected-access
if (input_shape is not None and output_shape is not None and
None not in input_shape and None not in output_shape):
input_size = np.prod(input_shape)
output_size = np.prod(output_shape)
factor = input_size // max(output_size, 1)
factor = constant_op.constant(factor, dtype=sum_grad.dtype)
else:
input_shape = array_ops.shape(op.inputs[0])
output_shape = array_ops.shape(op.outputs[0])
factor = _safe_shape_div(
math_ops.reduce_prod(input_shape), math_ops.reduce_prod(output_shape))
return math_ops.truediv(sum_grad, math_ops.cast(factor, sum_grad.dtype)), None
@ops.RegisterGradient("Prod")
def _ProdGrad(op, grad):
"""Gradient for Prod."""
# The gradient can be expressed by dividing the product by each entry of the
# input tensor, but this approach can't deal with zeros in the input.
# Here, we avoid this problem by composing the output as a product of two
# cumprod operations.
input_shape = array_ops.shape(op.inputs[0])
# Reshape reduction indices for the case where the parameter is a scalar
reduction_indices = array_ops.reshape(op.inputs[1], [-1])
# Expand grad to full input shape
output_shape_kept_dims = math_ops.reduced_shape(input_shape, op.inputs[1])
tile_scaling = _safe_shape_div(input_shape, output_shape_kept_dims)
grad = array_ops.reshape(grad, output_shape_kept_dims)
grad = array_ops.tile(grad, tile_scaling)
# Pack all reduced dimensions into a single one, so we can perform the
# cumprod ops. If the reduction dims list is empty, it defaults to float32,
# so we need to cast here. We put all the shape-related ops on CPU to avoid
# copying back and forth, and since listdiff is CPU only.
with ops.device("/cpu:0"):
rank = array_ops.rank(op.inputs[0])
reduction_indices = (reduction_indices + rank) % rank
reduced = math_ops.cast(reduction_indices, dtypes.int32)
idx = math_ops.range(0, rank)
other, _ = array_ops.setdiff1d(idx, reduced)
perm = array_ops.concat([reduced, other], 0)
reduced_num = math_ops.reduce_prod(array_ops.gather(input_shape, reduced))
other_num = math_ops.reduce_prod(array_ops.gather(input_shape, other))
permuted = array_ops.transpose(op.inputs[0], perm)
permuted_shape = array_ops.shape(permuted)
reshaped = array_ops.reshape(permuted, (reduced_num, other_num))
# Calculate product, leaving out the current entry
left = math_ops.cumprod(reshaped, axis=0, exclusive=True)
right = math_ops.cumprod(reshaped, axis=0, exclusive=True, reverse=True)
# For complex inputs, the gradient is in the conjugate direction.
y = array_ops.reshape(
math_ops.conj(left) * math_ops.conj(right), permuted_shape)
# Invert the transpose and reshape operations.
# Make sure to set the statically known shape information through a reshape.
out = grad * array_ops.transpose(y, array_ops.invert_permutation(perm))
return array_ops.reshape(out, input_shape), None
@ops.RegisterGradient("SegmentSum")
def _SegmentSumGrad(op, grad):
"""Gradient for SegmentSum."""
return array_ops.gather(grad, op.inputs[1]), None
@ops.RegisterGradient("SegmentMean")
def _SegmentMeanGrad(op, grad):
"""Gradient for SegmentMean."""
input_rank = array_ops.rank(op.inputs[0])
ones_shape = array_ops.concat([
array_ops.shape(op.inputs[1]),
array_ops.fill(array_ops.expand_dims(input_rank - 1, 0), 1)
], 0)
ones = array_ops.fill(ones_shape, constant_op.constant(1, dtype=grad.dtype))
scaled_grad = math_ops.divide(grad, math_ops.segment_sum(ones, op.inputs[1]))
return array_ops.gather(scaled_grad, op.inputs[1]), None
@ops.RegisterGradient("SparseSegmentSum")
def _SparseSegmentSumGrad(op, grad):
"""Gradient for SparseSegmentSum."""
input_rows = array_ops.shape(op.inputs[0])[0]
return (math_ops.unsorted_segment_sum(
array_ops.gather(grad, op.inputs[2]), op.inputs[1], input_rows), None,
None)
@ops.RegisterGradient("SparseSegmentSumWithNumSegments")
def _SparseSegmentSumWithNumSegmentsGrad(op, grad):
"""Gradient for SparseSegmentSumWithNumSegments."""
input_rows = array_ops.shape(op.inputs[0])[0]
return (math_ops.unsorted_segment_sum(
array_ops.gather(grad, op.inputs[2]), op.inputs[1], input_rows), None,
None, None)
@ops.RegisterGradient("SparseSegmentMean")
def _SparseSegmentMeanGrad(op, grad):
"""Gradient for SparseSegmentMean."""
dim0 = array_ops.shape(op.inputs[0])[0]
return (math_ops.sparse_segment_mean_grad(grad, op.inputs[1], op.inputs[2],
dim0), None, None)
@ops.RegisterGradient("SparseSegmentMeanWithNumSegments")
def _SparseSegmentMeanWithNumSegmentsGrad(op, grad):
"""Gradient for SparseSegmentMeanWithNumSegments."""
dim0 = array_ops.shape(op.inputs[0])[0]
return (math_ops.sparse_segment_mean_grad(grad, op.inputs[1], op.inputs[2],
dim0), None, None, None)
@ops.RegisterGradient("SparseSegmentSqrtN")
def _SparseSegmentSqrtNGrad(op, grad):
"""Gradient for SparseSegmentSqrtN."""
dim0 = array_ops.shape(op.inputs[0])[0]
return (math_ops.sparse_segment_sqrt_n_grad(grad, op.inputs[1], op.inputs[2],
dim0), None, None)
@ops.RegisterGradient("SparseSegmentSqrtNWithNumSegments")
def _SparseSegmentSqrtNWithNumSegmentsGrad(op, grad):
"""Gradient for SparseSegmentSqrtNWithNumSegments."""
dim0 = array_ops.shape(op.inputs[0])[0]
return (math_ops.sparse_segment_sqrt_n_grad(grad, op.inputs[1], op.inputs[2],
dim0), None, None, None)
def _SegmentMinOrMaxGrad(op, grad):
""" Gradient for SegmentMin and SegmentMax. """
zeros = array_ops.zeros_like(op.inputs[0], dtype=op.inputs[0].dtype)
# Get the number of selected (minimum or maximum) elements in each segment.
gathered_outputs = array_ops.gather(op.outputs[0], op.inputs[1])
is_selected = math_ops.equal(op.inputs[0], gathered_outputs)
num_selected = math_ops.segment_sum(
math_ops.cast(is_selected, grad.dtype), op.inputs[1])
# Compute the gradient for each segment. The gradient for the ith segment is
# divided evenly among the selected elements in that segment.
weighted_grads = math_ops.divide(grad, num_selected)
gathered_grads = array_ops.gather(weighted_grads, op.inputs[1])
return array_ops.where(is_selected, gathered_grads, zeros), None
@ops.RegisterGradient("SegmentMin")
def _SegmentMinGrad(op, grad):
"""Gradient for SegmentMin."""
return _SegmentMinOrMaxGrad(op, grad)
@ops.RegisterGradient("SegmentMax")
def _SegmentMaxGrad(op, grad):
"""Gradient for SegmentMax."""
return _SegmentMinOrMaxGrad(op, grad)
def _GatherDropNegatives(params,
ids,
zero_clipped_indices=None,
is_positive=None):
""" Helper function for unsorted segment ops.
Gathers params for
positive segment ids and gathers 0 for inputs with negative segment id.
Also returns the clipped indices and a boolean mask with the same shape
as ids where a positive id is masked as true. With this, the latter two
can be passed as arguments to this function to reuse them.
"""
if zero_clipped_indices is None:
zero_clipped_indices = math_ops.maximum(ids, array_ops.zeros_like(ids))
gathered = array_ops.gather(params, zero_clipped_indices)
if is_positive is None:
is_positive = math_ops.greater_equal(ids, 0)
# tf.where(condition, x, y) requires condition to have the same shape as x
# and y.
# todo(philjd): remove this if tf.where supports broadcasting (#9284)
for _ in range(gathered.shape.ndims - is_positive.shape.ndims):
is_positive = array_ops.expand_dims(is_positive, -1)
is_positive = (
is_positive & array_ops.ones_like(gathered, dtype=dtypes.bool))
# replace gathered params of negative indices with 0
zero_slice = array_ops.zeros_like(gathered)
return (array_ops.where(is_positive, gathered, zero_slice),
zero_clipped_indices, is_positive)
def _UnsortedSegmentMinOrMaxGrad(op, grad):
""" Gradient for UnsortedSegmentMin and UnsortedSegmentMax. """
# Get the number of selected (minimum or maximum) elements in each segment.
gathered_outputs, zero_clipped_indices, is_positive = \
_GatherDropNegatives(op.outputs[0], op.inputs[1])
is_selected = math_ops.equal(op.inputs[0], gathered_outputs)
is_selected = math_ops.logical_and(is_selected, is_positive)
num_selected = math_ops.unsorted_segment_sum(
math_ops.cast(is_selected, grad.dtype), op.inputs[1], op.inputs[2])
# Compute the gradient for each segment. The gradient for the ith segment is
# divided evenly among the selected elements in that segment.
weighted_grads = math_ops.divide(grad, num_selected)
gathered_grads, _, _ = _GatherDropNegatives(weighted_grads, None,
zero_clipped_indices, is_positive)
zeros = array_ops.zeros_like(gathered_grads)
return array_ops.where(is_selected, gathered_grads, zeros), None, None
@ops.RegisterGradient("UnsortedSegmentSum")
def _UnsortedSegmentSumGrad(op, grad):
"""Gradient for UnsortedSegmentSum."""
return _GatherDropNegatives(grad, op.inputs[1])[0], None, None
@ops.RegisterGradient("UnsortedSegmentMax")
def _UnsortedSegmentMaxGrad(op, grad):
""" Gradient for UnsortedSegmentMax. """
return _UnsortedSegmentMinOrMaxGrad(op, grad)
@ops.RegisterGradient("UnsortedSegmentMin")
def _UnsortedSegmentMinGrad(op, grad):
""" Gradient for UnsortedSegmentMin. """
return _UnsortedSegmentMinOrMaxGrad(op, grad)
@ops.RegisterGradient("UnsortedSegmentProd")
def _UnsortedSegmentProdGrad(op, grad):
""" Gradient for UnsortedSegmentProd.
The gradient can be expressed for each segment by dividing the segment's
product by each element of the segment input tensor, but this approach can't
deal with zeros in the input.
Unlike reduce_prod we can't use cumsum here as individual segments may have
a different number of elements. Therefore we consider three cases:
1) A segment input contains no zeros and we can safely divide by the input
tensor.
2) A segment contains exactly one zero. Then the gradient of each input of
the segment is zero except for the 0-input, there the gradient is
the product of the remaining segment entries.
3) A segment contains at least two zeros. The gradient is zero for all
segment inputs.
"""
# Note that unsorted_segment_sum will filter out the negative indices,
# so we don't need to do a logical_and with is_positive here
is_zero = math_ops.equal(op.inputs[0], 0)
num_zeros = gen_math_ops.unsorted_segment_sum(
math_ops.cast(is_zero, dtype=dtypes.int32), op.inputs[1], op.inputs[2])
# handle case 3 and set the gradient to 0 for segments with more than one
# 0 as input
grad = array_ops.where(
math_ops.greater(num_zeros, 1), array_ops.zeros_like(grad), grad)
# replace all zeros with ones and compute the unsorted_segment_prod
non_zero_data = array_ops.where(is_zero, array_ops.ones_like(op.inputs[0]),
op.inputs[0])
non_zero_prod = gen_math_ops.unsorted_segment_prod(non_zero_data,
op.inputs[1], op.inputs[2])
# clip the indices for gather to be positive
zero_clipped_indices = math_ops.maximum(op.inputs[1],
array_ops.zeros_like(op.inputs[1]))
gathered_prod = array_ops.gather(op.outputs[0], zero_clipped_indices)
gathered_non_zero_prod = array_ops.gather(non_zero_prod, zero_clipped_indices)
prod_divided_by_el = gathered_prod / op.inputs[0] # May contain nan/inf.
# Now fetch the individual results for segments containing 0 and those that
# don't. is_zero will also fetch results for entries with negative index
# but the following gather_drop_negatives sets the corresponding entry in
# grad to 0 for these
partial_derivative = array_ops.where(is_zero, gathered_non_zero_prod,
prod_divided_by_el)
gathered_grad = _GatherDropNegatives(grad, op.inputs[1],
zero_clipped_indices)[0]
return gathered_grad * partial_derivative, None, None
@ops.RegisterGradient("Abs")
def _AbsGrad(op, grad):
x = op.inputs[0]
return grad * math_ops.sign(x)
@ops.RegisterGradient("Neg")
def _NegGrad(_, grad):
"""Returns -grad."""
return -grad
@ops.RegisterGradient("Inv")
def _InvGrad(op, grad):
"""Returns -grad * (1 / x^2)."""
y = op.outputs[0] # y = 1 / x
return gen_math_ops.reciprocal_grad(y, grad)
@ops.RegisterGradient("Reciprocal")
def _ReciprocalGrad(op, grad):
"""Returns -grad * (1 / x^2)."""
y = op.outputs[0] # y = 1 / x
return gen_math_ops.reciprocal_grad(y, grad)
@ops.RegisterGradient("InvGrad")
def _InvGradGrad(op, grad):
b = op.inputs[1]
# op.output[0]: y = -b * conj(a)^2
with ops.control_dependencies([grad]):
ca = math_ops.conj(op.inputs[0])
cg = math_ops.conj(grad)
return cg * -2.0 * b * ca, gen_math_ops.reciprocal_grad(ca, grad)
@ops.RegisterGradient("ReciprocalGrad")
def _ReciprocalGradGrad(op, grad):
b = op.inputs[1]
# op.output[0]: y = -b * conj(a)^2
with ops.control_dependencies([grad]):
ca = math_ops.conj(op.inputs[0])
cg = math_ops.conj(grad)
return cg * -2.0 * b * ca, gen_math_ops.reciprocal_grad(ca, grad)
@ops.RegisterGradient("Square")
def _SquareGrad(op, grad):
x = op.inputs[0]
# Added control dependencies to prevent 2*x from being computed too early.
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
y = constant_op.constant(2.0, dtype=x.dtype)
return math_ops.multiply(grad, math_ops.multiply(x, y))
@ops.RegisterGradient("Sqrt")
def _SqrtGrad(op, grad):
y = op.outputs[0] # y = x^(1/2)
return gen_math_ops.sqrt_grad(y, grad)
@ops.RegisterGradient("SqrtGrad")
def _SqrtGradGrad(op, grad):
a = op.inputs[0]
y = op.outputs[0] # y = 0.5 * b / conj(a)
with ops.control_dependencies([grad]):
if compat.forward_compatible(2019, 9, 14):
ga = gen_math_ops.xdivy(grad, a)
return -gen_math_ops.mul_no_nan(y, math_ops.conj(ga)), 0.5 * ga
else:
ga = grad / a
return -math_ops.conj(ga) * y, 0.5 * ga
@ops.RegisterGradient("Rsqrt")
def _RsqrtGrad(op, grad):
"""Returns -0.5 * grad * conj(y)^3."""
y = op.outputs[0] # y = x^(-1/2)
return gen_math_ops.rsqrt_grad(y, grad)
@ops.RegisterGradient("RsqrtGrad")
def _RsqrtGradGrad(op, grad):
"""Returns backprop gradient for f(a,b) = -0.5 * b * conj(a)^3."""
a = op.inputs[0] # a = x^{-1/2}
b = op.inputs[1] # backprop gradient for a
with ops.control_dependencies([grad]):
ca = math_ops.conj(a)
cg = math_ops.conj(grad)
grad_a = -1.5 * cg * b * math_ops.square(ca)
grad_b = gen_math_ops.rsqrt_grad(ca, grad)
return grad_a, grad_b
@ops.RegisterGradient("Exp")
def _ExpGrad(op, grad):
"""Returns grad * exp(x)."""
y = op.outputs[0] # y = e^x
with ops.control_dependencies([grad]):
y = math_ops.conj(y)
if compat.forward_compatible(2019, 9, 14):
return math_ops.mul_no_nan(y, grad)
else:
return grad * y
@ops.RegisterGradient("Expm1")
def _Expm1Grad(op, grad):
"""Returns grad * exp(x)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
y = math_ops.exp(x)
if compat.forward_compatible(2019, 9, 14):
return math_ops.mul_no_nan(y, grad)
else:
return grad * y
@ops.RegisterGradient("Log")
def _LogGrad(op, grad):
"""Returns grad * (1/x)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
if compat.forward_compatible(2019, 9, 14):
return gen_math_ops.xdivy(grad, x)
else:
return grad * math_ops.reciprocal(x)
@ops.RegisterGradient("Log1p")
def _Log1pGrad(op, grad):
"""Returns grad * (1/(1 + x))."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
if compat.forward_compatible(2019, 9, 14):
return gen_math_ops.xdivy(grad, 1 + x)
else:
return grad * math_ops.reciprocal(1 + x)
@ops.RegisterGradient("Xlogy")
def _XLogyGrad(op, grad):
"""Returns gradient of xlogy(x, y) with respect to x and y."""
x = op.inputs[0]
y = op.inputs[1]
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
with ops.control_dependencies([grad]):
not_zero_x = math_ops.cast(
math_ops.not_equal(x, math_ops.cast(0., dtype=x.dtype)), dtype=x.dtype)
partial_x = gen_math_ops.xlogy(not_zero_x, y)
partial_y = gen_math_ops.xdivy(x, y)
return (array_ops.reshape(math_ops.reduce_sum(partial_x * grad, rx), sx),
array_ops.reshape(math_ops.reduce_sum(partial_y * grad, ry), sy))
@ops.RegisterGradient("Xdivy")
def _XDivyGrad(op, grad):
"""Returns gradient of xdivy(x, y) with respect to x and y."""
x = op.inputs[0]
y = op.inputs[1]
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
with ops.control_dependencies([grad]):
not_zero_x = math_ops.cast(
math_ops.not_equal(x, math_ops.cast(0., dtype=x.dtype)), dtype=x.dtype)
partial_x = gen_math_ops.xdivy(not_zero_x, y)
partial_y = gen_math_ops.xdivy(math_ops.negative(x), y**2)
return (array_ops.reshape(math_ops.reduce_sum(partial_x * grad, rx), sx),
array_ops.reshape(math_ops.reduce_sum(partial_y * grad, ry), sy))
@ops.RegisterGradient("Sinh")
def _SinhGrad(op, grad):
"""Returns grad * cosh(x)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
return grad * math_ops.cosh(x)
@ops.RegisterGradient("Cosh")
def _CoshGrad(op, grad):
"""Returns grad * sinh(x)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
return grad * math_ops.sinh(x)
@ops.RegisterGradient("Tanh")
def _TanhGrad(op, grad):
"""Returns grad * (1 - tanh(x) * tanh(x))."""
y = op.outputs[0] # y = tanh(x)
with ops.control_dependencies([grad]):
y = math_ops.conj(y)
return gen_math_ops.tanh_grad(y, grad)
@ops.RegisterGradient("Asinh")
def _AsinhGrad(op, grad):
"""Returns grad * 1/cosh(y)."""
y = op.outputs[0]
with ops.control_dependencies([grad]):
y = math_ops.conj(y)
return grad / math_ops.cosh(y)
@ops.RegisterGradient("Acosh")
def _AcoshGrad(op, grad):
"""Returns grad * 1/sinh(y)."""
y = op.outputs[0]
with ops.control_dependencies([grad]):
y = math_ops.conj(y)
if compat.forward_compatible(2019, 9, 14):
return math_ops.xdivy(grad, math_ops.sinh(y))
else:
return grad / math_ops.sinh(y)
@ops.RegisterGradient("Atanh")
def _AtanhGrad(op, grad):
"""Returns grad * 1/ (1 - x^2)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
x2 = math_ops.square(x)
one = constant_op.constant(1, dtype=grad.dtype)
inv = math_ops.reciprocal(math_ops.subtract(one, x2))
return grad * inv
@ops.RegisterGradient("TanhGrad")
def _TanhGradGrad(op, grad):
with ops.control_dependencies([grad]):
a = math_ops.conj(op.inputs[0])
b = math_ops.conj(op.inputs[1])
return grad * -2.0 * b * a, gen_math_ops.tanh_grad(a, grad)
@ops.RegisterGradient("Erf")
def _ErfGrad(op, grad):
"""Returns grad * 2/sqrt(pi) * exp(-x**2)."""
x = op.inputs[0]
two_over_root_pi = constant_op.constant(2 / np.sqrt(np.pi), dtype=grad.dtype)
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
return grad * two_over_root_pi * math_ops.exp(-math_ops.square(x))
@ops.RegisterGradient("Erfc")
def _ErfcGrad(op, grad):
"""Returns -grad * 2/sqrt(pi) * exp(-x**2)."""
x = op.inputs[0]
minus_two_over_root_pi = constant_op.constant(
-2 / np.sqrt(np.pi), dtype=grad.dtype)
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
return grad * minus_two_over_root_pi * math_ops.exp(-math_ops.square(x))
@ops.RegisterGradient("Lgamma")
def _LgammaGrad(op, grad):
"""Returns grad * digamma(x)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
if compat.forward_compatible(2019, 9, 14):
return math_ops.mul_no_nan(math_ops.digamma(x), grad)
else:
return grad * math_ops.digamma(x)
@ops.RegisterGradient("Digamma")
def _DigammaGrad(op, grad):
"""Compute gradient of the digamma function with respect to its argument."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
partial_x = math_ops.polygamma(array_ops.constant(1, dtype=x.dtype), x)
if compat.forward_compatible(2019, 9, 14):
return math_ops.mul_no_nan(partial_x, grad)
else:
return grad * partial_x
@ops.RegisterGradient("BesselI0e")
def _BesselI0eGrad(op, grad):
"""Compute gradient of bessel_i0e(x) with respect to its argument."""
x = op.inputs[0]
y = op.outputs[0]
with ops.control_dependencies([grad]):
partial_x = (math_ops.bessel_i1e(x) - math_ops.sign(x) * y)
if compat.forward_compatible(2019, 9, 14):
return math_ops.mul_no_nan(partial_x, grad)
else:
return grad * partial_x
@ops.RegisterGradient("BesselI1e")
def _BesselI1eGrad(op, grad):
"""Compute gradient of bessel_i1e(x) with respect to its argument."""
x = op.inputs[0]
y = op.outputs[0]
with ops.control_dependencies([grad]):
# For x = 0, the correct gradient is 0.5.
# However, the main branch gives NaN because of the division by x, so
# we impute the gradient manually.
# An alternative solution is to express the gradient via bessel_i0e and
# bessel_i2e, but the latter is not yet implemented in Eigen.
eps = np.finfo(x.dtype.as_numpy_dtype).eps
zeros = array_ops.zeros_like(x)
x_is_not_tiny = math_ops.abs(x) > eps
safe_x = array_ops.where(x_is_not_tiny, x, eps + zeros)
dy_dx = math_ops.bessel_i0e(safe_x) - y * (
math_ops.sign(safe_x) + math_ops.reciprocal(safe_x))
dy_dx = array_ops.where(x_is_not_tiny, dy_dx, 0.5 + zeros)
if compat.forward_compatible(2019, 9, 14):
return math_ops.mul_no_nan(dy_dx, grad)
else:
return grad * dy_dx
@ops.RegisterGradient("Igamma")
def _IgammaGrad(op, grad):
"""Returns gradient of igamma(a, x) with respect to a and x."""
a = op.inputs[0]
x = op.inputs[1]
sa = array_ops.shape(a)
sx = array_ops.shape(x)
ra, rx = gen_array_ops.broadcast_gradient_args(sa, sx)
with ops.control_dependencies([grad]):
partial_a = gen_math_ops.igamma_grad_a(a, x)
# Perform operations in log space before summing, because Gamma(a)
# and Gamma'(a) can grow large.
partial_x = math_ops.exp(-x + (a - 1) * math_ops.log(x) -
math_ops.lgamma(a))
if compat.forward_compatible(2019, 9, 14):
return (array_ops.reshape(
math_ops.reduce_sum(math_ops.mul_no_nan(partial_a, grad), ra), sa),
array_ops.reshape(
math_ops.reduce_sum(math_ops.mul_no_nan(partial_x, grad), rx),
sx))
else:
return (array_ops.reshape(math_ops.reduce_sum(partial_a * grad, ra), sa),
array_ops.reshape(math_ops.reduce_sum(partial_x * grad, rx), sx))
@ops.RegisterGradient("Igammac")
def _IgammacGrad(op, grad):
"""Returns gradient of igammac(a, x) = 1 - igamma(a, x) w.r.t. a and x."""
igamma_grad_a, igamma_grad_x = _IgammaGrad(op, grad)
return (-igamma_grad_a, -igamma_grad_x)
@ops.RegisterGradient("Betainc")
def _BetaincGrad(op, grad):
"""Returns gradient of betainc(a, b, x) with respect to x."""
# TODO(ebrevdo): Perhaps add the derivative w.r.t. a, b
a, b, x = op.inputs
# two cases: x is a scalar and a/b are same-shaped tensors, or vice
# versa; so its sufficient to check against shape(a).
sa = array_ops.shape(a)
sx = array_ops.shape(x)
_, rx = gen_array_ops.broadcast_gradient_args(sa, sx)
# Perform operations in log space before summing, because terms
# can grow large.
log_beta = (
gen_math_ops.lgamma(a) + gen_math_ops.lgamma(b) -
gen_math_ops.lgamma(a + b))
partial_x = math_ops.exp((b - 1) * math_ops.log(1 - x) +
(a - 1) * math_ops.log(x) - log_beta)
# TODO(b/36815900): Mark None return values as NotImplemented
if compat.forward_compatible(2019, 9, 14):
return (
None, # da
None, # db
array_ops.reshape(
math_ops.reduce_sum(math_ops.mul_no_nan(partial_x, grad), rx), sx))
else:
return (
None, # da
None, # db
array_ops.reshape(math_ops.reduce_sum(partial_x * grad, rx), sx))
@ops.RegisterGradient("Zeta")
def _ZetaGrad(op, grad):
"""Returns gradient of zeta(x, q) with respect to x and q."""
# TODO(tillahoffmann): Add derivative with respect to x
x = op.inputs[0]
q = op.inputs[1]
# Broadcast gradients
sx = array_ops.shape(x)
sq = array_ops.shape(q)
unused_rx, rq = gen_array_ops.broadcast_gradient_args(sx, sq)
# Evaluate gradient
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
q = math_ops.conj(q)
partial_q = -x * math_ops.zeta(x + 1, q)
# TODO(b/36815900): Mark None return values as NotImplemented
if compat.forward_compatible(2019, 9, 14):
return (None,
array_ops.reshape(
math_ops.reduce_sum(math_ops.mul_no_nan(partial_q, grad), rq),
sq))
else:
return (None,
array_ops.reshape(math_ops.reduce_sum(partial_q * grad, rq), sq))
@ops.RegisterGradient("Polygamma")
def _PolygammaGrad(op, grad):
"""Returns gradient of psi(n, x) with respect to n and x."""
# TODO(tillahoffmann): Add derivative with respect to n
n = op.inputs[0]
x = op.inputs[1]
# Broadcast gradients
sn = array_ops.shape(n)
sx = array_ops.shape(x)
unused_rn, rx = gen_array_ops.broadcast_gradient_args(sn, sx)
# Evaluate gradient
with ops.control_dependencies([grad]):
n = math_ops.conj(n)
x = math_ops.conj(x)
partial_x = math_ops.polygamma(n + 1, x)
# TODO(b/36815900): Mark None return values as NotImplemented
if compat.forward_compatible(2019, 9, 14):
return (None,
array_ops.reshape(
math_ops.reduce_sum(math_ops.mul_no_nan(partial_x, grad), rx),
sx))
else:
return (None,
array_ops.reshape(math_ops.reduce_sum(partial_x * grad, rx), sx))
@ops.RegisterGradient("Sigmoid")
def _SigmoidGrad(op, grad):
"""Returns grad * sigmoid(x) * (1 - sigmoid(x))."""
y = op.outputs[0] # y = sigmoid(x)
with ops.control_dependencies([grad]):
y = math_ops.conj(y)
return gen_math_ops.sigmoid_grad(y, grad)
@ops.RegisterGradient("SigmoidGrad")
def _SigmoidGradGrad(op, grad):
with ops.control_dependencies([grad]):
a = math_ops.conj(op.inputs[0])
b = math_ops.conj(op.inputs[1])
gb = grad * b
return gb - 2.0 * gb * a, gen_math_ops.sigmoid_grad(a, grad)
@ops.RegisterGradient("Sign")
def _SignGrad(op, _):
"""Returns 0."""
x = op.inputs[0]
return array_ops.zeros(array_ops.shape(x), dtype=x.dtype)
@ops.RegisterGradient("Sin")
def _SinGrad(op, grad):
"""Returns grad * cos(x)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
return grad * math_ops.cos(x)
@ops.RegisterGradient("Cos")
def _CosGrad(op, grad):
"""Returns grad * -sin(x)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
return -grad * math_ops.sin(x)
@ops.RegisterGradient("Tan")
def _TanGrad(op, grad):
"""Returns grad * 1/sec^2(x)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
secx = math_ops.reciprocal(math_ops.cos(x))
secx2 = math_ops.square(secx)
if compat.forward_compatible(2019, 9, 14):
return math_ops.mul_no_nan(secx2, grad)
else:
return secx2 * grad
@ops.RegisterGradient("Asin")
def _AsinGrad(op, grad):
"""Returns grad * 1/sqrt(1-x^2)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
x2 = math_ops.square(x)
one = constant_op.constant(1, dtype=grad.dtype)
den = math_ops.sqrt(math_ops.subtract(one, x2))
if compat.forward_compatible(2019, 9, 14):
return math_ops.xdivy(grad, den)
else:
inv = math_ops.reciprocal(den)
return grad * inv
@ops.RegisterGradient("Acos")
def _AcosGrad(op, grad):
"""Returns grad * -1/sqrt(1-x^2)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
x2 = math_ops.square(x)
one = constant_op.constant(1, dtype=grad.dtype)
den = math_ops.sqrt(math_ops.subtract(one, x2))
if compat.forward_compatible(2019, 9, 14):
return -math_ops.xdivy(grad, den)
else:
inv = math_ops.reciprocal(den)
return -grad * inv
@ops.RegisterGradient("Atan")
def _AtanGrad(op, grad):
"""Returns grad * 1/ (1 + x^2)."""
x = op.inputs[0]
with ops.control_dependencies([grad]):
x = math_ops.conj(x)
x2 = math_ops.square(x)
one = constant_op.constant(1, dtype=grad.dtype)
inv = math_ops.reciprocal(math_ops.add(one, x2))
return grad * inv
@ops.RegisterGradient("Atan2")
def _Atan2Grad(op, grad):
"""Returns grad * x / (x^2 + y^2), grad * -y / (x^2 + y^2)."""
y = op.inputs[0]
x = op.inputs[1]
with ops.control_dependencies([grad]):
if compat.forward_compatible(2019, 9, 14):
grad_inv = math_ops.xdivy(grad, (math_ops.square(x) + math_ops.square(y)))
else:
grad_inv = grad / (math_ops.square(x) + math_ops.square(y))
return x * grad_inv, -y * grad_inv
@ops.RegisterGradient("AddN")
def _AddNGrad(op, grad):
"""Copies the gradient to all inputs."""
# Not broadcasting.
return [grad] * len(op.inputs)
def _ShapesFullySpecifiedAndEqual(x, y, grad):
# pylint: disable=protected-access
x_shape = x._shape_tuple()
y_shape = y._shape_tuple()
grad_shape = grad._shape_tuple()
# pylint: enable=protected-access
return (x_shape == y_shape and x_shape == grad_shape and
x_shape is not None and None not in x_shape)
@ops.RegisterGradient("Add")
@ops.RegisterGradient("AddV2")
def _AddGrad(op, grad):
"""Gradient for Add."""
y = op.inputs[1]
skip_input_indices = None
try:
skip_input_indices = op.skip_input_indices
if skip_input_indices is not None and 1 in skip_input_indices and _IsScalar(
y):
return grad, None
except AttributeError:
# No gradient skipping, so do the full gradient computation
pass
x = op.inputs[0]
if (isinstance(grad, ops.Tensor) and
_ShapesFullySpecifiedAndEqual(x, y, grad)):
return grad, grad
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
if skip_input_indices is not None and 0 in skip_input_indices:
gx = None
else:
gx = array_ops.reshape(math_ops.reduce_sum(grad, rx), sx)
if skip_input_indices is not None and 1 in skip_input_indices:
gy = None
else:
gy = array_ops.reshape(math_ops.reduce_sum(grad, ry), sy)
return (gx, gy)
@ops.RegisterGradient("Sub")
def _SubGrad(op, grad):
"""Gradient for Sub."""
x = op.inputs[0]
y = op.inputs[1]
if (isinstance(grad, ops.Tensor) and
_ShapesFullySpecifiedAndEqual(x, y, grad)):
return grad, -grad
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
return (array_ops.reshape(math_ops.reduce_sum(grad, rx), sx),
array_ops.reshape(-math_ops.reduce_sum(grad, ry), sy))
@ops.RegisterGradient("Mul")
def _MulGrad(op, grad):
"""The gradient of scalar multiplication."""
x = op.inputs[0]
y = op.inputs[1]
if (isinstance(grad, ops.Tensor) and
_ShapesFullySpecifiedAndEqual(x, y, grad) and
grad.dtype in (dtypes.int32, dtypes.float32)):
return gen_math_ops.mul(grad, y), gen_math_ops.mul(grad, x)
assert x.dtype.base_dtype == y.dtype.base_dtype, (x.dtype, " vs. ", y.dtype)
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
x = math_ops.conj(x)
y = math_ops.conj(y)
return (array_ops.reshape(
math_ops.reduce_sum(gen_math_ops.mul(grad, y), rx), sx),
array_ops.reshape(
math_ops.reduce_sum(gen_math_ops.mul(x, grad), ry), sy))
@ops.RegisterGradient("MulNoNan")
def _MulNoNanGrad(op, grad):
"""The gradient of scalar multiplication with NaN-suppression."""
x = op.inputs[0]
y = op.inputs[1]
if (isinstance(grad, ops.Tensor) and
_ShapesFullySpecifiedAndEqual(x, y, grad)):
return gen_math_ops.mul_no_nan(grad, y), gen_math_ops.mul_no_nan(x, grad)
assert x.dtype.base_dtype == y.dtype.base_dtype, (x.dtype, " vs. ", y.dtype)
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
return (array_ops.reshape(
math_ops.reduce_sum(gen_math_ops.mul_no_nan(grad, y), rx), sx),
array_ops.reshape(
math_ops.reduce_sum(gen_math_ops.mul_no_nan(x, grad), ry), sy))
@ops.RegisterGradient("Div")
def _DivGrad(op, grad):
"""The gradient for the Div operator."""
x = op.inputs[0]
y = op.inputs[1]
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
x = math_ops.conj(x)
y = math_ops.conj(y)
if compat.forward_compatible(2019, 9, 14):
return (array_ops.reshape(
math_ops.reduce_sum(math_ops.xdivy(grad, y), rx), sx),
array_ops.reshape(
math_ops.reduce_sum(
math_ops.mul_no_nan(
math_ops.divide(math_ops.divide(-x, y), y), grad), ry),
sy))
else:
return (array_ops.reshape(
math_ops.reduce_sum(math_ops.divide(grad, y), rx), sx),
array_ops.reshape(
math_ops.reduce_sum(
grad * math_ops.divide(math_ops.divide(-x, y), y), ry), sy))
@ops.RegisterGradient("FloorDiv")
def _FloorDivGrad(_, unused_grad):
"""The gradient for the FloorDiv operator."""
return None, None
@ops.RegisterGradient("FloorMod")
def _FloorModGrad(op, grad):
"""Returns grad * (1, -floor(x/y))."""
x = math_ops.conj(op.inputs[0])
y = math_ops.conj(op.inputs[1])
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
floor_xy = math_ops.floor_div(x, y)
gx = array_ops.reshape(math_ops.reduce_sum(grad, rx), sx)
gy = array_ops.reshape(
math_ops.reduce_sum(grad * math_ops.negative(floor_xy), ry), sy)
return gx, gy
@ops.RegisterGradient("TruncateDiv")
def _TruncateDivGrad(_, unused_grad):
return None, None
@ops.RegisterGradient("RealDiv")
def _RealDivGrad(op, grad):
"""RealDiv op gradient."""
x = op.inputs[0]
y = op.inputs[1]
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
x = math_ops.conj(x)
y = math_ops.conj(y)
if compat.forward_compatible(2019, 9, 14):
return (array_ops.reshape(
math_ops.reduce_sum(math_ops.xdivy(grad, y), rx), sx),
array_ops.reshape(
math_ops.reduce_sum(
math_ops.mul_no_nan(
math_ops.realdiv(math_ops.realdiv(-x, y), y), grad),
ry), sy))
else:
return (array_ops.reshape(
math_ops.reduce_sum(math_ops.realdiv(grad, y), rx), sx),
array_ops.reshape(
math_ops.reduce_sum(
grad * math_ops.realdiv(math_ops.realdiv(-x, y), y), ry),
sy))
@ops.RegisterGradient("DivNoNan")
def _DivNoNanGrad(op, grad):
"""DivNoNan op gradient."""
x = op.inputs[0]
y = op.inputs[1]
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
x = math_ops.conj(x)
y = math_ops.conj(y)
if compat.forward_compatible(2019, 9, 14):
return (array_ops.reshape(
math_ops.reduce_sum(math_ops.div_no_nan(grad, y), rx), sx),
array_ops.reshape(
math_ops.reduce_sum(
math_ops.mul_no_nan(
math_ops.div_no_nan(math_ops.div_no_nan(-x, y), y),
grad), ry), sy))
else:
return (array_ops.reshape(
math_ops.reduce_sum(math_ops.div_no_nan(grad, y), rx), sx),
array_ops.reshape(
math_ops.reduce_sum(
grad * math_ops.div_no_nan(math_ops.div_no_nan(-x, y), y),
ry), sy))
@ops.RegisterGradient("Pow")
def _PowGrad(op, grad):
"""Returns grad * (y*x^(y-1), z*log(x))."""
x = op.inputs[0]
y = op.inputs[1]
z = op.outputs[0]
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
x = math_ops.conj(x)
y = math_ops.conj(y)
z = math_ops.conj(z)
if compat.forward_compatible(2019, 9, 14):
gx = array_ops.reshape(
math_ops.reduce_sum(
gen_math_ops.mul_no_nan(y * math_ops.pow(x, y - 1), grad), rx), sx)
else:
gx = array_ops.reshape(
math_ops.reduce_sum(grad * y * math_ops.pow(x, y - 1), rx), sx)
# Avoid false singularity at x = 0
if x.dtype.is_complex:
# real(x) < 0 is fine for the complex case
mask = math_ops.not_equal(x, 0)
else:
# There's no sensible real value to return if x < 0, so return 0
mask = x > 0
safe_x = array_ops.where(mask, x, array_ops.ones_like(x))
log_x = array_ops.where(mask, math_ops.log(safe_x), array_ops.zeros_like(x))
if compat.forward_compatible(2019, 9, 14):
gy = array_ops.reshape(
math_ops.reduce_sum(gen_math_ops.mul_no_nan(z * log_x, grad), ry), sy)
else:
gy = array_ops.reshape(math_ops.reduce_sum(grad * z * log_x, ry), sy)
return gx, gy
def _MaximumMinimumGradInputOnly(op, grad, selector_op):
x = op.inputs[0]
y = op.inputs[1]
zeros = array_ops.zeros_like(grad)
xmask = selector_op(x, y)
xgrad = array_ops.where(xmask, grad, zeros)
ygrad = None # Return None for ygrad since the config allows that.
return (xgrad, ygrad)
def _MaximumMinimumGrad(op, grad, selector_op):
"""Factor out the code for the gradient of Maximum or Minimum."""
y = op.inputs[1]
skip_input_indices = None
try:
skip_input_indices = op.skip_input_indices
if skip_input_indices is not None and 1 in skip_input_indices and _IsScalar(
y):
# When we want to get gradients for the first input only, and the second
# input tensor is a scalar, we can do a much simpler calculation
return _MaximumMinimumGradInputOnly(op, grad, selector_op)
except AttributeError:
# No gradient skipping, so do the full gradient computation
pass
x = op.inputs[0]
gdtype = grad.dtype
sx = array_ops.shape(x)
sy = array_ops.shape(y)
gradshape = array_ops.shape(grad)
zeros = array_ops.zeros(gradshape, gdtype)
xmask = selector_op(x, y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
if skip_input_indices is not None and 0 in skip_input_indices:
gx = None
else:
xgrad = array_ops.where(xmask, grad, zeros)
gx = array_ops.reshape(math_ops.reduce_sum(xgrad, rx), sx)
if skip_input_indices is not None and 1 in skip_input_indices:
gy = None
else:
ygrad = array_ops.where(xmask, zeros, grad)
gy = array_ops.reshape(math_ops.reduce_sum(ygrad, ry), sy)
return (gx, gy)
@ops.RegisterGradient("Maximum")
def _MaximumGrad(op, grad):
"""Returns grad*(x > y, x <= y) with type of grad."""
return _MaximumMinimumGrad(op, grad, math_ops.greater_equal)
@ops.RegisterGradient("Minimum")
def _MinimumGrad(op, grad):
"""Returns grad*(x < y, x >= y) with type of grad."""
return _MaximumMinimumGrad(op, grad, math_ops.less_equal)
@ops.RegisterGradient("SquaredDifference")
def _SquaredDifferenceGrad(op, grad):
"""Returns the gradient for (x-y)^2."""
x = op.inputs[0]
y = op.inputs[1]
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
with ops.control_dependencies([grad]):
# The parens ensure that if grad is IndexedSlices, it'll get multiplied by
# Tensor (not a number like 2.0) which causes it to convert to Tensor.
x_grad = math_ops.scalar_mul(2.0, grad) * (x - y)
return (array_ops.reshape(math_ops.reduce_sum(x_grad, rx), sx),
-array_ops.reshape(math_ops.reduce_sum(x_grad, ry), sy))
# Logical operations have no gradients.
ops.NotDifferentiable("Less")
ops.NotDifferentiable("LessEqual")
ops.NotDifferentiable("Greater")
ops.NotDifferentiable("GreaterEqual")
ops.NotDifferentiable("Equal")
ops.NotDifferentiable("ApproximateEqual")
ops.NotDifferentiable("NotEqual")
ops.NotDifferentiable("LogicalAnd")
ops.NotDifferentiable("LogicalOr")
ops.NotDifferentiable("LogicalNot")
@ops.RegisterGradient("Select")
def _SelectGrad(op, grad):
c = op.inputs[0]
x = op.inputs[1]
zeros = array_ops.zeros_like(x)
return (None, array_ops.where(c, grad, zeros), array_ops.where(
c, zeros, grad))
@ops.RegisterGradient("SelectV2")
def _SelectGradV2(op, grad):
c = op.inputs[0]
x = op.inputs[1]
y = op.inputs[2]
zeros = array_ops.zeros([], dtype=grad.dtype.base_dtype)
gx = array_ops.where_v2(c, grad, zeros)
x_shape = array_ops.shape(x)
output_shape = array_ops.shape(op.outputs[0])
# Reduce away broadcasted leading dims.
reduce_x, _ = gen_array_ops.broadcast_gradient_args(x_shape, output_shape)
gx = math_ops.reduce_sum(gx, keepdims=True, axis=reduce_x)
gx = array_ops.reshape(gx, x_shape)
gy = array_ops.where_v2(c, zeros, grad)
y_shape = array_ops.shape(y)
# Reduce away broadcasted leading dims.
reduce_y, _ = gen_array_ops.broadcast_gradient_args(y_shape, output_shape)
gy = math_ops.reduce_sum(gy, keepdims=True, axis=reduce_y)
gy = array_ops.reshape(gy, y_shape)
return (None, gx, gy)
def _MatMulGradAgainstFirstOnly(op, grad):
"""Gradient for MatMul, only for the first input."""
t_a = op.get_attr("transpose_a")
t_b = op.get_attr("transpose_b")
b = math_ops.conj(op.inputs[1])
if not t_a and not t_b:
grad_a = gen_math_ops.mat_mul(grad, b, transpose_b=True)
elif not t_a and t_b:
grad_a = gen_math_ops.mat_mul(grad, b)
elif t_a and not t_b:
grad_a = gen_math_ops.mat_mul(b, grad, transpose_b=True)
elif t_a and t_b:
grad_a = gen_math_ops.mat_mul(b, grad, transpose_a=True, transpose_b=True)
return grad_a, None
def _MatMulGradAgainstSecondOnly(op, grad):
"""Gradient for MatMul, only for the second input."""
t_a = op.get_attr("transpose_a")
t_b = op.get_attr("transpose_b")
a = math_ops.conj(op.inputs[0])
if not t_a and not t_b:
grad_b = gen_math_ops.mat_mul(a, grad, transpose_a=True)
elif not t_a and t_b:
grad_b = gen_math_ops.mat_mul(grad, a, transpose_a=True)
elif t_a and not t_b:
grad_b = gen_math_ops.mat_mul(a, grad)
elif t_a and t_b:
grad_b = gen_math_ops.mat_mul(grad, a, transpose_a=True, transpose_b=True)
return None, grad_b
@ops.RegisterGradient("MatMul")
def _MatMulGrad(op, grad):
"""Gradient for MatMul."""
try:
skip_input_indices = op.skip_input_indices
if skip_input_indices is not None:
if 1 in skip_input_indices:
return _MatMulGradAgainstFirstOnly(op, grad)
elif 0 in skip_input_indices:
return _MatMulGradAgainstSecondOnly(op, grad)
except AttributeError:
# No gradient skipping, so do the full gradient computation
pass
t_a = op.get_attr("transpose_a")
t_b = op.get_attr("transpose_b")
a = math_ops.conj(op.inputs[0])
b = math_ops.conj(op.inputs[1])
if not t_a and not t_b:
grad_a = gen_math_ops.mat_mul(grad, b, transpose_b=True)
grad_b = gen_math_ops.mat_mul(a, grad, transpose_a=True)
elif not t_a and t_b:
grad_a = gen_math_ops.mat_mul(grad, b)
grad_b = gen_math_ops.mat_mul(grad, a, transpose_a=True)
elif t_a and not t_b:
grad_a = gen_math_ops.mat_mul(b, grad, transpose_b=True)
grad_b = gen_math_ops.mat_mul(a, grad)
elif t_a and t_b:
grad_a = gen_math_ops.mat_mul(b, grad, transpose_a=True, transpose_b=True)
grad_b = gen_math_ops.mat_mul(grad, a, transpose_a=True, transpose_b=True)
return grad_a, grad_b
@ops.RegisterGradient("SparseMatMul")
def _SparseMatMulGrad(op, grad):
"""Gradient for SparseMatMul."""
t_a = op.get_attr("transpose_a")
t_b = op.get_attr("transpose_b")
is_sparse = {
op.inputs[0]: op.get_attr("a_is_sparse"),
op.inputs[1]: op.get_attr("b_is_sparse"),
# Use heuristic to figure out if grad might be sparse
grad: not context.executing_eagerly() and (grad.op.type == "ReluGrad")
}
def _SparseMatMul(t1, t2, out_dtype, transpose_a=False, transpose_b=False):
"""Helper function to create SparseMatMul op."""
assert t1 in is_sparse and t2 in is_sparse
t1_sparse = is_sparse[t1]
t2_sparse = is_sparse[t2]
if transpose_b:
t2 = array_ops.transpose(t2)
transpose_b = False
prod = math_ops.matmul(
t1,
t2,
transpose_a=transpose_a,
transpose_b=transpose_b,
a_is_sparse=t1_sparse,
b_is_sparse=t2_sparse)
if prod.dtype != out_dtype:
prod = math_ops.cast(prod, out_dtype)
return prod
dtype_a = op.inputs[0].dtype
dtype_b = op.inputs[1].dtype
if not t_a and not t_b:
return (_SparseMatMul(grad, op.inputs[1], dtype_a, transpose_b=True),
_SparseMatMul(op.inputs[0], grad, dtype_b, transpose_a=True))
elif not t_a and t_b:
return (_SparseMatMul(grad, op.inputs[1], dtype_a),
_SparseMatMul(grad, op.inputs[0], dtype_b, transpose_a=True))
elif t_a and not t_b:
return (_SparseMatMul(op.inputs[1], grad, dtype_a, transpose_b=True),
_SparseMatMul(op.inputs[0], grad, dtype_b))
elif t_a and t_b:
return (_SparseMatMul(
op.inputs[1], grad, dtype_a, transpose_a=True, transpose_b=True),
_SparseMatMul(
grad, op.inputs[0], dtype_b, transpose_a=True,
transpose_b=True))
@ops.RegisterGradient("Floor")
def _FloorGrad(_, unused_grad):
return [None]
@ops.RegisterGradient("Ceil")
def _CeilGrad(_, unused_grad):
return [None]
@ops.RegisterGradient("Round")
def _RoundGrad(_, unused_grad):
return [None]
@ops.RegisterGradient("Rint")
def _RintGrad(_, unused_grad):
# the gradient of Rint is zero
return [None]
@ops.RegisterGradient("BatchMatMul")
def _BatchMatMul(op, grad):
"""Returns the gradient of x and y given the gradient of x * y."""
x = op.inputs[0]
y = op.inputs[1]
adj_x = op.get_attr("adj_x")
adj_y = op.get_attr("adj_y")
if not adj_x:
if not adj_y:
grad_x = math_ops.matmul(grad, y, adjoint_a=False, adjoint_b=True)
grad_y = math_ops.matmul(x, grad, adjoint_a=True, adjoint_b=False)
else:
grad_x = math_ops.matmul(grad, y, adjoint_a=False, adjoint_b=False)
grad_y = math_ops.matmul(grad, x, adjoint_a=True, adjoint_b=False)
else:
if not adj_y:
grad_x = math_ops.matmul(y, grad, adjoint_a=False, adjoint_b=True)
grad_y = math_ops.matmul(x, grad, adjoint_a=False, adjoint_b=False)
else:
grad_x = math_ops.matmul(y, grad, adjoint_a=True, adjoint_b=True)
grad_y = math_ops.matmul(grad, x, adjoint_a=True, adjoint_b=True)
return grad_x, grad_y
@ops.RegisterGradient("BatchMatMulV2")
def _BatchMatMulV2(op, grad):
"""Returns the gradient of x and y given the gradient of x * y."""
x = op.inputs[0]
y = op.inputs[1]
adj_x = op.get_attr("adj_x")
adj_y = op.get_attr("adj_y")
if not adj_x:
if not adj_y:
grad_x = math_ops.matmul(grad, y, adjoint_a=False, adjoint_b=True)
grad_y = math_ops.matmul(x, grad, adjoint_a=True, adjoint_b=False)
else:
grad_x = math_ops.matmul(grad, y, adjoint_a=False, adjoint_b=False)
grad_y = math_ops.matmul(grad, x, adjoint_a=True, adjoint_b=False)
else:
if not adj_y:
grad_x = math_ops.matmul(y, grad, adjoint_a=False, adjoint_b=True)
grad_y = math_ops.matmul(x, grad, adjoint_a=False, adjoint_b=False)
else:
grad_x = math_ops.matmul(y, grad, adjoint_a=True, adjoint_b=True)
grad_y = math_ops.matmul(grad, x, adjoint_a=True, adjoint_b=True)
# Reduce along the broadcasted batch dimensions, if broadcasting is required.
shape_x_static = x.get_shape()
shape_y_static = y.get_shape()
if not (shape_x_static.is_fully_defined() and
shape_y_static.is_fully_defined() and
shape_x_static == shape_y_static):
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx[:-2], sy[:-2])
grad_x = array_ops.reshape(math_ops.reduce_sum(grad_x, rx), sx)
grad_y = array_ops.reshape(math_ops.reduce_sum(grad_y, ry), sy)
return grad_x, grad_y
ops.NotDifferentiable("Range")
ops.NotDifferentiable("LinSpace")
@ops.RegisterGradient("Complex")
def _ComplexGrad(op, grad):
"""Returns the real and imaginary components of 'grad', respectively."""
x = op.inputs[0]
y = op.inputs[1]
sx = array_ops.shape(x)
sy = array_ops.shape(y)
rx, ry = gen_array_ops.broadcast_gradient_args(sx, sy)
return (array_ops.reshape(math_ops.reduce_sum(math_ops.real(grad), rx), sx),
array_ops.reshape(math_ops.reduce_sum(math_ops.imag(grad), ry), sy))
@ops.RegisterGradient("Real")
def _RealGrad(_, grad):
"""Returns 'grad' as the real part and set the imaginary part 0."""
zero = constant_op.constant(0, dtype=grad.dtype)
return math_ops.complex(grad, zero)
@ops.RegisterGradient("Imag")
def _ImagGrad(_, grad):
"""Returns 'grad' as the imaginary part and set the real part 0."""
zero = constant_op.constant(0, dtype=grad.dtype)
return math_ops.complex(zero, grad)
@ops.RegisterGradient("Angle")
def _AngleGrad(op, grad):
"""Returns -grad / (Im(x) + iRe(x))"""
x = op.inputs[0]
with ops.control_dependencies([grad]):
re = math_ops.real(x)
im = math_ops.imag(x)
z = math_ops.reciprocal(math_ops.complex(im, re))
zero = constant_op.constant(0, dtype=grad.dtype)
complex_grad = math_ops.complex(grad, zero)
return -complex_grad * z
@ops.RegisterGradient("Conj")
def _ConjGrad(_, grad):
"""Returns the complex conjugate of grad."""
return math_ops.conj(grad)
@ops.RegisterGradient("ComplexAbs")
def _ComplexAbsGrad(op, grad):
"""Returns the gradient of ComplexAbs."""
return math_ops.div_no_nan(
math_ops.complex(
grad, array_ops.zeros_like(grad)) * op.inputs[0],
math_ops.complex(
op.outputs[0], array_ops.zeros_like(op.outputs[0])))
@ops.RegisterGradient("Cast")
def _CastGrad(op, grad):
t = [
dtypes.float16, dtypes.float32, dtypes.float64, dtypes.bfloat16,
dtypes.complex64, dtypes.complex128
]
src_type = op.inputs[0].dtype.base_dtype
dst_type = grad.dtype.base_dtype
if src_type in t and dst_type in t:
return math_ops.cast(grad, src_type)
else:
return None
@ops.RegisterGradient("Cross")
def _CrossGrad(op, grad):
u = op.inputs[0]
v = op.inputs[1]
return (math_ops.cross(v, grad), math_ops.cross(grad, u))
@ops.RegisterGradient("Cumsum")
def _CumsumGrad(op, grad):
axis = op.inputs[1]
exclusive = op.get_attr("exclusive")
reverse = op.get_attr("reverse")
return [
math_ops.cumsum(grad, axis, exclusive=exclusive, reverse=not reverse),
None
]
@ops.RegisterGradient("Cumprod")
def _CumprodGrad(op, grad):
x = op.inputs[0]
axis = op.inputs[1]
exclusive = op.get_attr("exclusive")
reverse = op.get_attr("reverse")
# TODO This fails when x contains 0 and should be fixed
prod = math_ops.cumprod(x, axis, exclusive=exclusive, reverse=reverse)
out = math_ops.cumsum(
prod * grad, axis, exclusive=exclusive, reverse=not reverse)
return [out / x, None]
@ops.RegisterGradient("NextAfter")
def _NextAfterGrad(op, grad):
"""Returns gradient of nextafter(x1, x2) with respect to x1 and x2."""
x1 = op.inputs[0]
x2 = op.inputs[1]
s_x1 = array_ops.shape(x1)
s_x2 = array_ops.shape(x2)
r_x1, r_x2 = gen_array_ops.broadcast_gradient_args(s_x1, s_x2)
with ops.control_dependencies([grad]):
partial_x1 = array_ops.ones(s_x1, dtype=x1.dtype)
partial_x2 = array_ops.zeros(s_x2, dtype=x2.dtype)
return (array_ops.reshape(
math_ops.reduce_sum(partial_x1 * grad, r_x1), s_x1),
array_ops.reshape(
math_ops.reduce_sum(partial_x2 * grad, r_x2), s_x2))
|
tensorflow-master
|
tensorflow/python/ops/math_grad.py
|
# Copyright 2018 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for initializers in init_ops."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.core.protobuf import config_pb2
from tensorflow.python.client import session
from tensorflow.python.eager import context
from tensorflow.python.framework import ops
from tensorflow.python.framework import tensor_shape as tensor_shape_lib
from tensorflow.python.framework import test_util
from tensorflow.python.ops import init_ops
from tensorflow.python.ops import variable_scope
from tensorflow.python.ops import variables
from tensorflow.python.platform import test
@test_util.run_all_in_graph_and_eager_modes
class InitializersTest(test.TestCase):
def _runner(self,
init,
shape,
target_mean=None,
target_std=None,
target_max=None,
target_min=None):
output = self.evaluate(init(shape))
self.assertEqual(output.shape, shape)
lim = 3e-2
if target_std is not None:
self.assertGreater(lim, abs(output.std() - target_std))
if target_mean is not None:
self.assertGreater(lim, abs(output.mean() - target_mean))
if target_max is not None:
self.assertGreater(lim, abs(output.max() - target_max))
if target_min is not None:
self.assertGreater(lim, abs(output.min() - target_min))
def test_uniform(self):
shape = (9, 6, 99)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
self._runner(
init_ops.RandomUniform(minval=-1, maxval=1, seed=124),
tensor_shape,
target_mean=0.,
target_max=1,
target_min=-1)
def test_normal(self):
shape = (8, 12, 99)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
self._runner(
init_ops.RandomNormal(mean=0, stddev=1, seed=153),
tensor_shape,
target_mean=0.,
target_std=1)
def test_truncated_normal(self):
shape = (12, 99, 7)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
self._runner(
init_ops.TruncatedNormal(mean=0, stddev=1, seed=126),
tensor_shape,
target_mean=0.,
target_max=2,
target_min=-2)
def test_constant(self):
shape = (5, 6, 4)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
self._runner(
init_ops.Constant(2),
tensor_shape,
target_mean=2,
target_max=2,
target_min=2)
def test_lecun_uniform(self):
shape = (5, 6, 4, 2)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
fan_in, _ = init_ops._compute_fans(tensor_shape)
std = np.sqrt(1. / fan_in)
self._runner(
init_ops.lecun_uniform(seed=123),
tensor_shape,
target_mean=0.,
target_std=std)
def test_glorot_uniform_initializer(self):
shape = (5, 6, 4, 2)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
fan_in, fan_out = init_ops._compute_fans(tensor_shape)
std = np.sqrt(2. / (fan_in + fan_out))
self._runner(
init_ops.glorot_uniform_initializer(seed=123),
tensor_shape,
target_mean=0.,
target_std=std)
def test_he_uniform(self):
shape = (5, 6, 4, 2)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
fan_in, _ = init_ops._compute_fans(tensor_shape)
std = np.sqrt(2. / fan_in)
self._runner(
init_ops.he_uniform(seed=123),
tensor_shape,
target_mean=0.,
target_std=std)
def test_lecun_normal(self):
shape = (5, 6, 4, 2)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
fan_in, _ = init_ops._compute_fans(tensor_shape)
std = np.sqrt(1. / fan_in)
self._runner(
init_ops.lecun_normal(seed=123),
tensor_shape,
target_mean=0.,
target_std=std)
def test_glorot_normal_initializer(self):
shape = (5, 6, 4, 2)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
fan_in, fan_out = init_ops._compute_fans(tensor_shape)
std = np.sqrt(2. / (fan_in + fan_out))
self._runner(
init_ops.glorot_normal_initializer(seed=123),
tensor_shape,
target_mean=0.,
target_std=std)
def test_he_normal(self):
shape = (5, 6, 4, 2)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
fan_in, _ = init_ops._compute_fans(tensor_shape)
std = np.sqrt(2. / fan_in)
self._runner(
init_ops.he_normal(seed=123),
tensor_shape,
target_mean=0.,
target_std=std)
def test_Orthogonal(self):
shape = (20, 20)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
self._runner(
init_ops.Orthogonal(seed=123), tensor_shape, target_mean=0.)
@test_util.run_gpu_only
def testVariablePlacementWithOrthogonalInitializer(self):
with ops.Graph().as_default() as g:
with ops.device('gpu:0'):
variable_scope.get_variable(
name='v', shape=[8, 2], initializer=init_ops.Orthogonal)
variable_scope.get_variable(
name='w', shape=[8, 2], initializer=init_ops.RandomNormal)
run_metadata = config_pb2.RunMetadata()
run_options = config_pb2.RunOptions(
trace_level=config_pb2.RunOptions.FULL_TRACE)
config = config_pb2.ConfigProto(
allow_soft_placement=False, log_device_placement=True)
# Note: allow_soft_placement=False will fail whenever we cannot satisfy
# the colocation constraints.
with session.Session(config=config, graph=g) as sess:
sess.run(
variables.global_variables_initializer(),
options=run_options,
run_metadata=run_metadata)
@test_util.run_gpu_only
def test_eager_orthogonal_gpu(self):
with context.eager_mode():
v = variable_scope.get_variable(
name='v', shape=[8, 2], initializer=init_ops.Orthogonal)
w = variable_scope.get_variable(
name='w', shape=[8, 2], initializer=init_ops.RandomNormal)
self.assertTrue('GPU' in v.handle.device)
self.assertTrue('GPU' in w.handle.device)
def test_Identity(self):
with self.cached_session():
shape = (3, 4, 5)
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
with self.assertRaises(ValueError):
self._runner(
init_ops.Identity(),
tensor_shape,
target_mean=1. / int(tensor_shape[0]),
target_max=1.)
shape = (3, 3)
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
self._runner(
init_ops.Identity(),
tensor_shape,
target_mean=1. / int(tensor_shape[0]),
target_max=1.)
def test_Zeros(self):
shape = (4, 5)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
self._runner(
init_ops.Zeros(), tensor_shape, target_mean=0., target_max=0.)
def test_Ones(self):
shape = (4, 5)
with self.cached_session():
for tensor_shape in [shape, tensor_shape_lib.TensorShape(shape)]:
self._runner(
init_ops.Ones(), tensor_shape, target_mean=1., target_max=1.)
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/ops/init_ops_test.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Gradients for operators defined in manip_ops.py."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import ops
from tensorflow.python.ops import manip_ops
@ops.RegisterGradient("Roll")
def _RollGrad(op, grad):
# The gradient is just the roll reversed
shift = op.inputs[1]
axis = op.inputs[2]
roll_grad = manip_ops.roll(grad, -shift, axis)
return roll_grad, None, None
|
tensorflow-master
|
tensorflow/python/ops/manip_grad.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Implements the graph generation for computation of gradients."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
# pylint: disable=unused-import
from tensorflow.python.eager import function
from tensorflow.python.eager.backprop import GradientTape
from tensorflow.python.ops.custom_gradient import custom_gradient
from tensorflow.python.ops.gradients_util import AggregationMethod
from tensorflow.python.ops.gradients_impl import gradients
from tensorflow.python.ops.gradients_impl import hessians
from tensorflow.python.ops.unconnected_gradients import UnconnectedGradients
# pylint: enable=unused-import
|
tensorflow-master
|
tensorflow/python/ops/gradients.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for tf.compat.v1.test.compute_gradient and tf.compute_gradient_error."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import test_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gradient_checker
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import nn_ops
import tensorflow.python.ops.nn_grad # pylint: disable=unused-import
from tensorflow.python.platform import test
from tensorflow.python.platform import tf_logging
@ops.RegisterGradient("BadGrad")
def _bad_grad(unused_op, grad):
"""A gradient that returns the wrong shape."""
return array_ops.transpose(grad)
@ops.RegisterGradient("NaNGrad")
def _nan_grad(unused_op, grad):
"""A gradient that returns NaN."""
return np.nan * grad
class GradientCheckerTest(test.TestCase):
@test_util.run_deprecated_v1
def testAddSimple(self):
np.random.seed(1) # Fix seed to avoid flakiness
with self.session(use_gpu=False):
# a test case for Add operation
size = (2, 3)
x1 = constant_op.constant(2.0, shape=size, name="x1")
x2 = constant_op.constant(3.0, shape=size, name="x2")
y = math_ops.add(x1, x2, name="y")
# checking gradients for x1
error = gradient_checker.compute_gradient_error(x1, size, y, size)
tf_logging.info("x1 error = %f", error)
assert error < 1e-4
@test_util.run_deprecated_v1
def testAddSimpleGPU(self):
np.random.seed(2) # Fix seed to avoid flakiness
with self.session(use_gpu=True):
# a test case for Add operation
size = (2, 3)
x1 = constant_op.constant(2.0, shape=size, name="x1")
x2 = constant_op.constant(3.0, shape=size, name="x2")
y = math_ops.add(x1, x2, name="y")
# checking gradients for x1
error = gradient_checker.compute_gradient_error(x1, size, y, size)
tf_logging.info("x1 error = %f", error)
assert error < 1e-4
@test_util.run_deprecated_v1
def testAddCustomized(self):
np.random.seed(3) # Fix seed to avoid flakiness
with self.cached_session():
# a test case for Add operation
size = (2, 3)
x1 = constant_op.constant(
2.0, shape=size, dtype=dtypes.float64, name="x1")
x2 = constant_op.constant(
3.0, shape=size, dtype=dtypes.float64, name="x2")
y = math_ops.add(x1, x2, name="y")
# checkint gradients for x2 using a special init_value and delta
x_init_value = np.asarray(np.arange(6, dtype=np.float64).reshape(2, 3))
error = gradient_checker.compute_gradient_error(
x2, size, y, size, x_init_value=x_init_value, delta=1e-2)
tf_logging.info("x2 error = %f", error)
assert error < 1e-10
@test_util.run_deprecated_v1
def testGather(self):
np.random.seed(4) # Fix seed to avoid flakiness
with self.cached_session():
p_shape = (4, 2)
p_size = 8
index_values = [1, 3]
y_shape = [2, 2]
params = constant_op.constant(
np.arange(p_size).astype(np.float), shape=p_shape, name="p")
indices = constant_op.constant(index_values, name="i")
y = array_ops.gather(params, indices, name="y")
error = gradient_checker.compute_gradient_error(params, p_shape, y,
y_shape)
tf_logging.info("gather error = %f", error)
assert error < 1e-4
@test_util.run_deprecated_v1
def testNestedGather(self):
np.random.seed(5) # Fix seed to avoid flakiness
with self.cached_session():
p_shape = (8, 2)
p_size = 16
index_values = [1, 3, 5, 6]
index_values2 = [0, 2]
y2_shape = [2, 2]
params = constant_op.constant(
np.arange(p_size).astype(np.float), shape=p_shape, name="p")
indices = constant_op.constant(index_values, name="i")
y = array_ops.gather(params, indices, name="y")
indices2 = constant_op.constant(index_values2, name="i2")
y2 = array_ops.gather(y, indices2, name="y2")
error = gradient_checker.compute_gradient_error(params, p_shape, y2,
y2_shape)
tf_logging.info("nested gather error = %f", error)
assert error < 1e-4
@test_util.run_deprecated_v1
def testComplexMul(self):
with self.cached_session():
size = ()
c = constant_op.constant(5 + 7j, dtype=dtypes.complex64)
x = constant_op.constant(11 - 13j, dtype=dtypes.complex64)
y = c * x
analytical, numerical = gradient_checker.compute_gradient(x, size, y,
size)
correct = np.array([[5, 7], [-7, 5]])
self.assertAllEqual(correct, analytical)
self.assertAllClose(correct, numerical, rtol=1e-4)
self.assertLess(
gradient_checker.compute_gradient_error(x, size, y, size), 2e-4)
@test_util.run_deprecated_v1
def testComplexConj(self):
with self.cached_session():
size = ()
x = constant_op.constant(11 - 13j, dtype=dtypes.complex64)
y = math_ops.conj(x)
analytical, numerical = gradient_checker.compute_gradient(x, size, y,
size)
correct = np.array([[1, 0], [0, -1]])
self.assertAllEqual(correct, analytical)
self.assertAllClose(correct, numerical, rtol=2e-5)
self.assertLess(
gradient_checker.compute_gradient_error(x, size, y, size), 2e-5)
@test_util.run_deprecated_v1
def testEmptySucceeds(self):
with self.cached_session():
x = array_ops.placeholder(dtypes.float32)
y = array_ops.identity(x)
for grad in gradient_checker.compute_gradient(x, (0, 3), y, (0, 3)):
self.assertEqual(grad.shape, (0, 0))
error = gradient_checker.compute_gradient_error(x, (0, 3), y, (0, 3))
self.assertEqual(error, 0)
def testEmptyFails(self):
with ops.Graph().as_default() as g:
with self.session(graph=g):
x = array_ops.placeholder(dtypes.float32)
with g.gradient_override_map({"Identity": "BadGrad"}):
y = array_ops.identity(x)
bad = r"Empty gradient has wrong shape: expected \(0, 3\), got \(3, 0\)"
with self.assertRaisesRegexp(ValueError, bad):
gradient_checker.compute_gradient(x, (0, 3), y, (0, 3))
with self.assertRaisesRegexp(ValueError, bad):
gradient_checker.compute_gradient_error(x, (0, 3), y, (0, 3))
def testNaNGradFails(self):
with ops.Graph().as_default() as g:
with self.session(graph=g):
x = array_ops.placeholder(dtypes.float32)
with g.gradient_override_map({"Identity": "NaNGrad"}):
y = array_ops.identity(x)
error = gradient_checker.compute_gradient_error(x, (), y, ())
# Typical test would assert error < max_err, so assert this test would
# raise AssertionError, since NaN is not < 1.0.
with self.assertRaisesRegexp(AssertionError, "False is not true"):
self.assertTrue(error < 1.0)
class MiniMNISTTest(test.TestCase):
# Gradient checker for MNIST.
def _BuildAndTestMiniMNIST(self, param_index, tag):
# Fix seed to avoid occasional flakiness
np.random.seed(6)
# Hyperparameters
batch = 3
inputs = 16
features = 32
classes = 10
# Define the parameters
inp_data = np.random.random_sample(inputs * batch)
hidden_weight_data = np.random.randn(inputs * features) / np.sqrt(inputs)
hidden_bias_data = np.random.random_sample(features)
sm_weight_data = np.random.randn(features * classes) / np.sqrt(features)
sm_bias_data = np.random.random_sample(classes)
# special care for labels since they need to be normalized per batch
label_data = np.random.random(batch * classes).reshape((batch, classes))
s = label_data.sum(axis=1)
label_data /= s[:, None]
with self.session(use_gpu=True):
# We treat the inputs as "parameters" here
inp = constant_op.constant(
inp_data.tolist(),
shape=[batch, inputs],
dtype=dtypes.float64,
name="inp")
hidden_weight = constant_op.constant(
hidden_weight_data.tolist(),
shape=[inputs, features],
dtype=dtypes.float64,
name="hidden_weight")
hidden_bias = constant_op.constant(
hidden_bias_data.tolist(),
shape=[features],
dtype=dtypes.float64,
name="hidden_bias")
softmax_weight = constant_op.constant(
sm_weight_data.tolist(),
shape=[features, classes],
dtype=dtypes.float64,
name="softmax_weight")
softmax_bias = constant_op.constant(
sm_bias_data.tolist(),
shape=[classes],
dtype=dtypes.float64,
name="softmax_bias")
# List all the parameter so that we can test them one at a time
all_params = [
inp, hidden_weight, hidden_bias, softmax_weight, softmax_bias
]
param_sizes = [
[batch, inputs], # inp
[inputs, features], # hidden_weight,
[features], # hidden_bias
[features, classes], # softmax_weight,
[classes]
] # softmax_bias
# Now, Building MNIST
features = nn_ops.relu(
nn_ops.xw_plus_b(inp, hidden_weight, hidden_bias), name="features")
logits = nn_ops.xw_plus_b(
features, softmax_weight, softmax_bias, name="logits")
labels = constant_op.constant(
label_data.tolist(),
shape=[batch, classes],
dtype=dtypes.float64,
name="labels")
cost = nn_ops.softmax_cross_entropy_with_logits(
labels=labels, logits=logits, name="cost")
# Test the gradients.
err = gradient_checker.compute_gradient_error(
all_params[param_index],
param_sizes[param_index],
cost, [batch],
delta=1e-5)
tf_logging.info("Mini MNIST: %s gradient error = %g", tag, err)
return err
@test_util.run_deprecated_v1
def testInputGradient(self):
self.assertLess(self._BuildAndTestMiniMNIST(0, "input"), 1e-8)
@test_util.run_deprecated_v1
def testHiddenWeightGradient(self):
self.assertLess(self._BuildAndTestMiniMNIST(1, "hidden_weight"), 1e-8)
@test_util.run_deprecated_v1
def testHiddenBiasGradient(self):
self.assertLess(self._BuildAndTestMiniMNIST(2, "hidden_bias"), 1e-8)
@test_util.run_deprecated_v1
def testSoftmaxWeightGradient(self):
self.assertLess(self._BuildAndTestMiniMNIST(3, "softmax_weight"), 1e-8)
@test_util.run_deprecated_v1
def testSoftmaxBiasGradient(self):
self.assertLess(self._BuildAndTestMiniMNIST(4, "softmax_bias"), 1e-8)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/ops/gradient_checker_test.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for batch_norm related functionality in tensorflow.ops.nn."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from six.moves import xrange # pylint: disable=redefined-builtin
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import test_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gen_nn_ops
from tensorflow.python.ops import gradient_checker
from tensorflow.python.ops import gradients_impl
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import nn_impl
import tensorflow.python.ops.nn_grad # pylint: disable=unused-import
from tensorflow.python.platform import test
class BatchNormalizationTest(test.TestCase):
def _npBatchNorm(self, x, m, v, beta, gamma, epsilon,
scale_after_normalization, shift_after_normalization):
y = (x - m) / np.sqrt(v + epsilon)
y = y * gamma if scale_after_normalization else y
return y + beta if shift_after_normalization else y
def _opsBatchNorm(self, x, m, v, beta, gamma, epsilon,
scale_after_normalization, shift_after_normalization):
y = (x - m) * math_ops.rsqrt(v + epsilon)
if scale_after_normalization:
y = gamma * y
return y + beta if shift_after_normalization else y
def _tfBatchNormV1(self, x, m, v, beta, gamma, epsilon,
scale_after_normalization):
"""Original implementation."""
test_util.set_producer_version(ops.get_default_graph(), 8)
return gen_nn_ops._batch_norm_with_global_normalization(
x, m, v, beta, gamma, epsilon, scale_after_normalization)
def _tfBatchNormV1BW(self, x, m, v, beta, gamma, epsilon,
scale_after_normalization):
"""Re-implementation of the original kernel for backward compatibility."""
return nn_impl.batch_norm_with_global_normalization(
x, m, v, beta, gamma, epsilon, scale_after_normalization)
def _tfBatchNormV2(self, x, m, v, beta, gamma, epsilon,
scale_after_normalization, shift_after_normalization):
"""New implementation."""
return nn_impl.batch_normalization(x, m, v, beta if
shift_after_normalization else None,
gamma if scale_after_normalization else
None, epsilon)
@test_util.run_deprecated_v1
def testBatchNorm(self):
x_shape = [3, 5, 4, 2]
param_shape = [2]
x_val = np.random.random_sample(x_shape).astype(np.float32)
m_val = np.random.random_sample(param_shape).astype(np.float32)
v_val = np.random.random_sample(param_shape).astype(np.float32)
beta_val = np.random.random_sample(param_shape).astype(np.float32)
gamma_val = np.random.random_sample(param_shape).astype(np.float32)
for use_gpu in [True, False]:
with self.cached_session(use_gpu=use_gpu) as sess:
x = constant_op.constant(x_val, name="x")
m = constant_op.constant(m_val, name="m")
v = constant_op.constant(v_val, name="v")
beta = constant_op.constant(beta_val, name="beta")
gamma = constant_op.constant(gamma_val, name="gamma")
epsilon = 0.001
for scale_after_normalization in [True, False]:
for shift_after_normalization in [True, False]:
bn2 = self._tfBatchNormV2(x, m, v, beta, gamma, epsilon,
scale_after_normalization,
shift_after_normalization)
bn1bw = self._tfBatchNormV1BW(x, m, v, beta, gamma, epsilon,
scale_after_normalization)
bn1 = self._tfBatchNormV1(x, m, v, beta, gamma, epsilon,
scale_after_normalization)
on = self._opsBatchNorm(x, m, v, beta, gamma, epsilon,
scale_after_normalization,
shift_after_normalization)
np_bn = self._npBatchNorm(x_val, m_val, v_val, beta_val, gamma_val,
epsilon, scale_after_normalization,
shift_after_normalization)
tf_bn_v2, tf_bn_v1bw, tf_bn_v1, ops_bn = sess.run(
[bn2, bn1bw, bn1, on])
self.assertAllClose(np_bn, ops_bn, atol=0.00001)
self.assertAllClose(np_bn, tf_bn_v2, atol=0.00001)
self.assertAllClose(tf_bn_v2, ops_bn, atol=0.00001)
# shift_after_normalization=False is not supported in v1.
if shift_after_normalization:
self.assertAllClose(np_bn, tf_bn_v1bw, atol=0.00001)
self.assertAllClose(np_bn, tf_bn_v1, atol=0.00001)
self.assertAllClose(tf_bn_v1, ops_bn, atol=0.00001)
self.assertAllClose(tf_bn_v1bw, ops_bn, atol=0.00001)
def _testBatchNormGradient(self,
param_index,
tag,
scale_after_normalization,
shift_after_normalization,
version,
err_tolerance=1e-11):
x_shape = [3, 5, 4, 5]
param_shape = [5]
np.random.seed(1) # Make it reproducible.
x_val = np.random.random_sample(x_shape).astype(np.float64)
m_val = np.random.random_sample(param_shape).astype(np.float64)
v_val = np.random.random_sample(param_shape).astype(np.float64)
beta_val = np.random.random_sample(param_shape).astype(np.float64)
gamma_val = np.random.random_sample(param_shape).astype(np.float64)
with self.cached_session():
x = constant_op.constant(x_val, name="x")
m = constant_op.constant(m_val, name="m")
v = constant_op.constant(v_val, name="v")
beta = constant_op.constant(beta_val, name="beta")
gamma = constant_op.constant(gamma_val, name="gamma")
epsilon = 0.001
if version == 1:
output = self._tfBatchNormV1(x, m, v, beta, gamma, epsilon,
scale_after_normalization)
elif version == 2:
output = self._tfBatchNormV2(x, m, v, beta, gamma, epsilon,
scale_after_normalization,
shift_after_normalization)
else:
print("Invalid version", version)
raise ValueError()
all_params = [x, m, v, beta, gamma]
all_shapes = [x_shape, param_shape, param_shape, param_shape, param_shape]
err = gradient_checker.compute_gradient_error(all_params[param_index],
all_shapes[param_index],
output, x_shape)
print("Batch normalization v%d %s gradient %s scale and %s shift err = " %
(version, tag, "with" if scale_after_normalization else "without",
"with" if shift_after_normalization else "without"), err)
self.assertLess(err, err_tolerance)
def _testBatchNormGradientInAllNeedConfigs(self,
param_index,
tag,
err_tolerance=1e-11):
for scale_after_normalization in [True, False]:
for shift_after_normalization in [True, False]:
# shift_after_normalization=False is not supported in version 1.
for v in ([1, 2] if shift_after_normalization else [2]):
self._testBatchNormGradient(param_index, tag,
scale_after_normalization,
shift_after_normalization, v,
err_tolerance)
@test_util.run_deprecated_v1
def testBatchNormInputGradient(self):
self._testBatchNormGradientInAllNeedConfigs(0, "x")
@test_util.run_deprecated_v1
def testBatchNormMeanGradient(self):
self._testBatchNormGradientInAllNeedConfigs(1, "mean")
@test_util.run_deprecated_v1
def testBatchNormVarianceGradient(self):
self._testBatchNormGradientInAllNeedConfigs(
2, "variance", err_tolerance=1e-03)
@test_util.run_deprecated_v1
def testBatchNormBetaGradient(self):
# Since beta does not exist when scale_after_normalization=False, we only
# test for scale_after_normalization=True.
for scale_after_normalization in [True, False]:
for v in [1, 2]:
self._testBatchNormGradient(3, "beta", scale_after_normalization, True,
v)
@test_util.run_deprecated_v1
def testBatchNormGammaGradient(self):
# If scale_after_normalization is False, backprop for gamma in v1
# will be 0. In version 2 of the API, if scale_after_normalization is False,
# gamma is not used at all, and the gradient is None, which displeases the
# gradient checker.
for scale_after_normalization in [True, False]:
self._testBatchNormGradient(4, "gamma", scale_after_normalization, True,
1)
for shift_after_normalization in [True, False]:
self._testBatchNormGradient(4, "gamma", True, shift_after_normalization,
2)
@test_util.run_deprecated_v1
def testBatchNormGradImpl(self):
x_shape = [7, 5, 4, 6]
param_shape = [6]
np.random.seed(1) # Make it reproducible.
x_val = np.random.random_sample(x_shape).astype(np.float32)
m_val = np.random.random_sample(param_shape).astype(np.float32)
v_val = np.random.random_sample(param_shape).astype(np.float32)
beta_val = np.random.random_sample(param_shape).astype(np.float32)
gamma_val = np.random.random_sample(param_shape).astype(np.float32)
backprop_val = np.random.random_sample(x_shape).astype(np.float32)
for use_gpu in [False, True]:
with self.cached_session(use_gpu=use_gpu) as sess:
x = constant_op.constant(x_val, name="x")
m = constant_op.constant(m_val, name="m")
v = constant_op.constant(v_val, name="v")
beta = constant_op.constant(beta_val, name="beta")
gamma = constant_op.constant(gamma_val, name="gamma")
backprop = constant_op.constant(backprop_val, name="backprop")
epsilon = 0.001
for scale_after_normalization in [True, False]:
# _batch_norm_with_global_normalization_grad is deprecated in v9
test_util.set_producer_version(ops.get_default_graph(), 8)
grad = gen_nn_ops.batch_norm_with_global_normalization_grad(
x, m, v, gamma, backprop, epsilon, scale_after_normalization)
dx, dm, dv, db, dg = grad
self.assertEqual(grad.dx, dx)
self.assertEqual(grad.dm, dm)
self.assertEqual(grad.dv, dv)
self.assertEqual(grad.db, db)
self.assertEqual(grad.dg, dg)
on = self._opsBatchNorm(x, m, v, beta, gamma, epsilon,
scale_after_normalization, True)
odx, odm, odv, odb, odg = gradients_impl.gradients(
[on], [x, m, v, beta, gamma], [backprop])
if scale_after_normalization:
all_grads = self.evaluate(
[dx, dm, dv, db, dg, odx, odm, odv, odb, odg])
to_check = ["dx", "dm", "dv", "db", "dg"]
else:
all_grads = self.evaluate([dx, dm, dv, db, odx, odm, odv, odb])
to_check = ["dx", "dm", "dv", "db"]
for i, _ in enumerate(to_check):
self.assertAllClose(
all_grads[i + len(to_check)], all_grads[i], atol=0.000001)
@test_util.run_deprecated_v1
def testBatchNormKeepDims(self):
"""Test for tf.nn.moments(..., keep_dims=True / False).
Make sure that parameters with shape (1, 1, 1, depth) yield the same
result as parameters with shape (depth)
"""
x_shape = (3, 5, 4, 2)
param_shape = (2)
keep_dims_param_shape = (1, 1, 1, 2)
x_val = np.random.random_sample(x_shape).astype(np.float32)
m_val = np.random.random_sample(param_shape).astype(np.float32)
v_val = np.random.random_sample(param_shape).astype(np.float32)
beta_val = np.random.random_sample(param_shape).astype(np.float32)
gamma_val = np.random.random_sample(param_shape).astype(np.float32)
for use_gpu in [True, False]:
with self.cached_session(use_gpu=use_gpu) as sess:
x = constant_op.constant(x_val, name="x")
m = constant_op.constant(m_val, name="m")
v = constant_op.constant(v_val, name="v")
beta = constant_op.constant(beta_val, name="beta")
gamma = constant_op.constant(gamma_val, name="gamma")
keep_dims_m = array_ops.reshape(
m, keep_dims_param_shape, name="keep_dims_m")
keep_dims_v = array_ops.reshape(
v, keep_dims_param_shape, name="keep_dims_v")
keep_dims_beta = array_ops.reshape(
beta, keep_dims_param_shape, name="keep_dims_beta")
keep_dims_gamma = array_ops.reshape(
gamma, keep_dims_param_shape, name="keep_dims_gamma")
epsilon = 0.001
for scale_after_normalization in [True, False]:
for shift_after_normalization in [True, False]:
bn = self._tfBatchNormV2(x, m, v, beta, gamma, epsilon,
scale_after_normalization,
shift_after_normalization)
keep_dims_bn = self._tfBatchNormV2(x, keep_dims_m, keep_dims_v,
keep_dims_beta, keep_dims_gamma,
epsilon,
scale_after_normalization,
shift_after_normalization)
tf_batch_norm, keep_dims_tf_batch_norm = sess.run(
[bn, keep_dims_bn])
self.assertEquals(x_shape, tf_batch_norm.shape)
self.assertEquals(x_shape, keep_dims_tf_batch_norm.shape)
self.assertAllClose(
tf_batch_norm, keep_dims_tf_batch_norm, atol=0.000001)
def _testBatchNormArbitraryShapes(self, x_shape, param_shape, atol=0.0001,
dtype=dtypes.float32,
param_dtype=dtypes.float32):
numpy_dtype = dtype.as_numpy_dtype
numpy_param_dtype = param_dtype.as_numpy_dtype
x_val = np.random.random_sample(x_shape).astype(numpy_dtype)
m_val = np.random.random_sample(param_shape).astype(numpy_param_dtype)
v_val = np.random.random_sample(param_shape).astype(numpy_param_dtype)
beta_val = np.random.random_sample(param_shape).astype(numpy_param_dtype)
gamma_val = np.random.random_sample(param_shape).astype(numpy_param_dtype)
for use_gpu in [True, False]:
with self.cached_session(use_gpu=use_gpu) as sess:
x = constant_op.constant(x_val, name="x")
m = constant_op.constant(m_val, name="m")
v = constant_op.constant(v_val, name="v")
beta = constant_op.constant(beta_val, name="beta")
gamma = constant_op.constant(gamma_val, name="gamma")
epsilon = 0.001
for scale_after_normalization in [True, False]:
for shift_after_normalization in [True, False]:
bn = self._tfBatchNormV2(x, m, v, beta, gamma, epsilon,
scale_after_normalization,
shift_after_normalization)
np_batch_norm = self._npBatchNorm(x_val, m_val, v_val, beta_val,
gamma_val, epsilon,
scale_after_normalization,
shift_after_normalization)
[tf_batch_norm] = self.evaluate([bn])
self.assertEquals(x_shape, np_batch_norm.shape)
self.assertEquals(x_shape, tf_batch_norm.shape)
self.assertAllClose(np_batch_norm, tf_batch_norm, atol=atol)
def testBatchNormArbitraryShapes(self):
"""Test for a variety of shapes and moments.
Batch normalization is expected to work regardless of the position and
dimensionality of the 'depth' axis/axes.
"""
self._testBatchNormArbitraryShapes((3, 3), (1, 3))
self._testBatchNormArbitraryShapes((3, 3), (3, 1))
self._testBatchNormArbitraryShapes((3, 2, 4, 5), (1, 2, 1, 1))
self._testBatchNormArbitraryShapes(
(2, 3, 2, 4, 5), (1, 1, 1, 4, 5), atol=0.005)
def testBatchNormMixedPrecision(self):
self._testBatchNormArbitraryShapes((3, 3), (1, 3), dtype=dtypes.float16,
param_dtype=dtypes.float32, atol=0.001)
class SufficientStatisticsTest(test.TestCase):
def _npSuffStats(self, x, axes, shift, keep_dims):
axis = tuple(axes)
if shift is not None:
m_ss = np.sum(x - shift, axis=axis, keepdims=keep_dims)
v_ss = np.sum((x - shift) * (x - shift), axis=axis, keepdims=keep_dims)
else:
m_ss = np.sum(x, axis=axis, keepdims=keep_dims)
v_ss = np.sum(x * x, axis=axis, keepdims=keep_dims)
count = 1.0
for d in xrange(x.ndim):
if d in set(axes):
count *= x.shape[d]
if not keep_dims:
shift = np.squeeze(shift, axis=axis)
return count, m_ss, v_ss, shift
def _opSuffStats(self, x, axes, shift, keep_dims):
return nn_impl.sufficient_statistics(x, axes, shift, keep_dims)
def _testSuffStats(self, x_shape, axes, shift, keep_dims, has_shape):
x_val = np.random.random_sample(x_shape).astype(np.float32)
np_c, np_m, np_v, np_s = self._npSuffStats(x_val, axes, shift, keep_dims)
for use_gpu in [True, False]:
with self.cached_session(use_gpu=use_gpu) as sess:
if has_shape:
x = constant_op.constant(x_val, name="x")
x.set_shape(x_shape)
op_c, op_m, op_v, op_s = self._opSuffStats(x, axes, shift, keep_dims)
if shift:
tf_c, tf_m, tf_v, tf_s = self.evaluate([op_c, op_m, op_v, op_s])
else:
tf_c, tf_m, tf_v = self.evaluate([op_c, op_m, op_v])
else:
x = array_ops.placeholder(
dtype=dtypes.float32, shape=[None] * len(x_shape), name="x")
op_c, op_m, op_v, op_s = self._opSuffStats(x, axes, shift, keep_dims)
if shift:
tf_c, tf_m, tf_v, tf_s = sess.run([op_c, op_m, op_v, op_s],
feed_dict={x: x_val})
else:
tf_c, tf_m, tf_v = sess.run([op_c, op_m, op_v],
feed_dict={x: x_val})
self.assertAllClose(np_c, tf_c, atol=0.000001)
self.assertAllClose(np_m, tf_m, atol=0.000001)
self.assertAllClose(np_v, tf_v, atol=0.000001)
if shift:
self.assertAllClose(np_s, tf_s, atol=0.000001)
@test_util.run_deprecated_v1
def testSuffStats(self):
for has_shape in [True, False]:
for keep_dims in [True, False]:
for shift in [None, 1.0]:
self._testSuffStats([2, 3], [1], shift, keep_dims, has_shape)
self._testSuffStats([2, 3], [0], shift, keep_dims, has_shape)
self._testSuffStats([1, 2, 3], [0, 2], shift, keep_dims, has_shape)
class NormalizeMomentsTest(test.TestCase):
def _npNormalizeMoments(self, counts, mean_ss, variance_ss, shift):
mean = mean_ss / counts
variance = variance_ss / counts - mean * mean
if shift is not None:
mean += shift
return mean, variance
def _opNormalizeMoments(self, counts, mean_ss, variance_ss, shift):
return nn_impl.normalize_moments(counts, mean_ss, variance_ss, shift)
def _testNormalizeMoments(self, shape, shift):
counts = np.ones([1]).astype(np.float32)
mean_ss = np.random.random_sample(shape).astype(np.float32)
variance_ss = np.random.random_sample(shape).astype(np.float32)
variance_ss *= variance_ss
if shift:
shift_v = np.random.random_sample(shape).astype(np.float32)
else:
shift_v = None
npm, npv = self._npNormalizeMoments(counts, mean_ss, variance_ss, shift_v)
for use_gpu in [True, False]:
with self.cached_session(use_gpu=use_gpu) as sess:
tf_counts = constant_op.constant(counts, name="counts")
tf_mean_ss = constant_op.constant(mean_ss, name="mean_ss")
tf_variance_ss = constant_op.constant(variance_ss, name="variance_ss")
if shift:
tf_shift_v = constant_op.constant(shift_v, name="shift")
else:
tf_shift_v = None
opm, opv = self._opNormalizeMoments(tf_counts, tf_mean_ss,
tf_variance_ss, tf_shift_v)
tfm, tfv = self.evaluate([opm, opv])
self.assertAllClose(npm, tfm, atol=0.000001)
self.assertAllClose(npv, tfv, atol=0.000001)
def testNormalizeMoments(self):
for shift in [None, 4.0]:
self._testNormalizeMoments([3], shift)
self._testNormalizeMoments([2, 3], shift)
class MomentsTest(test.TestCase):
def _unweighted_moments(self, x, axes, keep_dims=False, extra_out_grads=None):
# Method to compute moments of `x` wrt `axes`.
#
# This is exposed so WeightedMomentsTest can inherit the tests and
# assertions from MomentsTest; the extra_out_grads argument allows
# its inherited gradient tests to assert gradients against the
# weights as well as the input values.
return nn_impl.moments(x, axes, keep_dims=keep_dims)
def RunMomentTestWithDynamicShape(self, shape, axes, keep_dims, dtype):
with self.cached_session():
# shape = [batch, width, height, depth]
assert len(shape) == 4
x_numpy = np.random.normal(size=shape).astype(np.float32)
x = array_ops.placeholder(dtype, shape=[None] * len(shape))
mean, var = self._unweighted_moments(x, axes, keep_dims=keep_dims)
num_elements = np.prod([shape[i] for i in axes])
ax = tuple(axes)
expected_mean = np.sum(x_numpy, axis=ax,
keepdims=keep_dims) / num_elements
expected_mean_squared = np.multiply(expected_mean, expected_mean)
expected_x_squared = np.sum(np.multiply(x_numpy, x_numpy),
axis=ax,
keepdims=keep_dims) / num_elements
expected_variance = expected_x_squared - expected_mean_squared
# Check that the moments are correct.
self.assertAllCloseAccordingToType(
expected_mean, mean.eval(feed_dict={x: x_numpy}))
self.assertAllCloseAccordingToType(
expected_variance, var.eval(feed_dict={x: x_numpy}))
def RunMomentTest(self, shape, axes, keep_dims, dtype):
with self.cached_session():
# shape = [batch, width, height, depth]
assert len(shape) == 4
x_numpy = np.random.normal(size=shape).astype(np.float32)
x = math_ops.cast(constant_op.constant(x_numpy), dtype=dtype)
# Compute the expected values at high precision since the method
# is prone to catastrophic cancellation:
x_numpy = x_numpy.astype(np.float128)
mean, var = self._unweighted_moments(x, axes, keep_dims=keep_dims)
num_elements = np.prod([shape[i] for i in axes])
ax = tuple(axes)
expected_mean = np.sum(x_numpy, axis=ax,
keepdims=keep_dims) / num_elements
expected_mean_squared = np.multiply(expected_mean, expected_mean)
expected_x_squared = np.sum(np.multiply(x_numpy, x_numpy),
axis=ax,
keepdims=keep_dims) / num_elements
expected_variance = expected_x_squared - expected_mean_squared
# Check that the moments are correct.
self.assertAllCloseAccordingToType(expected_mean, self.evaluate(mean))
self.assertAllCloseAccordingToType(expected_variance, self.evaluate(var))
@test_util.run_deprecated_v1
def testBasic(self):
for keep_dims in [False, True]:
for dtype in [dtypes.float32, dtypes.float16]:
self.RunMomentTest(
shape=[2, 3, 5, 4], axes=[0], keep_dims=keep_dims, dtype=dtype)
self.RunMomentTestWithDynamicShape(
shape=[2, 3, 5, 4], axes=[0], keep_dims=keep_dims, dtype=dtype)
@test_util.run_deprecated_v1
def testGlobalNormalization(self):
for keep_dims in [False, True]:
for dtype in [dtypes.float32, dtypes.float16]:
self.RunMomentTest(
shape=[2, 3, 5, 4],
axes=[0, 1, 2],
keep_dims=keep_dims,
dtype=dtype)
self.RunMomentTestWithDynamicShape(
shape=[2, 3, 5, 4],
axes=[0, 1, 2],
keep_dims=keep_dims,
dtype=dtype)
@test_util.run_deprecated_v1
def testAxes(self):
for keep_dims in [False, True]:
for dtype in [dtypes.float32, dtypes.float16]:
self.RunMomentTest(
shape=[2, 3, 5, 4],
axes=[1, 2, 3],
keep_dims=keep_dims,
dtype=dtype)
self.RunMomentTestWithDynamicShape(
shape=[2, 3, 5, 4],
axes=[1, 2, 3],
keep_dims=keep_dims,
dtype=dtype)
def _testGlobalGradient(self, from_y="mean"):
with self.cached_session():
x_shape = [3, 5, 4, 2]
x_val = np.random.random_sample(x_shape).astype(np.float64)
x = constant_op.constant(x_val)
x.set_shape(x_shape)
axes = [0, 1, 2]
y_shape = [2] # Depth of x
inputs_to_compute_gradients_for = [x]
out_mean, out_var = self._unweighted_moments(
x, axes, extra_out_grads=inputs_to_compute_gradients_for)
if from_y == "mean":
y = out_mean
elif from_y == "var":
y = out_var
for (i, v) in enumerate(inputs_to_compute_gradients_for):
err = gradient_checker.compute_gradient_error(v,
v.get_shape().as_list(),
y, y_shape)
print("Moments %s gradient err vs input %d = %g" % (from_y, i, err))
self.assertLess(err, 1e-11)
@test_util.run_deprecated_v1
def testMeanGlobalGradient(self):
self._testGlobalGradient(from_y="mean")
@test_util.run_deprecated_v1
def testVarGlobalGradient(self):
self._testGlobalGradient(from_y="var")
class WeightedMomentsTest(MomentsTest):
"""Tests for nn.weighted_moments.
Note that this test inherits from MomentsTest, inheriting all its
test methods!
It modifies MomentsTest in two ways:
a) By overriding _unweighted_moments, all the codepaths in
MomentsTest are executed, but with calls to tf.nn.moments()
replaced by calls to tf.nn.weighted_moments() with a constant
weight of 1.
b) By overriding RunMomentTest and RunMomentTestWithDynamicShape,
this test adds multiple additional calls to
RunWeightedMomentsTest() to exercise correctness with
non-constant weights and varying broadcasting situations. (It
also continues to call MomentsTest.Run(Weighted)?MomentsTest as
well.)
"""
def _unweighted_moments(self, x, axes, keep_dims=False, extra_out_grads=None):
weights = constant_op.constant(1, dtype=x.dtype)
if extra_out_grads is not None:
# We want to assert gradients WRT weights as well as X!
extra_out_grads.append(weights)
return nn_impl.weighted_moments(x, axes, weights, keep_dims=keep_dims)
def RunMomentTest(self, shape, axes, keep_dims, dtype, dynshapes=False):
if not dynshapes:
super(WeightedMomentsTest, self).RunMomentTest(shape, axes, keep_dims,
dtype)
else:
super(WeightedMomentsTest, self).RunMomentTestWithDynamicShape(shape,
axes,
keep_dims,
dtype)
# 1:1 weights and inputs
self.RunWeightedMomentTest(shape, shape, axes, keep_dims, dtype)
# Various broadcasting combinations
for idx in range(len(shape)):
# try broadcasting weights in all positions
weight_shape = [1] * len(shape)
weight_shape[idx] = shape[idx]
self.RunWeightedMomentTest(shape, weight_shape, axes, keep_dims, dtype)
# Also try broadcasting with a suffix of length n
weight_shape = shape[-(idx + 1):]
self.RunWeightedMomentTest(
shape, weight_shape, axes, keep_dims, dtype, dynshapes=dynshapes)
def RunMomentTestWithDynamicShape(self, shape, axes, keep_dims, dtype):
self.RunMomentTest(shape, axes, keep_dims, dtype, dynshapes=True)
def RunWeightedMomentTest(self,
shape,
weights_shape,
axes,
keep_dims,
dtype,
dynshapes=False):
with self.cached_session() as s:
x_numpy = np.random.normal(size=shape).astype(np.float32)
weights_numpy = np.absolute( # weights must be positive
np.random.normal(
size=weights_shape, loc=1.0).astype(np.float32))
# Expand the numpy version to higher precision
x_numpy = x_numpy.astype(np.float128)
weights_numpy = weights_numpy.astype(np.float128)
x_shape = [None] * len(shape) if dynshapes else shape
weights_shape = ([None] * len(weights_shape) if dynshapes else
weights_shape)
x = array_ops.placeholder(dtype, shape=x_shape)
weights = array_ops.placeholder(dtype, shape=weights_shape)
mean, var = nn_impl.weighted_moments(
x, axes, weights, keep_dims=keep_dims)
ax = tuple(axes)
def _np_weighted_sum(v):
return np.sum(weights_numpy * v, axis=ax, keepdims=keep_dims)
weight_sum = _np_weighted_sum(np.ones_like(x_numpy))
expected_mean = _np_weighted_sum(x_numpy) / weight_sum
expected_mean_squared = np.multiply(expected_mean, expected_mean)
expected_x_squared = (_np_weighted_sum(np.multiply(x_numpy, x_numpy)) /
weight_sum)
expected_variance = expected_x_squared - expected_mean_squared
mean_v, var_v = s.run([mean, var],
feed_dict={x: x_numpy,
weights: weights_numpy})
self.assertAllCloseAccordingToType(expected_mean, mean_v)
self.assertAllCloseAccordingToType(expected_variance, var_v)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/ops/nn_batchnorm_test.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Parsing Ops."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import re
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import sparse_tensor
from tensorflow.python.framework import tensor_shape
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import gen_parsing_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import sparse_ops
# go/tf-wildcard-import
# pylint: disable=wildcard-import,undefined-variable
from tensorflow.python.ops.gen_parsing_ops import *
# pylint: enable=wildcard-import,undefined-variable
from tensorflow.python.platform import tf_logging
from tensorflow.python.util import deprecation
from tensorflow.python.util.tf_export import tf_export
ops.NotDifferentiable("DecodeRaw")
ops.NotDifferentiable("DecodePaddedRaw")
ops.NotDifferentiable("ParseTensor")
ops.NotDifferentiable("SerializeTensor")
ops.NotDifferentiable("StringToNumber")
@tf_export("io.VarLenFeature", v1=["VarLenFeature", "io.VarLenFeature"])
class VarLenFeature(collections.namedtuple("VarLenFeature", ["dtype"])):
"""Configuration for parsing a variable-length input feature.
Fields:
dtype: Data type of input.
"""
pass
@tf_export("io.SparseFeature", v1=["io.SparseFeature", "SparseFeature"])
class SparseFeature(
collections.namedtuple(
"SparseFeature",
["index_key", "value_key", "dtype", "size", "already_sorted"])):
"""Configuration for parsing a sparse input feature from an `Example`.
Note, preferably use `VarLenFeature` (possibly in combination with a
`SequenceExample`) in order to parse out `SparseTensor`s instead of
`SparseFeature` due to its simplicity.
Closely mimicking the `SparseTensor` that will be obtained by parsing an
`Example` with a `SparseFeature` config, a `SparseFeature` contains a
* `value_key`: The name of key for a `Feature` in the `Example` whose parsed
`Tensor` will be the resulting `SparseTensor.values`.
* `index_key`: A list of names - one for each dimension in the resulting
`SparseTensor` whose `indices[i][dim]` indicating the position of
the `i`-th value in the `dim` dimension will be equal to the `i`-th value in
the Feature with key named `index_key[dim]` in the `Example`.
* `size`: A list of ints for the resulting `SparseTensor.dense_shape`.
For example, we can represent the following 2D `SparseTensor`
```python
SparseTensor(indices=[[3, 1], [20, 0]],
values=[0.5, -1.0]
dense_shape=[100, 3])
```
with an `Example` input proto
```python
features {
feature { key: "val" value { float_list { value: [ 0.5, -1.0 ] } } }
feature { key: "ix0" value { int64_list { value: [ 3, 20 ] } } }
feature { key: "ix1" value { int64_list { value: [ 1, 0 ] } } }
}
```
and `SparseFeature` config with 2 `index_key`s
```python
SparseFeature(index_key=["ix0", "ix1"],
value_key="val",
dtype=tf.float32,
size=[100, 3])
```
Fields:
index_key: A single string name or a list of string names of index features.
For each key the underlying feature's type must be `int64` and its length
must always match that of the `value_key` feature.
To represent `SparseTensor`s with a `dense_shape` of `rank` higher than 1
a list of length `rank` should be used.
value_key: Name of value feature. The underlying feature's type must
be `dtype` and its length must always match that of all the `index_key`s'
features.
dtype: Data type of the `value_key` feature.
size: A Python int or list thereof specifying the dense shape. Should be a
list if and only if `index_key` is a list. In that case the list must be
equal to the length of `index_key`. Each for each entry `i` all values in
the `index_key`[i] feature must be in `[0, size[i])`.
already_sorted: A Python boolean to specify whether the values in
`value_key` are already sorted by their index position. If so skip
sorting. False by default (optional).
"""
def __new__(cls, index_key, value_key, dtype, size, already_sorted=False):
return super(SparseFeature, cls).__new__(
cls, index_key, value_key, dtype, size, already_sorted)
@tf_export("io.FixedLenFeature", v1=["io.FixedLenFeature", "FixedLenFeature"])
class FixedLenFeature(collections.namedtuple(
"FixedLenFeature", ["shape", "dtype", "default_value"])):
"""Configuration for parsing a fixed-length input feature.
To treat sparse input as dense, provide a `default_value`; otherwise,
the parse functions will fail on any examples missing this feature.
Fields:
shape: Shape of input data.
dtype: Data type of input.
default_value: Value to be used if an example is missing this feature. It
must be compatible with `dtype` and of the specified `shape`.
"""
def __new__(cls, shape, dtype, default_value=None):
return super(FixedLenFeature, cls).__new__(
cls, shape, dtype, default_value)
@tf_export("io.FixedLenSequenceFeature",
v1=["io.FixedLenSequenceFeature", "FixedLenSequenceFeature"])
class FixedLenSequenceFeature(collections.namedtuple(
"FixedLenSequenceFeature",
["shape", "dtype", "allow_missing", "default_value"])):
"""Configuration for parsing a variable-length input feature into a `Tensor`.
The resulting `Tensor` of parsing a single `SequenceExample` or `Example` has
a static `shape` of `[None] + shape` and the specified `dtype`.
The resulting `Tensor` of parsing a `batch_size` many `Example`s has
a static `shape` of `[batch_size, None] + shape` and the specified `dtype`.
The entries in the `batch` from different `Examples` will be padded with
`default_value` to the maximum length present in the `batch`.
To treat a sparse input as dense, provide `allow_missing=True`; otherwise,
the parse functions will fail on any examples missing this feature.
Fields:
shape: Shape of input data for dimension 2 and higher. First dimension is
of variable length `None`.
dtype: Data type of input.
allow_missing: Whether to allow this feature to be missing from a feature
list item. Is available only for parsing `SequenceExample` not for
parsing `Examples`.
default_value: Scalar value to be used to pad multiple `Example`s to their
maximum length. Irrelevant for parsing a single `Example` or
`SequenceExample`. Defaults to "" for dtype string and 0 otherwise
(optional).
"""
def __new__(cls, shape, dtype, allow_missing=False, default_value=None):
return super(FixedLenSequenceFeature, cls).__new__(
cls, shape, dtype, allow_missing, default_value)
def _features_to_raw_params(features, types):
"""Split feature tuples into raw params used by `gen_parsing_ops`.
Args:
features: A `dict` mapping feature keys to objects of a type in `types`.
types: Type of features to allow, among `FixedLenFeature`, `VarLenFeature`,
`SparseFeature`, and `FixedLenSequenceFeature`.
Returns:
Tuple of `sparse_keys`, `sparse_types`, `dense_keys`, `dense_types`,
`dense_defaults`, `dense_shapes`.
Raises:
ValueError: if `features` contains an item not in `types`, or an invalid
feature.
"""
sparse_keys = []
sparse_types = []
dense_keys = []
dense_types = []
# When the graph is built twice, multiple dense_defaults in a normal dict
# could come out in different orders. This will fail the _e2e_test which
# expects exactly the same graph.
# OrderedDict which preserves the order can solve the problem.
dense_defaults = collections.OrderedDict()
dense_shapes = []
if features:
# NOTE: We iterate over sorted keys to keep things deterministic.
for key in sorted(features.keys()):
feature = features[key]
if isinstance(feature, VarLenFeature):
if VarLenFeature not in types:
raise ValueError("Unsupported VarLenFeature %s." % (feature,))
if not feature.dtype:
raise ValueError("Missing type for feature %s." % key)
sparse_keys.append(key)
sparse_types.append(feature.dtype)
elif isinstance(feature, SparseFeature):
if SparseFeature not in types:
raise ValueError("Unsupported SparseFeature %s." % (feature,))
if not feature.index_key:
raise ValueError(
"Missing index_key for SparseFeature %s." % (feature,))
if not feature.value_key:
raise ValueError(
"Missing value_key for SparseFeature %s." % (feature,))
if not feature.dtype:
raise ValueError("Missing type for feature %s." % key)
index_keys = feature.index_key
if isinstance(index_keys, str):
index_keys = [index_keys]
elif len(index_keys) > 1:
tf_logging.warning("SparseFeature is a complicated feature config "
"and should only be used after careful "
"consideration of VarLenFeature.")
for index_key in sorted(index_keys):
if index_key in sparse_keys:
dtype = sparse_types[sparse_keys.index(index_key)]
if dtype != dtypes.int64:
raise ValueError("Conflicting type %s vs int64 for feature %s." %
(dtype, index_key))
else:
sparse_keys.append(index_key)
sparse_types.append(dtypes.int64)
if feature.value_key in sparse_keys:
dtype = sparse_types[sparse_keys.index(feature.value_key)]
if dtype != feature.dtype:
raise ValueError("Conflicting type %s vs %s for feature %s." % (
dtype, feature.dtype, feature.value_key))
else:
sparse_keys.append(feature.value_key)
sparse_types.append(feature.dtype)
elif isinstance(feature, FixedLenFeature):
if FixedLenFeature not in types:
raise ValueError("Unsupported FixedLenFeature %s." % (feature,))
if not feature.dtype:
raise ValueError("Missing type for feature %s." % key)
if feature.shape is None:
raise ValueError("Missing shape for feature %s." % key)
feature_tensor_shape = tensor_shape.as_shape(feature.shape)
if (feature.shape and feature_tensor_shape.ndims and
feature_tensor_shape.dims[0].value is None):
raise ValueError("First dimension of shape for feature %s unknown. "
"Consider using FixedLenSequenceFeature." % key)
if (feature.shape is not None and
not feature_tensor_shape.is_fully_defined()):
raise ValueError("All dimensions of shape for feature %s need to be "
"known but received %s." % (key, str(feature.shape)))
dense_keys.append(key)
dense_shapes.append(feature.shape)
dense_types.append(feature.dtype)
if feature.default_value is not None:
dense_defaults[key] = feature.default_value
elif isinstance(feature, FixedLenSequenceFeature):
if FixedLenSequenceFeature not in types:
raise ValueError("Unsupported FixedLenSequenceFeature %s." % (
feature,))
if not feature.dtype:
raise ValueError("Missing type for feature %s." % key)
if feature.shape is None:
raise ValueError("Missing shape for feature %s." % key)
dense_keys.append(key)
dense_shapes.append(feature.shape)
dense_types.append(feature.dtype)
if feature.allow_missing:
dense_defaults[key] = None
if feature.default_value is not None:
dense_defaults[key] = feature.default_value
else:
raise ValueError("Invalid feature %s:%s." % (key, feature))
return (
sparse_keys, sparse_types, dense_keys, dense_types, dense_defaults,
dense_shapes)
def _construct_sparse_tensors_for_sparse_features(features, tensor_dict):
"""Merges SparseTensors of indices and values of SparseFeatures.
Constructs new dict based on `tensor_dict`. For `SparseFeatures` in the values
of `features` expects their `index_key`s and `index_value`s to be present in
`tensor_dict` mapping to `SparseTensor`s. Constructs a single `SparseTensor`
from them, and adds it to the result with the key from `features`.
Copies other keys and values from `tensor_dict` with keys present in
`features`.
Args:
features: A `dict` mapping feature keys to `SparseFeature` values.
Values of other types will be ignored.
tensor_dict: A `dict` mapping feature keys to `Tensor` and `SparseTensor`
values. Expected to contain keys of the `SparseFeature`s' `index_key`s and
`value_key`s and mapping them to `SparseTensor`s.
Returns:
A `dict` mapping feature keys to `Tensor` and `SparseTensor` values. Similar
to `tensor_dict` except each `SparseFeature`s in `features` results in a
single `SparseTensor`.
"""
tensor_dict = dict(tensor_dict) # Do not modify argument passed in.
# Construct SparseTensors for SparseFeatures.
for key in sorted(features.keys()):
feature = features[key]
if isinstance(feature, SparseFeature):
if isinstance(feature.index_key, str):
sp_ids = tensor_dict[feature.index_key]
else:
sp_ids = [tensor_dict[index_key] for index_key in feature.index_key]
sp_values = tensor_dict[feature.value_key]
tensor_dict[key] = sparse_ops.sparse_merge(
sp_ids,
sp_values,
vocab_size=feature.size,
already_sorted=feature.already_sorted)
# Remove tensors from dictionary that were only used to construct
# SparseTensors for SparseFeature.
for key in set(tensor_dict) - set(features):
del tensor_dict[key]
return tensor_dict
def _prepend_none_dimension(features):
if features:
modified_features = dict(features) # Create a copy to modify
for key, feature in features.items():
if isinstance(feature, FixedLenSequenceFeature):
if not feature.allow_missing:
raise ValueError("Unsupported: FixedLenSequenceFeature requires "
"allow_missing to be True.")
modified_features[key] = FixedLenSequenceFeature(
[None] + list(feature.shape),
feature.dtype,
feature.allow_missing,
feature.default_value)
return modified_features
else:
return features
@tf_export(v1=["io.parse_example", "parse_example"])
def parse_example(serialized, features, name=None, example_names=None):
# pylint: disable=line-too-long
"""Parses `Example` protos into a `dict` of tensors.
Parses a number of serialized [`Example`](https://www.tensorflow.org/code/tensorflow/core/example/example.proto)
protos given in `serialized`. We refer to `serialized` as a batch with
`batch_size` many entries of individual `Example` protos.
`example_names` may contain descriptive names for the corresponding serialized
protos. These may be useful for debugging purposes, but they have no effect on
the output. If not `None`, `example_names` must be the same length as
`serialized`.
This op parses serialized examples into a dictionary mapping keys to `Tensor`
and `SparseTensor` objects. `features` is a dict from keys to `VarLenFeature`,
`SparseFeature`, and `FixedLenFeature` objects. Each `VarLenFeature`
and `SparseFeature` is mapped to a `SparseTensor`, and each
`FixedLenFeature` is mapped to a `Tensor`.
Each `VarLenFeature` maps to a `SparseTensor` of the specified type
representing a ragged matrix. Its indices are `[batch, index]` where `batch`
identifies the example in `serialized`, and `index` is the value's index in
the list of values associated with that feature and example.
Each `SparseFeature` maps to a `SparseTensor` of the specified type
representing a Tensor of `dense_shape` `[batch_size] + SparseFeature.size`.
Its `values` come from the feature in the examples with key `value_key`.
A `values[i]` comes from a position `k` in the feature of an example at batch
entry `batch`. This positional information is recorded in `indices[i]` as
`[batch, index_0, index_1, ...]` where `index_j` is the `k-th` value of
the feature in the example at with key `SparseFeature.index_key[j]`.
In other words, we split the indices (except the first index indicating the
batch entry) of a `SparseTensor` by dimension into different features of the
`Example`. Due to its complexity a `VarLenFeature` should be preferred over a
`SparseFeature` whenever possible.
Each `FixedLenFeature` `df` maps to a `Tensor` of the specified type (or
`tf.float32` if not specified) and shape `(serialized.size(),) + df.shape`.
`FixedLenFeature` entries with a `default_value` are optional. With no default
value, we will fail if that `Feature` is missing from any example in
`serialized`.
Each `FixedLenSequenceFeature` `df` maps to a `Tensor` of the specified type
(or `tf.float32` if not specified) and shape
`(serialized.size(), None) + df.shape`.
All examples in `serialized` will be padded with `default_value` along the
second dimension.
Examples:
For example, if one expects a `tf.float32` `VarLenFeature` `ft` and three
serialized `Example`s are provided:
```
serialized = [
features
{ feature { key: "ft" value { float_list { value: [1.0, 2.0] } } } },
features
{ feature []},
features
{ feature { key: "ft" value { float_list { value: [3.0] } } }
]
```
then the output will look like:
```python
{"ft": SparseTensor(indices=[[0, 0], [0, 1], [2, 0]],
values=[1.0, 2.0, 3.0],
dense_shape=(3, 2)) }
```
If instead a `FixedLenSequenceFeature` with `default_value = -1.0` and
`shape=[]` is used then the output will look like:
```python
{"ft": [[1.0, 2.0], [3.0, -1.0]]}
```
Given two `Example` input protos in `serialized`:
```
[
features {
feature { key: "kw" value { bytes_list { value: [ "knit", "big" ] } } }
feature { key: "gps" value { float_list { value: [] } } }
},
features {
feature { key: "kw" value { bytes_list { value: [ "emmy" ] } } }
feature { key: "dank" value { int64_list { value: [ 42 ] } } }
feature { key: "gps" value { } }
}
]
```
And arguments
```
example_names: ["input0", "input1"],
features: {
"kw": VarLenFeature(tf.string),
"dank": VarLenFeature(tf.int64),
"gps": VarLenFeature(tf.float32),
}
```
Then the output is a dictionary:
```python
{
"kw": SparseTensor(
indices=[[0, 0], [0, 1], [1, 0]],
values=["knit", "big", "emmy"]
dense_shape=[2, 2]),
"dank": SparseTensor(
indices=[[1, 0]],
values=[42],
dense_shape=[2, 1]),
"gps": SparseTensor(
indices=[],
values=[],
dense_shape=[2, 0]),
}
```
For dense results in two serialized `Example`s:
```
[
features {
feature { key: "age" value { int64_list { value: [ 0 ] } } }
feature { key: "gender" value { bytes_list { value: [ "f" ] } } }
},
features {
feature { key: "age" value { int64_list { value: [] } } }
feature { key: "gender" value { bytes_list { value: [ "f" ] } } }
}
]
```
We can use arguments:
```
example_names: ["input0", "input1"],
features: {
"age": FixedLenFeature([], dtype=tf.int64, default_value=-1),
"gender": FixedLenFeature([], dtype=tf.string),
}
```
And the expected output is:
```python
{
"age": [[0], [-1]],
"gender": [["f"], ["f"]],
}
```
An alternative to `VarLenFeature` to obtain a `SparseTensor` is
`SparseFeature`. For example, given two `Example` input protos in
`serialized`:
```
[
features {
feature { key: "val" value { float_list { value: [ 0.5, -1.0 ] } } }
feature { key: "ix" value { int64_list { value: [ 3, 20 ] } } }
},
features {
feature { key: "val" value { float_list { value: [ 0.0 ] } } }
feature { key: "ix" value { int64_list { value: [ 42 ] } } }
}
]
```
And arguments
```
example_names: ["input0", "input1"],
features: {
"sparse": SparseFeature(
index_key="ix", value_key="val", dtype=tf.float32, size=100),
}
```
Then the output is a dictionary:
```python
{
"sparse": SparseTensor(
indices=[[0, 3], [0, 20], [1, 42]],
values=[0.5, -1.0, 0.0]
dense_shape=[2, 100]),
}
```
Args:
serialized: A vector (1-D Tensor) of strings, a batch of binary
serialized `Example` protos.
features: A `dict` mapping feature keys to `FixedLenFeature`,
`VarLenFeature`, and `SparseFeature` values.
name: A name for this operation (optional).
example_names: A vector (1-D Tensor) of strings (optional), the names of
the serialized protos in the batch.
Returns:
A `dict` mapping feature keys to `Tensor` and `SparseTensor` values.
Raises:
ValueError: if any feature is invalid.
"""
return parse_example_v2(serialized, features, example_names, name)
@tf_export("io.parse_example", v1=[])
def parse_example_v2(serialized, features, example_names=None, name=None):
# pylint: disable=line-too-long
"""Parses `Example` protos into a `dict` of tensors.
Parses a number of serialized [`Example`](https://www.tensorflow.org/code/tensorflow/core/example/example.proto)
protos given in `serialized`. We refer to `serialized` as a batch with
`batch_size` many entries of individual `Example` protos.
`example_names` may contain descriptive names for the corresponding serialized
protos. These may be useful for debugging purposes, but they have no effect on
the output. If not `None`, `example_names` must be the same length as
`serialized`.
This op parses serialized examples into a dictionary mapping keys to `Tensor`
and `SparseTensor` objects. `features` is a dict from keys to `VarLenFeature`,
`SparseFeature`, and `FixedLenFeature` objects. Each `VarLenFeature`
and `SparseFeature` is mapped to a `SparseTensor`, and each
`FixedLenFeature` is mapped to a `Tensor`.
Each `VarLenFeature` maps to a `SparseTensor` of the specified type
representing a ragged matrix. Its indices are `[batch, index]` where `batch`
identifies the example in `serialized`, and `index` is the value's index in
the list of values associated with that feature and example.
Each `SparseFeature` maps to a `SparseTensor` of the specified type
representing a Tensor of `dense_shape` `[batch_size] + SparseFeature.size`.
Its `values` come from the feature in the examples with key `value_key`.
A `values[i]` comes from a position `k` in the feature of an example at batch
entry `batch`. This positional information is recorded in `indices[i]` as
`[batch, index_0, index_1, ...]` where `index_j` is the `k-th` value of
the feature in the example at with key `SparseFeature.index_key[j]`.
In other words, we split the indices (except the first index indicating the
batch entry) of a `SparseTensor` by dimension into different features of the
`Example`. Due to its complexity a `VarLenFeature` should be preferred over a
`SparseFeature` whenever possible.
Each `FixedLenFeature` `df` maps to a `Tensor` of the specified type (or
`tf.float32` if not specified) and shape `(serialized.size(),) + df.shape`.
`FixedLenFeature` entries with a `default_value` are optional. With no default
value, we will fail if that `Feature` is missing from any example in
`serialized`.
Each `FixedLenSequenceFeature` `df` maps to a `Tensor` of the specified type
(or `tf.float32` if not specified) and shape
`(serialized.size(), None) + df.shape`.
All examples in `serialized` will be padded with `default_value` along the
second dimension.
Examples:
For example, if one expects a `tf.float32` `VarLenFeature` `ft` and three
serialized `Example`s are provided:
```
serialized = [
features
{ feature { key: "ft" value { float_list { value: [1.0, 2.0] } } } },
features
{ feature []},
features
{ feature { key: "ft" value { float_list { value: [3.0] } } }
]
```
then the output will look like:
```python
{"ft": SparseTensor(indices=[[0, 0], [0, 1], [2, 0]],
values=[1.0, 2.0, 3.0],
dense_shape=(3, 2)) }
```
If instead a `FixedLenSequenceFeature` with `default_value = -1.0` and
`shape=[]` is used then the output will look like:
```python
{"ft": [[1.0, 2.0], [3.0, -1.0]]}
```
Given two `Example` input protos in `serialized`:
```
[
features {
feature { key: "kw" value { bytes_list { value: [ "knit", "big" ] } } }
feature { key: "gps" value { float_list { value: [] } } }
},
features {
feature { key: "kw" value { bytes_list { value: [ "emmy" ] } } }
feature { key: "dank" value { int64_list { value: [ 42 ] } } }
feature { key: "gps" value { } }
}
]
```
And arguments
```
example_names: ["input0", "input1"],
features: {
"kw": VarLenFeature(tf.string),
"dank": VarLenFeature(tf.int64),
"gps": VarLenFeature(tf.float32),
}
```
Then the output is a dictionary:
```python
{
"kw": SparseTensor(
indices=[[0, 0], [0, 1], [1, 0]],
values=["knit", "big", "emmy"]
dense_shape=[2, 2]),
"dank": SparseTensor(
indices=[[1, 0]],
values=[42],
dense_shape=[2, 1]),
"gps": SparseTensor(
indices=[],
values=[],
dense_shape=[2, 0]),
}
```
For dense results in two serialized `Example`s:
```
[
features {
feature { key: "age" value { int64_list { value: [ 0 ] } } }
feature { key: "gender" value { bytes_list { value: [ "f" ] } } }
},
features {
feature { key: "age" value { int64_list { value: [] } } }
feature { key: "gender" value { bytes_list { value: [ "f" ] } } }
}
]
```
We can use arguments:
```
example_names: ["input0", "input1"],
features: {
"age": FixedLenFeature([], dtype=tf.int64, default_value=-1),
"gender": FixedLenFeature([], dtype=tf.string),
}
```
And the expected output is:
```python
{
"age": [[0], [-1]],
"gender": [["f"], ["f"]],
}
```
An alternative to `VarLenFeature` to obtain a `SparseTensor` is
`SparseFeature`. For example, given two `Example` input protos in
`serialized`:
```
[
features {
feature { key: "val" value { float_list { value: [ 0.5, -1.0 ] } } }
feature { key: "ix" value { int64_list { value: [ 3, 20 ] } } }
},
features {
feature { key: "val" value { float_list { value: [ 0.0 ] } } }
feature { key: "ix" value { int64_list { value: [ 42 ] } } }
}
]
```
And arguments
```
example_names: ["input0", "input1"],
features: {
"sparse": SparseFeature(
index_key="ix", value_key="val", dtype=tf.float32, size=100),
}
```
Then the output is a dictionary:
```python
{
"sparse": SparseTensor(
indices=[[0, 3], [0, 20], [1, 42]],
values=[0.5, -1.0, 0.0]
dense_shape=[2, 100]),
}
```
Args:
serialized: A vector (1-D Tensor) of strings, a batch of binary
serialized `Example` protos.
features: A `dict` mapping feature keys to `FixedLenFeature`,
`VarLenFeature`, and `SparseFeature` values.
example_names: A vector (1-D Tensor) of strings (optional), the names of
the serialized protos in the batch.
name: A name for this operation (optional).
Returns:
A `dict` mapping feature keys to `Tensor` and `SparseTensor` values.
Raises:
ValueError: if any feature is invalid.
"""
if not features:
raise ValueError("Missing: features was %s." % features)
features = _prepend_none_dimension(features)
(sparse_keys, sparse_types, dense_keys, dense_types, dense_defaults,
dense_shapes) = _features_to_raw_params(
features,
[VarLenFeature, SparseFeature, FixedLenFeature, FixedLenSequenceFeature])
outputs = _parse_example_raw(
serialized, example_names, sparse_keys, sparse_types, dense_keys,
dense_types, dense_defaults, dense_shapes, name)
return _construct_sparse_tensors_for_sparse_features(features, outputs)
def _parse_example_raw(serialized,
names=None,
sparse_keys=None,
sparse_types=None,
dense_keys=None,
dense_types=None,
dense_defaults=None,
dense_shapes=None,
name=None):
"""Parses `Example` protos.
Args:
serialized: A vector (1-D Tensor) of strings, a batch of binary
serialized `Example` protos.
names: A vector (1-D Tensor) of strings (optional), the names of
the serialized protos.
sparse_keys: A list of string keys in the examples' features.
The results for these keys will be returned as `SparseTensor` objects.
sparse_types: A list of `DTypes` of the same length as `sparse_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`),
and `tf.string` (`BytesList`) are supported.
dense_keys: A list of string keys in the examples' features.
The results for these keys will be returned as `Tensor`s
dense_types: A list of DTypes of the same length as `dense_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`),
and `tf.string` (`BytesList`) are supported.
dense_defaults: A dict mapping string keys to `Tensor`s.
The keys of the dict must match the dense_keys of the feature.
dense_shapes: A list of tuples with the same length as `dense_keys`.
The shape of the data for each dense feature referenced by `dense_keys`.
Required for any input tensors identified by `dense_keys`. Must be
either fully defined, or may contain an unknown first dimension.
An unknown first dimension means the feature is treated as having
a variable number of blocks, and the output shape along this dimension
is considered unknown at graph build time. Padding is applied for
minibatch elements smaller than the maximum number of blocks for the
given feature along this dimension.
name: A name for this operation (optional).
Returns:
A `dict` mapping keys to `Tensor`s and `SparseTensor`s.
"""
with ops.name_scope(name, "ParseExample", [serialized, names]):
(names, dense_defaults_vec, sparse_keys, sparse_types,
dense_keys, dense_shapes, _) = _process_raw_parameters(
names, dense_defaults, sparse_keys, sparse_types, dense_keys,
dense_types, dense_shapes)
outputs = gen_parsing_ops.parse_example(
serialized=serialized,
names=names,
dense_defaults=dense_defaults_vec,
sparse_keys=sparse_keys,
sparse_types=sparse_types,
dense_keys=dense_keys,
dense_shapes=dense_shapes,
name=name)
(sparse_indices, sparse_values, sparse_shapes, dense_values) = outputs
sparse_tensors = [
sparse_tensor.SparseTensor(ix, val, shape) for (ix, val, shape)
in zip(sparse_indices, sparse_values, sparse_shapes)]
return dict(zip(sparse_keys + dense_keys, sparse_tensors + dense_values))
def _process_raw_parameters(names, dense_defaults, sparse_keys, sparse_types,
dense_keys, dense_types, dense_shapes):
"""Process raw parameters to params used by `gen_parsing_ops`.
Args:
names: A vector (1-D Tensor) of strings (optional), the names of
the serialized protos.
dense_defaults: A dict mapping string keys to `Tensor`s.
The keys of the dict must match the dense_keys of the feature.
sparse_keys: A list of string keys in the examples' features.
The results for these keys will be returned as `SparseTensor` objects.
sparse_types: A list of `DTypes` of the same length as `sparse_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`),
and `tf.string` (`BytesList`) are supported.
dense_keys: A list of string keys in the examples' features.
The results for these keys will be returned as `Tensor`s
dense_types: A list of DTypes of the same length as `dense_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`),
and `tf.string` (`BytesList`) are supported.
dense_shapes: A list of tuples with the same length as `dense_keys`.
The shape of the data for each dense feature referenced by `dense_keys`.
Required for any input tensors identified by `dense_keys`. Must be
either fully defined, or may contain an unknown first dimension.
An unknown first dimension means the feature is treated as having
a variable number of blocks, and the output shape along this dimension
is considered unknown at graph build time. Padding is applied for
minibatch elements smaller than the maximum number of blocks for the
given feature along this dimension.
Returns:
Tuple of `names`, `dense_defaults_vec`, `sparse_keys`, `sparse_types`,
`dense_keys`, `dense_shapes`.
Raises:
ValueError: If sparse and dense key sets intersect, or input lengths do not
match up.
"""
names = [] if names is None else names
dense_defaults = collections.OrderedDict(
) if dense_defaults is None else dense_defaults
sparse_keys = [] if sparse_keys is None else sparse_keys
sparse_types = [] if sparse_types is None else sparse_types
dense_keys = [] if dense_keys is None else dense_keys
dense_types = [] if dense_types is None else dense_types
dense_shapes = ([[]] * len(dense_keys)
if dense_shapes is None else dense_shapes)
num_dense = len(dense_keys)
num_sparse = len(sparse_keys)
if len(dense_shapes) != num_dense:
raise ValueError("len(dense_shapes) != len(dense_keys): %d vs. %d" %
(len(dense_shapes), num_dense))
if len(dense_types) != num_dense:
raise ValueError("len(dense_types) != len(num_dense): %d vs. %d" %
(len(dense_types), num_dense))
if len(sparse_types) != num_sparse:
raise ValueError("len(sparse_types) != len(sparse_keys): %d vs. %d" %
(len(sparse_types), num_sparse))
if num_dense + num_sparse == 0:
raise ValueError("Must provide at least one sparse key or dense key")
if not set(dense_keys).isdisjoint(set(sparse_keys)):
raise ValueError(
"Dense and sparse keys must not intersect; intersection: %s" %
set(dense_keys).intersection(set(sparse_keys)))
# Convert dense_shapes to TensorShape object.
dense_shapes = [tensor_shape.as_shape(shape) for shape in dense_shapes]
dense_defaults_vec = []
for i, key in enumerate(dense_keys):
default_value = dense_defaults.get(key)
dense_shape = dense_shapes[i]
if (dense_shape.ndims is not None and dense_shape.ndims > 0 and
dense_shape.dims[0].value is None):
# Variable stride dense shape, the default value should be a
# scalar padding value
if default_value is None:
default_value = ops.convert_to_tensor(
"" if dense_types[i] == dtypes.string else 0, dtype=dense_types[i])
else:
# Reshape to a scalar to ensure user gets an error if they
# provide a tensor that's not intended to be a padding value
# (0 or 2+ elements).
key_name = "padding_" + re.sub("[^A-Za-z0-9_.\\-/]", "_", key)
default_value = ops.convert_to_tensor(
default_value, dtype=dense_types[i], name=key_name)
default_value = array_ops.reshape(default_value, [])
else:
if default_value is None:
default_value = constant_op.constant([], dtype=dense_types[i])
elif not isinstance(default_value, ops.Tensor):
key_name = "key_" + re.sub("[^A-Za-z0-9_.\\-/]", "_", key)
default_value = ops.convert_to_tensor(
default_value, dtype=dense_types[i], name=key_name)
default_value = array_ops.reshape(default_value, dense_shape)
dense_defaults_vec.append(default_value)
# Finally, convert dense_shapes to TensorShapeProto
dense_shapes_as_proto = [shape.as_proto() for shape in dense_shapes]
return (names, dense_defaults_vec, sparse_keys, sparse_types, dense_keys,
dense_shapes_as_proto, dense_shapes)
@tf_export(v1=["io.parse_single_example", "parse_single_example"])
def parse_single_example(serialized, features, name=None, example_names=None):
"""Parses a single `Example` proto.
Similar to `parse_example`, except:
For dense tensors, the returned `Tensor` is identical to the output of
`parse_example`, except there is no batch dimension, the output shape is the
same as the shape given in `dense_shape`.
For `SparseTensor`s, the first (batch) column of the indices matrix is removed
(the indices matrix is a column vector), the values vector is unchanged, and
the first (`batch_size`) entry of the shape vector is removed (it is now a
single element vector).
One might see performance advantages by batching `Example` protos with
`parse_example` instead of using this function directly.
Args:
serialized: A scalar string Tensor, a single serialized Example.
See `_parse_single_example_raw` documentation for more details.
features: A `dict` mapping feature keys to `FixedLenFeature` or
`VarLenFeature` values.
name: A name for this operation (optional).
example_names: (Optional) A scalar string Tensor, the associated name.
See `_parse_single_example_raw` documentation for more details.
Returns:
A `dict` mapping feature keys to `Tensor` and `SparseTensor` values.
Raises:
ValueError: if any feature is invalid.
"""
return parse_single_example_v2_unoptimized(
serialized, features, example_names, name
)
# TODO(b/70890287): Combine the implementation of this op and
# `parse_single_example_v2()` after 1/10/2018.
@tf_export("io.parse_single_example", v1=[])
def parse_single_example_v2_unoptimized(
serialized, features, example_names=None, name=None
):
"""Parses a single `Example` proto.
Similar to `parse_example`, except:
For dense tensors, the returned `Tensor` is identical to the output of
`parse_example`, except there is no batch dimension, the output shape is the
same as the shape given in `dense_shape`.
For `SparseTensor`s, the first (batch) column of the indices matrix is removed
(the indices matrix is a column vector), the values vector is unchanged, and
the first (`batch_size`) entry of the shape vector is removed (it is now a
single element vector).
One might see performance advantages by batching `Example` protos with
`parse_example` instead of using this function directly.
Args:
serialized: A scalar string Tensor, a single serialized Example.
See `_parse_single_example_raw` documentation for more details.
features: A `dict` mapping feature keys to `FixedLenFeature` or
`VarLenFeature` values.
example_names: (Optional) A scalar string Tensor, the associated name.
See `_parse_single_example_raw` documentation for more details.
name: A name for this operation (optional).
Returns:
A `dict` mapping feature keys to `Tensor` and `SparseTensor` values.
Raises:
ValueError: if any feature is invalid.
"""
if not features:
raise ValueError("Missing features.")
if example_names is None:
return parse_single_example_v2(serialized, features, name)
features = _prepend_none_dimension(features)
(sparse_keys, sparse_types, dense_keys, dense_types, dense_defaults,
dense_shapes) = _features_to_raw_params(
features,
[VarLenFeature, FixedLenFeature, FixedLenSequenceFeature, SparseFeature])
outputs = _parse_single_example_raw(
serialized, example_names, sparse_keys, sparse_types, dense_keys,
dense_types, dense_defaults, dense_shapes, name)
return _construct_sparse_tensors_for_sparse_features(features, outputs)
def _parse_single_example_raw(serialized,
names=None,
sparse_keys=None,
sparse_types=None,
dense_keys=None,
dense_types=None,
dense_defaults=None,
dense_shapes=None,
name=None):
"""Parses a single `Example` proto.
Args:
serialized: A scalar string Tensor, a single serialized Example.
See `_parse_example_raw` documentation for more details.
names: (Optional) A scalar string Tensor, the associated name.
See `_parse_example_raw` documentation for more details.
sparse_keys: See `_parse_example_raw` documentation for more details.
sparse_types: See `_parse_example_raw` documentation for more details.
dense_keys: See `_parse_example_raw` documentation for more details.
dense_types: See `_parse_example_raw` documentation for more details.
dense_defaults: See `_parse_example_raw` documentation for more details.
dense_shapes: See `_parse_example_raw` documentation for more details.
name: A name for this operation (optional).
Returns:
A `dict` mapping feature keys to `Tensor` and `SparseTensor` values.
Raises:
ValueError: if any feature is invalid.
"""
with ops.name_scope(name, "ParseSingleExample", [serialized, names]):
serialized = ops.convert_to_tensor(serialized)
serialized_shape = serialized.get_shape()
if serialized_shape.ndims is not None:
if serialized_shape.ndims != 0:
raise ValueError("Input serialized must be a scalar")
else:
serialized = control_flow_ops.with_dependencies(
[control_flow_ops.Assert(
math_ops.equal(array_ops.rank(serialized), 0),
["Input serialized must be a scalar"],
name="SerializedIsScalar")],
serialized,
name="SerializedDependencies")
serialized = array_ops.expand_dims(serialized, 0)
if names is not None:
names = ops.convert_to_tensor(names)
names_shape = names.get_shape()
if names_shape.ndims is not None:
if names_shape.ndims != 0:
raise ValueError("Input names must be a scalar")
else:
names = control_flow_ops.with_dependencies(
[control_flow_ops.Assert(
math_ops.equal(array_ops.rank(names), 0),
["Input names must be a scalar"],
name="NamesIsScalar")],
names,
name="NamesDependencies")
names = array_ops.expand_dims(names, 0)
outputs = _parse_example_raw(
serialized,
names=names,
sparse_keys=sparse_keys,
sparse_types=sparse_types,
dense_keys=dense_keys,
dense_types=dense_types,
dense_defaults=dense_defaults,
dense_shapes=dense_shapes,
name=name)
if dense_keys is not None:
for d in dense_keys:
d_name = re.sub("[^A-Za-z0-9_.\\-/]", "_", d)
outputs[d] = array_ops.squeeze(
outputs[d], [0], name="Squeeze_%s" % d_name)
if sparse_keys is not None:
for s in sparse_keys:
s_name = re.sub("[^A-Za-z0-9_.\\-/]", "_", s)
outputs[s] = sparse_tensor.SparseTensor(
array_ops.slice(outputs[s].indices,
[0, 1], [-1, -1], name="Slice_Indices_%s" % s_name),
outputs[s].values,
array_ops.slice(outputs[s].dense_shape,
[1], [-1], name="Squeeze_Shape_%s" % s_name))
return outputs
@tf_export("io.parse_sequence_example")
def parse_sequence_example(serialized,
context_features=None,
sequence_features=None,
example_names=None,
name=None):
# pylint: disable=line-too-long
"""Parses a batch of `SequenceExample` protos.
Parses a vector of serialized
[`SequenceExample`](https://www.tensorflow.org/code/tensorflow/core/example/example.proto)
protos given in `serialized`.
This op parses serialized sequence examples into a tuple of dictionaries
mapping keys to `Tensor` and `SparseTensor` objects respectively.
The first dictionary contains mappings for keys appearing in
`context_features`, and the second dictionary contains mappings for keys
appearing in `sequence_features`.
At least one of `context_features` and `sequence_features` must be provided
and non-empty.
The `context_features` keys are associated with a `SequenceExample` as a
whole, independent of time / frame. In contrast, the `sequence_features` keys
provide a way to access variable-length data within the `FeatureList` section
of the `SequenceExample` proto. While the shapes of `context_features` values
are fixed with respect to frame, the frame dimension (the first dimension)
of `sequence_features` values may vary between `SequenceExample` protos,
and even between `feature_list` keys within the same `SequenceExample`.
`context_features` contains `VarLenFeature` and `FixedLenFeature` objects.
Each `VarLenFeature` is mapped to a `SparseTensor`, and each `FixedLenFeature`
is mapped to a `Tensor`, of the specified type, shape, and default value.
`sequence_features` contains `VarLenFeature` and `FixedLenSequenceFeature`
objects. Each `VarLenFeature` is mapped to a `SparseTensor`, and each
`FixedLenSequenceFeature` is mapped to a `Tensor`, each of the specified type.
The shape will be `(B,T,) + df.dense_shape` for `FixedLenSequenceFeature`
`df`, where `B` is the batch size, and `T` is the length of the associated
`FeatureList` in the `SequenceExample`. For instance,
`FixedLenSequenceFeature([])` yields a scalar 2-D `Tensor` of static shape
`[None, None]` and dynamic shape `[B, T]`, while
`FixedLenSequenceFeature([k])` (for `int k >= 1`) yields a 3-D matrix `Tensor`
of static shape `[None, None, k]` and dynamic shape `[B, T, k]`.
Like the input, the resulting output tensors have a batch dimension. This
means that the original per-example shapes of `VarLenFeature`s and
`FixedLenSequenceFeature`s can be lost. To handle that situation, this op also
provides dicts of shape tensors as part of the output. There is one dict for
the context features, and one for the feature_list features. Context features
of type `FixedLenFeature`s will not be present, since their shapes are already
known by the caller. In situations where the input 'FixedLenFeature`s are of
different lengths across examples, the shorter examples will be padded with
default datatype values: 0 for numeric types, and the empty string for string
types.
Each `SparseTensor` corresponding to `sequence_features` represents a ragged
vector. Its indices are `[time, index]`, where `time` is the `FeatureList`
entry and `index` is the value's index in the list of values associated with
that time.
`FixedLenFeature` entries with a `default_value` and `FixedLenSequenceFeature`
entries with `allow_missing=True` are optional; otherwise, we will fail if
that `Feature` or `FeatureList` is missing from any example in `serialized`.
`example_name` may contain a descriptive name for the corresponding serialized
proto. This may be useful for debugging purposes, but it has no effect on the
output. If not `None`, `example_name` must be a scalar.
Args:
serialized: A vector (1-D Tensor) of type string containing binary
serialized `SequenceExample` protos.
context_features: A `dict` mapping feature keys to `FixedLenFeature` or
`VarLenFeature` values. These features are associated with a
`SequenceExample` as a whole.
sequence_features: A `dict` mapping feature keys to
`FixedLenSequenceFeature` or `VarLenFeature` values. These features are
associated with data within the `FeatureList` section of the
`SequenceExample` proto.
example_names: A vector (1-D Tensor) of strings (optional), the name of the
serialized protos.
name: A name for this operation (optional).
Returns:
A tuple of three `dict`s, each mapping keys to `Tensor`s and
`SparseTensor`s. The first dict contains the context key/values,
the second dict contains the feature_list key/values, and the final dict
contains the lengths of any dense feature_list features.
Raises:
ValueError: if any feature is invalid.
"""
if not (context_features or sequence_features):
raise ValueError("Missing features.")
(context_sparse_keys, context_sparse_types, context_dense_keys,
context_dense_types,
context_dense_defaults, context_dense_shapes) = _features_to_raw_params(
context_features, [VarLenFeature, FixedLenFeature])
(feature_list_sparse_keys, feature_list_sparse_types, feature_list_dense_keys,
feature_list_dense_types, feature_list_dense_defaults,
feature_list_dense_shapes) = _features_to_raw_params(
sequence_features, [VarLenFeature, FixedLenSequenceFeature])
return _parse_sequence_example_raw(
serialized, example_names, context_sparse_keys, context_sparse_types,
context_dense_keys, context_dense_types, context_dense_defaults,
context_dense_shapes, feature_list_sparse_keys, feature_list_sparse_types,
feature_list_dense_keys, feature_list_dense_types,
feature_list_dense_shapes, feature_list_dense_defaults, name)
def _parse_sequence_example_raw(serialized,
debug_name=None,
context_sparse_keys=None,
context_sparse_types=None,
context_dense_keys=None,
context_dense_types=None,
context_dense_defaults=None,
context_dense_shapes=None,
feature_list_sparse_keys=None,
feature_list_sparse_types=None,
feature_list_dense_keys=None,
feature_list_dense_types=None,
feature_list_dense_shapes=None,
feature_list_dense_defaults=None,
name=None):
"""Parses a vector of `SequenceExample` protos.
Args:
serialized: A vector (1-D Tensor) of type string, containing binary
serialized `SequenceExample` protos.
debug_name: A vector (1-D Tensor) of strings (optional), the names of the
serialized protos.
context_sparse_keys: A list of string keys in the `SequenceExample`'s
features. The results for these keys will be returned as `SparseTensor`
objects.
context_sparse_types: A list of `DTypes`, the same length as `sparse_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`), and `tf.string`
(`BytesList`) are supported.
context_dense_keys: A list of string keys in the examples' features. The
results for these keys will be returned as `Tensor`s
context_dense_types: A list of DTypes, same length as `context_dense_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`), and `tf.string`
(`BytesList`) are supported.
context_dense_defaults: A dict mapping string keys to `Tensor`s. The keys of
the dict must match the context_dense_keys of the feature.
context_dense_shapes: A list of tuples, same length as `context_dense_keys`.
The shape of the data for each context_dense feature referenced by
`context_dense_keys`. Required for any input tensors identified by
`context_dense_keys` whose shapes are anything other than `[]` or `[1]`.
feature_list_sparse_keys: A list of string keys in the `SequenceExample`'s
feature_lists. The results for these keys will be returned as
`SparseTensor` objects.
feature_list_sparse_types: A list of `DTypes`, same length as `sparse_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`), and `tf.string`
(`BytesList`) are supported.
feature_list_dense_keys: A list of string keys in the `SequenceExample`'s
features_lists. The results for these keys will be returned as `Tensor`s.
feature_list_dense_types: A list of `DTypes`, same length as
`feature_list_dense_keys`. Only `tf.float32` (`FloatList`), `tf.int64`
(`Int64List`), and `tf.string` (`BytesList`) are supported.
feature_list_dense_shapes: A list of tuples, same length as
`feature_list_dense_keys`. The shape of the data for each `FeatureList`
feature referenced by `feature_list_dense_keys`.
feature_list_dense_defaults: A dict mapping key strings to values. The only
currently allowed value is `None`. Any key appearing in this dict with
value `None` is allowed to be missing from the `SequenceExample`. If
missing, the key is treated as zero-length.
name: A name for this operation (optional).
Returns:
A tuple of three `dict`s, each mapping keys to `Tensor`s and
`SparseTensor`s. The first dict contains the context key/values,
the second dict contains the feature_list key/values, and the final dict
contains the lengths of any dense feature_list features.
Raises:
ValueError: If context_sparse and context_dense key sets intersect,
if feature_list_sparse and feature_list_dense key sets intersect,
if input lengths do not match up, or if a value in
feature_list_dense_defaults is not None.
TypeError: if feature_list_dense_defaults is not either None or a dict.
"""
with ops.name_scope(name, "ParseSequenceExample", [serialized]):
context_dense_defaults = ({} if context_dense_defaults is None else
context_dense_defaults)
context_sparse_keys = ([] if context_sparse_keys is None else
context_sparse_keys)
context_sparse_types = ([] if context_sparse_types is None else
context_sparse_types)
context_dense_keys = ([]
if context_dense_keys is None else context_dense_keys)
context_dense_types = ([] if context_dense_types is None else
context_dense_types)
context_dense_shapes = ([[]] * len(context_dense_keys)
if context_dense_shapes is None else
context_dense_shapes)
feature_list_sparse_keys = ([] if feature_list_sparse_keys is None else
feature_list_sparse_keys)
feature_list_sparse_types = ([] if feature_list_sparse_types is None else
feature_list_sparse_types)
feature_list_dense_keys = ([] if feature_list_dense_keys is None else
feature_list_dense_keys)
feature_list_dense_types = ([] if feature_list_dense_types is None else
feature_list_dense_types)
feature_list_dense_shapes = ([[]] * len(feature_list_dense_keys)
if feature_list_dense_shapes is None else
feature_list_dense_shapes)
feature_list_dense_defaults = (
dict()
if feature_list_dense_defaults is None else feature_list_dense_defaults)
debug_name = [] if debug_name is None else debug_name
# Internal
feature_list_dense_missing_assumed_empty = []
num_context_dense = len(context_dense_keys)
num_feature_list_dense = len(feature_list_dense_keys)
num_context_sparse = len(context_sparse_keys)
num_feature_list_sparse = len(feature_list_sparse_keys)
if len(context_dense_shapes) != num_context_dense:
raise ValueError(
"len(context_dense_shapes) != len(context_dense_keys): %d vs. %d" %
(len(context_dense_shapes), num_context_dense))
if len(context_dense_types) != num_context_dense:
raise ValueError(
"len(context_dense_types) != len(num_context_dense): %d vs. %d" %
(len(context_dense_types), num_context_dense))
if len(feature_list_dense_shapes) != num_feature_list_dense:
raise ValueError(
"len(feature_list_dense_shapes) != len(feature_list_dense_keys): "
"%d vs. %d" % (len(feature_list_dense_shapes),
num_feature_list_dense))
if len(feature_list_dense_types) != num_feature_list_dense:
raise ValueError(
"len(feature_list_dense_types) != len(num_feature_list_dense):"
"%d vs. %d" % (len(feature_list_dense_types), num_feature_list_dense))
if len(context_sparse_types) != num_context_sparse:
raise ValueError(
"len(context_sparse_types) != len(context_sparse_keys): %d vs. %d" %
(len(context_sparse_types), num_context_sparse))
if len(feature_list_sparse_types) != num_feature_list_sparse:
raise ValueError(
"len(feature_list_sparse_types) != len(feature_list_sparse_keys): "
"%d vs. %d" % (len(feature_list_sparse_types),
num_feature_list_sparse))
if (num_context_dense + num_context_sparse + num_feature_list_dense +
num_feature_list_sparse) == 0:
raise ValueError(
"Must provide at least one context_sparse key, context_dense key, "
", feature_list_sparse key, or feature_list_dense key")
if not set(context_dense_keys).isdisjoint(set(context_sparse_keys)):
raise ValueError(
"context_dense and context_sparse keys must not intersect; "
"intersection: %s" % set(context_dense_keys).intersection(
set(context_sparse_keys)))
if not set(feature_list_dense_keys).isdisjoint(
set(feature_list_sparse_keys)):
raise ValueError(
"feature_list_dense and feature_list_sparse keys must not intersect; "
"intersection: %s" % set(feature_list_dense_keys).intersection(
set(feature_list_sparse_keys)))
if not isinstance(feature_list_dense_defaults, dict):
raise TypeError("feature_list_dense_defaults must be a dict")
for k, v in feature_list_dense_defaults.items():
if v is not None:
raise ValueError(
"Value feature_list_dense_defaults[%s] must be None" % k)
feature_list_dense_missing_assumed_empty.append(k)
context_dense_defaults_vec = []
for i, key in enumerate(context_dense_keys):
default_value = context_dense_defaults.get(key)
if default_value is None:
default_value = constant_op.constant([], dtype=context_dense_types[i])
elif not isinstance(default_value, ops.Tensor):
key_name = "key_" + re.sub("[^A-Za-z0-9_.\\-/]", "_", key)
default_value = ops.convert_to_tensor(
default_value, dtype=context_dense_types[i], name=key_name)
context_dense_defaults_vec.append(default_value)
context_dense_shapes = [
tensor_shape.as_shape(shape).as_proto()
for shape in context_dense_shapes
]
feature_list_dense_shapes = [
tensor_shape.as_shape(shape).as_proto()
for shape in feature_list_dense_shapes
]
# pylint: disable=protected-access
outputs = gen_parsing_ops.parse_sequence_example(
serialized=serialized,
debug_name=debug_name,
Ncontext_sparse=num_context_sparse,
Ncontext_dense=num_context_dense,
Nfeature_list_sparse=num_feature_list_sparse,
Nfeature_list_dense=num_feature_list_dense,
context_dense_defaults=context_dense_defaults_vec,
context_sparse_keys=context_sparse_keys,
context_sparse_types=context_sparse_types,
context_dense_keys=context_dense_keys,
context_dense_shapes=context_dense_shapes,
feature_list_sparse_keys=feature_list_sparse_keys,
feature_list_sparse_types=feature_list_sparse_types,
feature_list_dense_keys=feature_list_dense_keys,
feature_list_dense_types=feature_list_dense_types,
feature_list_dense_shapes=feature_list_dense_shapes,
feature_list_dense_missing_assumed_empty=(
feature_list_dense_missing_assumed_empty),
name=name)
# pylint: enable=protected-access
(context_sparse_indices, context_sparse_values, context_sparse_shapes,
context_dense_values, feature_list_sparse_indices,
feature_list_sparse_values, feature_list_sparse_shapes,
feature_list_dense_values, feature_list_dense_lengths) = outputs
context_sparse_tensors = [
sparse_tensor.SparseTensor(ix, val, shape)
for (ix, val,
shape) in zip(context_sparse_indices, context_sparse_values,
context_sparse_shapes)
]
feature_list_sparse_tensors = [
sparse_tensor.SparseTensor(ix, val, shape)
for (ix, val, shape
) in zip(feature_list_sparse_indices, feature_list_sparse_values,
feature_list_sparse_shapes)
]
context_output = dict(
zip(context_sparse_keys + context_dense_keys,
context_sparse_tensors + context_dense_values))
feature_list_output = dict(
zip(feature_list_sparse_keys + feature_list_dense_keys,
feature_list_sparse_tensors + feature_list_dense_values))
feature_list_lengths = dict(
zip(feature_list_dense_keys, feature_list_dense_lengths))
return (context_output, feature_list_output, feature_list_lengths)
# TODO(sundberg): rewrite this method to call the batch version, which is more
# efficient especially for large inputs.
@tf_export("io.parse_single_sequence_example",
v1=["io.parse_single_sequence_example",
"parse_single_sequence_example"])
def parse_single_sequence_example(
serialized, context_features=None, sequence_features=None,
example_name=None, name=None):
# pylint: disable=line-too-long
"""Parses a single `SequenceExample` proto.
Parses a single serialized [`SequenceExample`](https://www.tensorflow.org/code/tensorflow/core/example/example.proto)
proto given in `serialized`.
This op parses a serialized sequence example into a tuple of dictionaries
mapping keys to `Tensor` and `SparseTensor` objects respectively.
The first dictionary contains mappings for keys appearing in
`context_features`, and the second dictionary contains mappings for keys
appearing in `sequence_features`.
At least one of `context_features` and `sequence_features` must be provided
and non-empty.
The `context_features` keys are associated with a `SequenceExample` as a
whole, independent of time / frame. In contrast, the `sequence_features` keys
provide a way to access variable-length data within the `FeatureList` section
of the `SequenceExample` proto. While the shapes of `context_features` values
are fixed with respect to frame, the frame dimension (the first dimension)
of `sequence_features` values may vary between `SequenceExample` protos,
and even between `feature_list` keys within the same `SequenceExample`.
`context_features` contains `VarLenFeature` and `FixedLenFeature` objects.
Each `VarLenFeature` is mapped to a `SparseTensor`, and each `FixedLenFeature`
is mapped to a `Tensor`, of the specified type, shape, and default value.
`sequence_features` contains `VarLenFeature` and `FixedLenSequenceFeature`
objects. Each `VarLenFeature` is mapped to a `SparseTensor`, and each
`FixedLenSequenceFeature` is mapped to a `Tensor`, each of the specified type.
The shape will be `(T,) + df.dense_shape` for `FixedLenSequenceFeature` `df`, where
`T` is the length of the associated `FeatureList` in the `SequenceExample`.
For instance, `FixedLenSequenceFeature([])` yields a scalar 1-D `Tensor` of
static shape `[None]` and dynamic shape `[T]`, while
`FixedLenSequenceFeature([k])` (for `int k >= 1`) yields a 2-D matrix `Tensor`
of static shape `[None, k]` and dynamic shape `[T, k]`.
Each `SparseTensor` corresponding to `sequence_features` represents a ragged
vector. Its indices are `[time, index]`, where `time` is the `FeatureList`
entry and `index` is the value's index in the list of values associated with
that time.
`FixedLenFeature` entries with a `default_value` and `FixedLenSequenceFeature`
entries with `allow_missing=True` are optional; otherwise, we will fail if
that `Feature` or `FeatureList` is missing from any example in `serialized`.
`example_name` may contain a descriptive name for the corresponding serialized
proto. This may be useful for debugging purposes, but it has no effect on the
output. If not `None`, `example_name` must be a scalar.
Args:
serialized: A scalar (0-D Tensor) of type string, a single binary
serialized `SequenceExample` proto.
context_features: A `dict` mapping feature keys to `FixedLenFeature` or
`VarLenFeature` values. These features are associated with a
`SequenceExample` as a whole.
sequence_features: A `dict` mapping feature keys to
`FixedLenSequenceFeature` or `VarLenFeature` values. These features are
associated with data within the `FeatureList` section of the
`SequenceExample` proto.
example_name: A scalar (0-D Tensor) of strings (optional), the name of
the serialized proto.
name: A name for this operation (optional).
Returns:
A tuple of two `dict`s, each mapping keys to `Tensor`s and `SparseTensor`s.
The first dict contains the context key/values.
The second dict contains the feature_list key/values.
Raises:
ValueError: if any feature is invalid.
"""
# pylint: enable=line-too-long
if not (context_features or sequence_features):
raise ValueError("Missing features.")
(context_sparse_keys, context_sparse_types, context_dense_keys,
context_dense_types, context_dense_defaults,
context_dense_shapes) = _features_to_raw_params(
context_features, [VarLenFeature, FixedLenFeature])
(feature_list_sparse_keys, feature_list_sparse_types,
feature_list_dense_keys, feature_list_dense_types,
feature_list_dense_defaults,
feature_list_dense_shapes) = _features_to_raw_params(
sequence_features, [VarLenFeature, FixedLenSequenceFeature])
return _parse_single_sequence_example_raw(
serialized, context_sparse_keys, context_sparse_types,
context_dense_keys, context_dense_types, context_dense_defaults,
context_dense_shapes, feature_list_sparse_keys,
feature_list_sparse_types, feature_list_dense_keys,
feature_list_dense_types, feature_list_dense_shapes,
feature_list_dense_defaults, example_name, name)
def _parse_single_sequence_example_raw(serialized,
context_sparse_keys=None,
context_sparse_types=None,
context_dense_keys=None,
context_dense_types=None,
context_dense_defaults=None,
context_dense_shapes=None,
feature_list_sparse_keys=None,
feature_list_sparse_types=None,
feature_list_dense_keys=None,
feature_list_dense_types=None,
feature_list_dense_shapes=None,
feature_list_dense_defaults=None,
debug_name=None,
name=None):
"""Parses a single `SequenceExample` proto.
Args:
serialized: A scalar (0-D Tensor) of type string, a single binary
serialized `SequenceExample` proto.
context_sparse_keys: A list of string keys in the `SequenceExample`'s
features. The results for these keys will be returned as
`SparseTensor` objects.
context_sparse_types: A list of `DTypes`, the same length as `sparse_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`),
and `tf.string` (`BytesList`) are supported.
context_dense_keys: A list of string keys in the examples' features.
The results for these keys will be returned as `Tensor`s
context_dense_types: A list of DTypes, same length as `context_dense_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`),
and `tf.string` (`BytesList`) are supported.
context_dense_defaults: A dict mapping string keys to `Tensor`s.
The keys of the dict must match the context_dense_keys of the feature.
context_dense_shapes: A list of tuples, same length as `context_dense_keys`.
The shape of the data for each context_dense feature referenced by
`context_dense_keys`. Required for any input tensors identified by
`context_dense_keys` whose shapes are anything other than `[]` or `[1]`.
feature_list_sparse_keys: A list of string keys in the `SequenceExample`'s
feature_lists. The results for these keys will be returned as
`SparseTensor` objects.
feature_list_sparse_types: A list of `DTypes`, same length as `sparse_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`),
and `tf.string` (`BytesList`) are supported.
feature_list_dense_keys: A list of string keys in the `SequenceExample`'s
features_lists. The results for these keys will be returned as `Tensor`s.
feature_list_dense_types: A list of `DTypes`, same length as
`feature_list_dense_keys`. Only `tf.float32` (`FloatList`),
`tf.int64` (`Int64List`), and `tf.string` (`BytesList`) are supported.
feature_list_dense_shapes: A list of tuples, same length as
`feature_list_dense_keys`. The shape of the data for each
`FeatureList` feature referenced by `feature_list_dense_keys`.
feature_list_dense_defaults: A dict mapping key strings to values.
The only currently allowed value is `None`. Any key appearing
in this dict with value `None` is allowed to be missing from the
`SequenceExample`. If missing, the key is treated as zero-length.
debug_name: A scalar (0-D Tensor) of strings (optional), the name of
the serialized proto.
name: A name for this operation (optional).
Returns:
A tuple of two `dict`s, each mapping keys to `Tensor`s and `SparseTensor`s.
The first dict contains the context key/values.
The second dict contains the feature_list key/values.
Raises:
ValueError: If context_sparse and context_dense key sets intersect,
if input lengths do not match up, or if a value in
feature_list_dense_defaults is not None.
TypeError: if feature_list_dense_defaults is not either None or a dict.
"""
with ops.name_scope(name, "ParseSingleSequenceExample", [serialized]):
context_dense_defaults = (
{} if context_dense_defaults is None else context_dense_defaults)
context_sparse_keys = (
[] if context_sparse_keys is None else context_sparse_keys)
context_sparse_types = (
[] if context_sparse_types is None else context_sparse_types)
context_dense_keys = (
[] if context_dense_keys is None else context_dense_keys)
context_dense_types = (
[] if context_dense_types is None else context_dense_types)
context_dense_shapes = (
[[]] * len(context_dense_keys)
if context_dense_shapes is None else context_dense_shapes)
feature_list_sparse_keys = (
[] if feature_list_sparse_keys is None else feature_list_sparse_keys)
feature_list_sparse_types = (
[] if feature_list_sparse_types is None else feature_list_sparse_types)
feature_list_dense_keys = (
[] if feature_list_dense_keys is None else feature_list_dense_keys)
feature_list_dense_types = (
[] if feature_list_dense_types is None else feature_list_dense_types)
feature_list_dense_shapes = (
[[]] * len(feature_list_dense_keys)
if feature_list_dense_shapes is None else feature_list_dense_shapes)
feature_list_dense_defaults = (
dict() if feature_list_dense_defaults is None
else feature_list_dense_defaults)
debug_name = "" if debug_name is None else debug_name
# Internal
feature_list_dense_missing_assumed_empty = []
num_context_dense = len(context_dense_keys)
num_feature_list_dense = len(feature_list_dense_keys)
num_context_sparse = len(context_sparse_keys)
num_feature_list_sparse = len(feature_list_sparse_keys)
if len(context_dense_shapes) != num_context_dense:
raise ValueError(
"len(context_dense_shapes) != len(context_dense_keys): %d vs. %d"
% (len(context_dense_shapes), num_context_dense))
if len(context_dense_types) != num_context_dense:
raise ValueError(
"len(context_dense_types) != len(num_context_dense): %d vs. %d"
% (len(context_dense_types), num_context_dense))
if len(feature_list_dense_shapes) != num_feature_list_dense:
raise ValueError(
"len(feature_list_dense_shapes) != len(feature_list_dense_keys): "
"%d vs. %d" % (len(feature_list_dense_shapes),
num_feature_list_dense))
if len(feature_list_dense_types) != num_feature_list_dense:
raise ValueError(
"len(feature_list_dense_types) != len(num_feature_list_dense):"
"%d vs. %d" % (len(feature_list_dense_types), num_feature_list_dense))
if len(context_sparse_types) != num_context_sparse:
raise ValueError(
"len(context_sparse_types) != len(context_sparse_keys): %d vs. %d"
% (len(context_sparse_types), num_context_sparse))
if len(feature_list_sparse_types) != num_feature_list_sparse:
raise ValueError(
"len(feature_list_sparse_types) != len(feature_list_sparse_keys): "
"%d vs. %d"
% (len(feature_list_sparse_types), num_feature_list_sparse))
if (num_context_dense + num_context_sparse
+ num_feature_list_dense + num_feature_list_sparse) == 0:
raise ValueError(
"Must provide at least one context_sparse key, context_dense key, "
", feature_list_sparse key, or feature_list_dense key")
if not set(context_dense_keys).isdisjoint(set(context_sparse_keys)):
raise ValueError(
"context_dense and context_sparse keys must not intersect; "
"intersection: %s" %
set(context_dense_keys).intersection(set(context_sparse_keys)))
if not set(feature_list_dense_keys).isdisjoint(
set(feature_list_sparse_keys)):
raise ValueError(
"feature_list_dense and feature_list_sparse keys must not intersect; "
"intersection: %s" %
set(feature_list_dense_keys).intersection(
set(feature_list_sparse_keys)))
if not isinstance(feature_list_dense_defaults, dict):
raise TypeError("feature_list_dense_defaults must be a dict")
for k, v in feature_list_dense_defaults.items():
if v is not None:
raise ValueError("Value feature_list_dense_defaults[%s] must be None"
% k)
feature_list_dense_missing_assumed_empty.append(k)
context_dense_defaults_vec = []
for i, key in enumerate(context_dense_keys):
default_value = context_dense_defaults.get(key)
if default_value is None:
default_value = constant_op.constant([], dtype=context_dense_types[i])
elif not isinstance(default_value, ops.Tensor):
key_name = "key_" + re.sub("[^A-Za-z0-9_.\\-/]", "_", key)
default_value = ops.convert_to_tensor(
default_value, dtype=context_dense_types[i], name=key_name)
default_value = array_ops.reshape(
default_value, context_dense_shapes[i])
context_dense_defaults_vec.append(default_value)
context_dense_shapes = [tensor_shape.as_shape(shape).as_proto()
for shape in context_dense_shapes]
feature_list_dense_shapes = [tensor_shape.as_shape(shape).as_proto()
for shape in feature_list_dense_shapes]
outputs = gen_parsing_ops.parse_single_sequence_example(
serialized=serialized,
debug_name=debug_name,
context_dense_defaults=context_dense_defaults_vec,
context_sparse_keys=context_sparse_keys,
context_sparse_types=context_sparse_types,
context_dense_keys=context_dense_keys,
context_dense_shapes=context_dense_shapes,
feature_list_sparse_keys=feature_list_sparse_keys,
feature_list_sparse_types=feature_list_sparse_types,
feature_list_dense_keys=feature_list_dense_keys,
feature_list_dense_types=feature_list_dense_types,
feature_list_dense_shapes=feature_list_dense_shapes,
feature_list_dense_missing_assumed_empty=(
feature_list_dense_missing_assumed_empty),
name=name)
(context_sparse_indices, context_sparse_values,
context_sparse_shapes, context_dense_values,
feature_list_sparse_indices, feature_list_sparse_values,
feature_list_sparse_shapes, feature_list_dense_values) = outputs
context_sparse_tensors = [
sparse_tensor.SparseTensor(ix, val, shape) for (ix, val, shape)
in zip(context_sparse_indices,
context_sparse_values,
context_sparse_shapes)]
feature_list_sparse_tensors = [
sparse_tensor.SparseTensor(ix, val, shape) for (ix, val, shape)
in zip(feature_list_sparse_indices,
feature_list_sparse_values,
feature_list_sparse_shapes)]
context_output = dict(
zip(context_sparse_keys + context_dense_keys,
context_sparse_tensors + context_dense_values))
feature_list_output = dict(
zip(feature_list_sparse_keys + feature_list_dense_keys,
feature_list_sparse_tensors + feature_list_dense_values))
return (context_output, feature_list_output)
@tf_export("io.decode_raw", v1=[])
def decode_raw(input_bytes,
out_type,
little_endian=True,
fixed_length=None,
name=None):
"""Convert raw byte strings into tensors.
Args:
input_bytes:
Each element of the input Tensor is converted to an array of bytes.
out_type:
`DType` of the output. Acceptable types are `half`, `float`, `double`,
`int32`, `uint16`, `uint8`, `int16`, `int8`, `int64`.
little_endian:
Whether the `input_bytes` data is in little-endian format. Data will be
converted into host byte order if necessary.
fixed_length:
If set, the first `fixed_length` bytes of each element will be converted.
Data will be zero-padded or truncated to the specified length.
`fixed_length` must be a multiple of the size of `out_type`.
`fixed_length` must be specified if the elements of `input_bytes` are of
variable length.
name: A name for the operation (optional).
Returns:
A `Tensor` object storing the decoded bytes.
"""
if fixed_length is not None:
return gen_parsing_ops.decode_padded_raw(
input_bytes,
fixed_length=fixed_length,
out_type=out_type,
little_endian=little_endian,
name=name)
else:
return gen_parsing_ops.decode_raw(
input_bytes, out_type, little_endian=little_endian, name=name)
@tf_export(v1=["decode_raw", "io.decode_raw"])
@deprecation.deprecated_args(None,
"bytes is deprecated, use input_bytes instead",
"bytes")
def decode_raw_v1(
input_bytes=None,
out_type=None,
little_endian=True,
name=None,
bytes=None # pylint: disable=redefined-builtin
):
"""Convert raw byte strings into tensors.
Args:
input_bytes:
Each element of the input Tensor is converted to an array of bytes.
out_type:
`DType` of the output. Acceptable types are `half`, `float`, `double`,
`int32`, `uint16`, `uint8`, `int16`, `int8`, `int64`.
little_endian:
Whether the `input_bytes` data is in little-endian format. Data will be
converted into host byte order if necessary.
name: A name for the operation (optional).
bytes: Deprecated parameter. Use `input_bytes` instead.
Returns:
A `Tensor` object storing the decoded bytes.
"""
input_bytes = deprecation.deprecated_argument_lookup("input_bytes",
input_bytes, "bytes",
bytes)
# out_type is a required positional argument in the original API, and had to
# be changed to a keyword argument in order to facilitate the transition from
# the reserved named `bytes` to `input_bytes`. Ensure it's still set.
if out_type is None:
raise ValueError(
"decode_raw_v1() missing 1 positional argument: 'out_type'")
return gen_parsing_ops.decode_raw(
input_bytes, out_type, little_endian=little_endian, name=name)
# Swap `name` and `na_value` for backward compatibility.
@tf_export(v1=["io.decode_csv", "decode_csv"])
@deprecation.deprecated_endpoints("decode_csv")
def decode_csv(records,
record_defaults,
field_delim=",",
use_quote_delim=True,
name=None,
na_value="",
select_cols=None):
"""Convert CSV records to tensors. Each column maps to one tensor.
RFC 4180 format is expected for the CSV records.
(https://tools.ietf.org/html/rfc4180)
Note that we allow leading and trailing spaces with int or float field.
Args:
records: A `Tensor` of type `string`.
Each string is a record/row in the csv and all records should have
the same format.
record_defaults: A list of `Tensor` objects with specific types.
Acceptable types are `float32`, `float64`, `int32`, `int64`, `string`.
One tensor per column of the input record, with either a
scalar default value for that column or an empty vector if the column is
required.
field_delim: An optional `string`. Defaults to `","`.
char delimiter to separate fields in a record.
use_quote_delim: An optional `bool`. Defaults to `True`.
If false, treats double quotation marks as regular
characters inside of the string fields (ignoring RFC 4180, Section 2,
Bullet 5).
name: A name for the operation (optional).
na_value: Additional string to recognize as NA/NaN.
select_cols: Optional sorted list of column indices to select. If specified,
only this subset of columns will be parsed and returned.
Returns:
A list of `Tensor` objects. Has the same type as `record_defaults`.
Each tensor will have the same shape as records.
Raises:
ValueError: If any of the arguments is malformed.
"""
return decode_csv_v2(
records, record_defaults,
field_delim, use_quote_delim,
na_value, select_cols, name
)
@tf_export("io.decode_csv", v1=[])
def decode_csv_v2(records,
record_defaults,
field_delim=",",
use_quote_delim=True,
na_value="",
select_cols=None,
name=None):
"""Convert CSV records to tensors. Each column maps to one tensor.
RFC 4180 format is expected for the CSV records.
(https://tools.ietf.org/html/rfc4180)
Note that we allow leading and trailing spaces with int or float field.
Args:
records: A `Tensor` of type `string`.
Each string is a record/row in the csv and all records should have
the same format.
record_defaults: A list of `Tensor` objects with specific types.
Acceptable types are `float32`, `float64`, `int32`, `int64`, `string`.
One tensor per column of the input record, with either a
scalar default value for that column or an empty vector if the column is
required.
field_delim: An optional `string`. Defaults to `","`.
char delimiter to separate fields in a record.
use_quote_delim: An optional `bool`. Defaults to `True`.
If false, treats double quotation marks as regular
characters inside of the string fields (ignoring RFC 4180, Section 2,
Bullet 5).
na_value: Additional string to recognize as NA/NaN.
select_cols: Optional sorted list of column indices to select. If specified,
only this subset of columns will be parsed and returned.
name: A name for the operation (optional).
Returns:
A list of `Tensor` objects. Has the same type as `record_defaults`.
Each tensor will have the same shape as records.
Raises:
ValueError: If any of the arguments is malformed.
"""
if select_cols is not None and any(select_cols[i] >= select_cols[i + 1]
for i in range(len(select_cols) - 1)):
raise ValueError("select_cols is not strictly increasing.")
if select_cols is not None and select_cols[0] < 0:
raise ValueError("select_cols contains negative values.")
if select_cols is not None and len(select_cols) != len(record_defaults):
raise ValueError("Length of select_cols and record_defaults do not match.")
return gen_parsing_ops.decode_csv(
records=records,
record_defaults=record_defaults,
field_delim=field_delim,
use_quote_delim=use_quote_delim,
na_value=na_value,
name=name,
select_cols=select_cols,
)
# TODO(b/70890287): Combine the implementation of this op and
# `parse_single_example()` after 1/10/2018.
def parse_single_example_v2(serialized, features, name=None):
# pylint: disable=line-too-long
"""Parses an `Example` proto into a `dict` of tensors.
Parses a serialized
[`Example`](https://www.tensorflow.org/code/tensorflow/core/example/example.proto)
proto given in `serialized`.
This op parses serialized examples into a dictionary mapping keys to `Tensor`
and `SparseTensor` objects. `features` is a dict from keys to `VarLenFeature`,
`SparseFeature`, and `FixedLenFeature` objects. Each `VarLenFeature`
and `SparseFeature` is mapped to a `SparseTensor`, and each
`FixedLenFeature` is mapped to a `Tensor`.
Each `VarLenFeature` maps to a `SparseTensor` of the specified type
representing a ragged matrix. Its indices are `[index]` where
`index` is the value's index in the list of values associated with
that feature and example.
Each `SparseFeature` maps to a `SparseTensor` of the specified type
representing a Tensor of `dense_shape` `SparseFeature.size`.
Its `values` come from the feature in the examples with key `value_key`.
A `values[i]` comes from a position `k` in the feature of an example at batch
entry `batch`. This positional information is recorded in `indices[i]` as
`[batch, index_0, index_1, ...]` where `index_j` is the `k-th` value of
the feature in the example at with key `SparseFeature.index_key[j]`.
In other words, we split the indices (except the first index indicating the
batch entry) of a `SparseTensor` by dimension into different features of the
`Example`. Due to its complexity a `VarLenFeature` should be preferred over a
`SparseFeature` whenever possible.
Each `FixedLenFeature` `df` maps to a `Tensor` of the specified type (or
`tf.float32` if not specified) and shape `df.shape`.
`FixedLenFeature` entries with a `default_value` are optional. With no default
value, we will fail if that `Feature` is missing from any example in
`serialized`.
Each `FixedLenSequenceFeature` `df` maps to a `Tensor` of the specified type
(or `tf.float32` if not specified) and shape `(None,) + df.shape`.
Args:
serialized: A scalar (0-D Tensor) string, a serialized `Example` proto.
features: A `dict` mapping feature keys to `FixedLenFeature`,
`VarLenFeature`, and `SparseFeature` values.
name: A name for this operation (optional).
Returns:
A `dict` mapping feature keys to `Tensor` and `SparseTensor` values.
Raises:
ValueError: if any feature is invalid.
"""
if not features:
raise ValueError("Missing: features was %s." % features)
features = _prepend_none_dimension(features)
(sparse_keys, sparse_types, dense_keys, dense_types,
dense_defaults, dense_shapes) = _features_to_raw_params(
features,
[VarLenFeature, SparseFeature, FixedLenFeature, FixedLenSequenceFeature])
outputs = _parse_single_example_v2_raw(serialized, sparse_keys, sparse_types,
dense_keys, dense_types,
dense_defaults, dense_shapes, name)
return _construct_sparse_tensors_for_sparse_features(features, outputs)
def _parse_single_example_v2_raw(serialized, sparse_keys, sparse_types,
dense_keys, dense_types, dense_defaults,
dense_shapes, name):
"""Parses `Example` protos.
Args:
serialized: A scalar (0-D Tensor) string, containing a binary
serialized `Example` proto.
sparse_keys: A list of string keys in the examples' features.
The results for these keys will be returned as `SparseTensor` objects.
sparse_types: A list of `DTypes` of the same length as `sparse_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`),
and `tf.string` (`BytesList`) are supported.
dense_keys: A list of string keys in the examples' features.
The results for these keys will be returned as `Tensor`s
dense_types: A list of DTypes of the same length as `dense_keys`.
Only `tf.float32` (`FloatList`), `tf.int64` (`Int64List`),
and `tf.string` (`BytesList`) are supported.
dense_defaults: A dict mapping string keys to `Tensor`s.
The keys of the dict must match the dense_keys of the feature.
dense_shapes: A list of tuples with the same length as `dense_keys`.
The shape of the data for each dense feature referenced by `dense_keys`.
Required for any input tensors identified by `dense_keys`. Must be
either fully defined, or may contain an unknown first dimension.
An unknown first dimension means the feature is treated as having
a variable number of blocks, and the output shape along this dimension
is considered unknown at graph build time. Padding is applied for
minibatch elements smaller than the maximum number of blocks for the
given feature along this dimension.
name: A name for this operation (optional).
Returns:
A `dict` mapping keys to `Tensor`s and `SparseTensor`s.
Raises:
ValueError: If sparse and dense key sets intersect, or input lengths do not
match up.
"""
with ops.name_scope(name, "ParseSingleExample", [serialized]):
serialized = ops.convert_to_tensor(serialized, name="serialized")
dense_defaults = collections.OrderedDict(
) if dense_defaults is None else dense_defaults
sparse_keys = [] if sparse_keys is None else sparse_keys
sparse_types = [] if sparse_types is None else sparse_types
dense_keys = [] if dense_keys is None else dense_keys
dense_types = [] if dense_types is None else dense_types
dense_shapes = ([[]] * len(dense_keys)
if dense_shapes is None else dense_shapes)
num_dense = len(dense_keys)
num_sparse = len(sparse_keys)
if len(dense_shapes) != num_dense:
raise ValueError("len(dense_shapes) != len(dense_keys): %d vs. %d" %
(len(dense_shapes), num_dense))
if len(dense_types) != num_dense:
raise ValueError("len(dense_types) != len(num_dense): %d vs. %d" %
(len(dense_types), num_dense))
if len(sparse_types) != num_sparse:
raise ValueError("len(sparse_types) != len(sparse_keys): %d vs. %d" %
(len(sparse_types), num_sparse))
if num_dense + num_sparse == 0:
raise ValueError("Must provide at least one sparse key or dense key")
if not set(dense_keys).isdisjoint(set(sparse_keys)):
raise ValueError(
"Dense and sparse keys must not intersect; intersection: %s" %
set(dense_keys).intersection(set(sparse_keys)))
# Convert dense_shapes to TensorShape object.
dense_shapes = [tensor_shape.as_shape(shape) for shape in dense_shapes]
dense_defaults_vec = []
for i, key in enumerate(dense_keys):
default_value = dense_defaults.get(key)
dense_shape = dense_shapes[i]
if (dense_shape.ndims is not None and dense_shape.ndims > 0 and
dense_shape.dims[0].value is None):
# Variable stride dense shape, the default value should be a
# scalar padding value
if default_value is None:
default_value = ops.convert_to_tensor(
"" if dense_types[i] == dtypes.string else 0,
dtype=dense_types[i])
else:
# Reshape to a scalar to ensure user gets an error if they
# provide a tensor that's not intended to be a padding value
# (0 or 2+ elements).
key_name = "padding_" + re.sub("[^A-Za-z0-9_.\\-/]", "_", key)
default_value = ops.convert_to_tensor(
default_value, dtype=dense_types[i], name=key_name)
default_value = array_ops.reshape(default_value, [])
else:
if default_value is None:
default_value = constant_op.constant([], dtype=dense_types[i])
elif not isinstance(default_value, ops.Tensor):
key_name = "key_" + re.sub("[^A-Za-z0-9_.\\-/]", "_", key)
default_value = ops.convert_to_tensor(
default_value, dtype=dense_types[i], name=key_name)
default_value = array_ops.reshape(default_value, dense_shape)
dense_defaults_vec.append(default_value)
# Finally, convert dense_shapes to TensorShapeProto
dense_shapes = [shape.as_proto() for shape in dense_shapes]
outputs = gen_parsing_ops.parse_single_example(
serialized=serialized,
dense_defaults=dense_defaults_vec,
num_sparse=len(sparse_keys),
sparse_keys=sparse_keys,
sparse_types=sparse_types,
dense_keys=dense_keys,
dense_shapes=dense_shapes,
name=name)
(sparse_indices, sparse_values, sparse_shapes, dense_values) = outputs
sparse_tensors = [
sparse_tensor.SparseTensor(ix, val, shape)
for (ix, val,
shape) in zip(sparse_indices, sparse_values, sparse_shapes)
]
return dict(zip(sparse_keys + dense_keys, sparse_tensors + dense_values))
|
tensorflow-master
|
tensorflow/python/ops/parsing_ops.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Public API for tf.initializer namespace."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.ops import init_ops
from tensorflow.python.ops import variables as _variables
# variable initializers
zeros = init_ops.zeros_initializer
ones = init_ops.ones_initializer
constant = init_ops.constant_initializer
random_uniform = init_ops.random_uniform_initializer
random_normal = init_ops.random_normal_initializer
truncated_normal = init_ops.truncated_normal_initializer
uniform_unit_scaling = init_ops.uniform_unit_scaling_initializer
variance_scaling = init_ops.variance_scaling_initializer
orthogonal = init_ops.orthogonal_initializer
identity = init_ops.identity_initializer
# variable initializer ops
variables = _variables.variables_initializer
global_variables = _variables.global_variables_initializer
local_variables = _variables.local_variables_initializer
# Seal API.
del absolute_import
del division
del print_function
del init_ops
del _variables
|
tensorflow-master
|
tensorflow/python/ops/initializers_ns.py
|
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#==============================================================================
"""Lookup operations."""
# pylint: disable=g-bad-name
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import functools
import uuid
import six
from tensorflow.python.compat import compat as fwd_compat
from tensorflow.python.eager import context
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import sparse_tensor
from tensorflow.python.framework import tensor_shape
from tensorflow.python.framework import tensor_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import gen_lookup_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import string_ops
# go/tf-wildcard-import
# pylint: disable=wildcard-import
from tensorflow.python.ops.gen_lookup_ops import *
from tensorflow.python.training.saver import BaseSaverBuilder
# pylint: enable=wildcard-import
from tensorflow.python.training.tracking import base as trackable_base
from tensorflow.python.training.tracking import tracking as trackable
from tensorflow.python.util import compat
from tensorflow.python.util.deprecation import deprecated
from tensorflow.python.util.tf_export import tf_export
@tf_export(v1=["initialize_all_tables"])
@deprecated(None, "Use `tf.tables_initializer` instead.")
def initialize_all_tables(name="init_all_tables"):
"""Returns an Op that initializes all tables of the default graph.
Args:
name: Optional name for the initialization op.
Returns:
An Op that initializes all tables. Note that if there are
not tables the returned Op is a NoOp.
"""
return tables_initializer(name)
@tf_export(v1=["initializers.tables_initializer", "tables_initializer"])
def tables_initializer(name="init_all_tables"):
"""Returns an Op that initializes all tables of the default graph.
See the [Low Level
Intro](https://www.tensorflow.org/guide/low_level_intro#feature_columns)
guide, for an example of usage.
Args:
name: Optional name for the initialization op.
Returns:
An Op that initializes all tables. Note that if there are
not tables the returned Op is a NoOp.
"""
initializers = ops.get_collection(ops.GraphKeys.TABLE_INITIALIZERS)
if initializers:
return control_flow_ops.group(*initializers, name=name)
return control_flow_ops.no_op(name=name)
def _check_table_dtypes(table, key_dtype, value_dtype):
"""Check that the given key_dtype and value_dtype matches the table dtypes.
Args:
table: The table to check types against to.
key_dtype: The key data type to check.
value_dtype: The value data type to check.
Raises:
TypeError: when 'key_dtype' or 'value_dtype' doesn't match the table data
types.
"""
if key_dtype.base_dtype != table.key_dtype:
raise TypeError("Invalid key dtype, expected %s but got %s." %
(table.key_dtype, key_dtype))
if value_dtype.base_dtype != table.value_dtype:
raise TypeError("Invalid value dtype, expected %s but got %s." %
(table.value_dtype, value_dtype))
class LookupInterface(trackable.TrackableResource):
"""Represent a lookup table that persists across different steps."""
def __init__(self, key_dtype, value_dtype):
"""Construct a lookup table interface.
Args:
key_dtype: The table key type.
value_dtype: The table value type.
"""
self._key_dtype = dtypes.as_dtype(key_dtype)
self._value_dtype = dtypes.as_dtype(value_dtype)
super(LookupInterface, self).__init__()
def _create_resource(self):
raise NotImplementedError
@property
def key_dtype(self):
"""The table key dtype."""
return self._key_dtype
@property
def value_dtype(self):
"""The table value dtype."""
return self._value_dtype
@property
def name(self):
"""The name of the table."""
return NotImplementedError
def size(self, name=None):
"""Compute the number of elements in this table."""
raise NotImplementedError
def lookup(self, keys, name=None):
"""Looks up `keys` in a table, outputs the corresponding values."""
raise NotImplementedError
class InitializableLookupTableBase(LookupInterface):
"""Initializable lookup table interface.
An initializable lookup tables persist across different steps.
"""
def __init__(self, default_value, initializer):
"""Construct a table object from a table reference.
If requires a table initializer object (subclass of `TableInitializerBase`).
It provides the table key and value types, as well as the op to initialize
the table. The caller is responsible to execute the initialization op.
Args:
default_value: The value to use if a key is missing in the table.
initializer: The table initializer to use.
"""
super(InitializableLookupTableBase, self).__init__(initializer.key_dtype,
initializer.value_dtype)
self._default_value = ops.convert_to_tensor(
default_value, dtype=self._value_dtype)
self._default_value.get_shape().merge_with(tensor_shape.scalar())
if isinstance(initializer, trackable_base.Trackable):
self._initializer = self._track_trackable(initializer, "_initializer")
with ops.init_scope():
self._resource_handle = self._create_resource()
self._init_op = self._initialize()
def _initialize(self):
return self._initializer.initialize(self)
@property
def default_value(self):
"""The default value of the table."""
return self._default_value
def size(self, name=None):
"""Compute the number of elements in this table.
Args:
name: A name for the operation (optional).
Returns:
A scalar tensor containing the number of elements in this table.
"""
with ops.name_scope(name, "%s_Size" % self.name, [self.resource_handle]):
return gen_lookup_ops.lookup_table_size_v2(self.resource_handle)
def lookup(self, keys, name=None):
"""Looks up `keys` in a table, outputs the corresponding values.
The `default_value` is used for keys not present in the table.
Args:
keys: Keys to look up. May be either a `SparseTensor` or dense `Tensor`.
name: A name for the operation (optional).
Returns:
A `SparseTensor` if keys are sparse, otherwise a dense `Tensor`.
Raises:
TypeError: when `keys` or `default_value` doesn't match the table data
types.
"""
key_tensor = keys
if isinstance(keys, sparse_tensor.SparseTensor):
key_tensor = keys.values
if keys.dtype.base_dtype != self._key_dtype:
raise TypeError("Signature mismatch. Keys must be dtype %s, got %s." %
(self._key_dtype, keys.dtype))
with ops.name_scope(
name, "%s_Lookup" % self.name,
(self.resource_handle, key_tensor, self._default_value)):
values = gen_lookup_ops.lookup_table_find_v2(self.resource_handle,
key_tensor,
self._default_value)
values.set_shape(key_tensor.get_shape())
if isinstance(keys, sparse_tensor.SparseTensor):
return sparse_tensor.SparseTensor(keys.indices, values, keys.dense_shape)
else:
return values
class InitializableLookupTableBaseV1(InitializableLookupTableBase):
@property
def initializer(self):
return self._init_op
@tf_export("lookup.StaticHashTable", v1=[])
class StaticHashTable(InitializableLookupTableBase):
"""A generic hash table implementation.
Example usage:
```python
table = tf.lookup.StaticHashTable(
tf.KeyValueTensorInitializer(keys, values), -1)
out = table.lookup(input_tensor)
table.init.run()
print(out.eval())
```
"""
def __init__(self, initializer, default_value, name=None):
"""Creates a non-initialized `HashTable` object.
Creates a table, the type of its keys and values are specified by the
initializer.
Before using the table you will have to initialize it. After initialization
the table will be immutable.
Args:
initializer: The table initializer to use. See `HashTable` kernel for
supported key and value types.
default_value: The value to use if a key is missing in the table.
name: A name for the operation (optional).
Returns:
A `HashTable` object.
"""
self._initializer = initializer
self._default_value = default_value
self._shared_name = self._initializer._shared_name # pylint: disable=protected-access
if not self._shared_name:
# Force using a shared name so that StaticHashTable resources can be
# shared across different kernels. If no "shared_name" is set and
# "use_node_name_sharing" is False, then each kernel gets its own local
# resource.
self._shared_name = "hash_table_%s" % (str(uuid.uuid4()),)
self._name = name or "hash_table"
self._table_name = None
super(StaticHashTable, self).__init__(default_value, initializer)
self._value_shape = self._default_value.get_shape()
def _create_resource(self):
table_ref = gen_lookup_ops.hash_table_v2(
shared_name=self._shared_name,
key_dtype=self._initializer.key_dtype,
value_dtype=self._initializer.value_dtype,
name=self._name)
if context.executing_eagerly():
self._table_name = None
else:
self._table_name = table_ref.op.name.split("/")[-1]
return table_ref
@property
def name(self):
return self._table_name
def export(self, name=None):
"""Returns tensors of all keys and values in the table.
Args:
name: A name for the operation (optional).
Returns:
A pair of tensors with the first tensor containing all keys and the
second tensors containing all values in the table.
"""
with ops.name_scope(name, "%s_Export" % self.name, [self.resource_handle]):
exported_keys, exported_values = gen_lookup_ops.lookup_table_export_v2(
self.resource_handle, self._key_dtype, self._value_dtype)
exported_values.set_shape(exported_keys.get_shape().concatenate(
self._value_shape))
return exported_keys, exported_values
@tf_export(v1=["lookup.StaticHashTable"])
class StaticHashTableV1(StaticHashTable):
@property
def initializer(self):
return self._init_op
# For backwards compatibility. This will be removed in TF 2.0.
class HashTable(StaticHashTableV1):
@property
def init(self):
return self.initializer
class TableInitializerBase(trackable_base.Trackable):
"""Base class for lookup table initializers."""
def __init__(self, key_dtype, value_dtype):
"""Construct a table initializer object.
Args:
key_dtype: Type of the table keys.
value_dtype: Type of the table values.
"""
self._key_dtype = dtypes.as_dtype(key_dtype)
self._value_dtype = dtypes.as_dtype(value_dtype)
@property
def key_dtype(self):
"""The expected table key dtype."""
return self._key_dtype
@property
def value_dtype(self):
"""The expected table value dtype."""
return self._value_dtype
def initialize(self, table):
"""Returns the table initialization op."""
raise NotImplementedError
@property
def _shared_name(self):
"""Returns a shared name to be used by the table."""
shared_name = ""
if context.executing_eagerly():
# Ensure a unique name when eager execution is enabled to avoid spurious
# sharing issues.
# TODO(rohanj): Use context.shared_name() instead.
shared_name += str(ops.uid())
return shared_name
@tf_export("lookup.KeyValueTensorInitializer")
class KeyValueTensorInitializer(TableInitializerBase):
"""Table initializers given `keys` and `values` tensors."""
def __init__(self, keys, values, key_dtype=None, value_dtype=None, name=None):
"""Constructs a table initializer object based on keys and values tensors.
Args:
keys: The tensor for the keys.
values: The tensor for the values.
key_dtype: The `keys` data type. Used when `keys` is a python array.
value_dtype: The `values` data type. Used when `values` is a python array.
name: A name for the operation (optional).
"""
with ops.init_scope():
self._keys = ops.convert_to_tensor(keys, dtype=key_dtype, name="keys")
self._values = ops.convert_to_tensor(
values, dtype=value_dtype, name="values")
self._name = name if name is not None else "key_value_init"
if context.executing_eagerly():
# Ensure a unique name when eager execution is enabled to avoid spurious
# sharing issues.
# TODO(rohanj): Use context.shared_name() instead.
self._name += str(ops.uid())
super(KeyValueTensorInitializer, self).__init__(self._keys.dtype,
self._values.dtype)
def initialize(self, table):
"""Initializes the given `table` with `keys` and `values` tensors.
Args:
table: The table to initialize.
Returns:
The operation that initializes the table.
Raises:
TypeError: when the keys and values data types do not match the table
key and value data types.
"""
_check_table_dtypes(table, self._keys.dtype, self._values.dtype)
with ops.name_scope(
self._name, values=(table.resource_handle, self._keys, self._values)):
if fwd_compat.forward_compatible(2018, 9, 19):
init_op = gen_lookup_ops.lookup_table_import_v2(table.resource_handle,
self._keys,
self._values)
else:
# To maintain forward compatibiltiy, use the old implementation.
init_op = gen_lookup_ops.initialize_table_v2(table.resource_handle,
self._keys, self._values)
ops.add_to_collection(ops.GraphKeys.TABLE_INITIALIZERS, init_op)
return init_op
@tf_export("lookup.TextFileIndex")
class TextFileIndex(object):
"""The key and value content to get from each line.
This class defines the key and value used for tf.lookup.TextFileInitializer.
The key and value content to get from each line is specified either
by the following, or a value `>=0`.
* `TextFileIndex.LINE_NUMBER` means use the line number starting from zero,
expects data type int64.
* `TextFileIndex.WHOLE_LINE` means use the whole line content, expects data
type string.
A value `>=0` means use the index (starting at zero) of the split line based
on `delimiter`.
"""
WHOLE_LINE = -2
LINE_NUMBER = -1
@tf_export("lookup.TextFileInitializer")
class TextFileInitializer(TableInitializerBase):
"""Table initializers from a text file.
This initializer assigns one entry in the table for each line in the file.
The key and value type of the table to initialize is given by `key_dtype` and
`value_dtype`.
The key and value content to get from each line is specified by
the `key_index` and `value_index`.
* `TextFileIndex.LINE_NUMBER` means use the line number starting from zero,
expects data type int64.
* `TextFileIndex.WHOLE_LINE` means use the whole line content, expects data
type string.
* A value `>=0` means use the index (starting at zero) of the split line based
on `delimiter`.
For example if we have a file with the following content:
```
emerson 10
lake 20
palmer 30
```
The following snippet initializes a table with the first column as keys and
second column as values:
* `emerson -> 10`
* `lake -> 20`
* `palmer -> 30`
```python
table = tf.lookup.StaticHashTable(tf.lookup.TextFileInitializer(
"test.txt", tf.string, 0, tf.int64, 1, delimiter=" "), -1)
...
table.init.run()
```
Similarly to initialize the whole line as keys and the line number as values.
* `emerson 10 -> 0`
* `lake 20 -> 1`
* `palmer 30 -> 2`
```python
table = tf.lookup.StaticHashTable(tf.lookup.TextFileInitializer(
"test.txt", tf.string, tf.lookup.TextFileIndex.WHOLE_LINE,
tf.int64, tf.lookup.TextFileIndex.LINE_NUMBER, delimiter=" "), -1)
...
table.init.run()
```
"""
def __init__(self,
filename,
key_dtype,
key_index,
value_dtype,
value_index,
vocab_size=None,
delimiter="\t",
name=None):
"""Constructs a table initializer object to populate from a text file.
It generates one key-value pair per line. The type of table key and
value are specified by `key_dtype` and `value_dtype`, respectively.
Similarly the content of the key and value are specified by the key_index
and value_index.
- TextFileIndex.LINE_NUMBER means use the line number starting from zero,
expects data type int64.
- TextFileIndex.WHOLE_LINE means use the whole line content, expects data
type string.
- A value >=0 means use the index (starting at zero) of the split line based
on `delimiter`.
Args:
filename: The filename of the text file to be used for initialization. The
path must be accessible from wherever the graph is initialized (eg.
trainer or eval workers). The filename may be a scalar `Tensor`.
key_dtype: The `key` data type.
key_index: the index that represents information of a line to get the
table 'key' values from.
value_dtype: The `value` data type.
value_index: the index that represents information of a line to get the
table 'value' values from.'
vocab_size: The number of elements in the file, if known.
delimiter: The delimiter to separate fields in a line.
name: A name for the operation (optional).
Raises:
ValueError: when the filename is empty, or when the table key and value
data types do not match the expected data types.
"""
if not isinstance(filename, ops.Tensor) and not filename:
raise ValueError("Filename required for %s." % name)
self._filename_arg = filename
key_dtype = dtypes.as_dtype(key_dtype)
value_dtype = dtypes.as_dtype(value_dtype)
if key_index < -2:
raise ValueError("Invalid key index %s." % (key_index))
if key_index == TextFileIndex.LINE_NUMBER and key_dtype != dtypes.int64:
raise ValueError("Signature mismatch. Keys must be dtype %s, got %s." %
(dtypes.int64, key_dtype))
if ((key_index == TextFileIndex.WHOLE_LINE) and
(not key_dtype.is_integer) and (key_dtype != dtypes.string)):
raise ValueError(
"Signature mismatch. Keys must be integer or string, got %s." %
key_dtype)
if value_index < -2:
raise ValueError("Invalid value index %s." % (value_index))
if value_index == TextFileIndex.LINE_NUMBER and value_dtype != dtypes.int64:
raise ValueError("Signature mismatch. Values must be dtype %s, got %s." %
(dtypes.int64, value_dtype))
if value_index == TextFileIndex.WHOLE_LINE and value_dtype != dtypes.string:
raise ValueError("Signature mismatch. Values must be dtype %s, got %s." %
(dtypes.string, value_dtype))
if (vocab_size is not None) and (vocab_size <= 0):
raise ValueError("Invalid vocab_size %s." % vocab_size)
self._key_index = key_index
self._value_index = value_index
self._vocab_size = vocab_size
self._delimiter = delimiter
self._name = name
self._filename = self._track_trackable(
trackable.TrackableAsset(filename), "_filename")
super(TextFileInitializer, self).__init__(key_dtype, value_dtype)
def initialize(self, table):
"""Initializes the table from a text file.
Args:
table: The table to be initialized.
Returns:
The operation that initializes the table.
Raises:
TypeError: when the keys and values data types do not match the table
key and value data types.
"""
_check_table_dtypes(table, self.key_dtype, self.value_dtype)
with ops.name_scope(self._name, "text_file_init", (table.resource_handle,)):
filename = ops.convert_to_tensor(
self._filename, dtypes.string, name="asset_filepath")
init_op = gen_lookup_ops.initialize_table_from_text_file_v2(
table.resource_handle, filename, self._key_index, self._value_index,
-1 if self._vocab_size is None else self._vocab_size, self._delimiter)
ops.add_to_collection(ops.GraphKeys.TABLE_INITIALIZERS, init_op)
# If the filename tensor is anything other than a string constant (e.g.,
# if it is a placeholder) then it does not make sense to track it as an
# asset.
if not context.executing_eagerly() and constant_op.is_constant(filename):
ops.add_to_collection(ops.GraphKeys.ASSET_FILEPATHS, filename)
return init_op
@property
def _shared_name(self):
if self._vocab_size:
# Keep the shared_name:
# <table_type>_<filename>_<vocab_size>_<key_index>_<value_index>
shared_name = "hash_table_%s_%d_%s_%s" % (
self._filename_arg, self._vocab_size, self._key_index,
self._value_index)
else:
# Keep the shared_name
# <table_type>_<filename>_<key_index>_<value_index>
shared_name = "hash_table_%s_%s_%s" % (self._filename_arg,
self._key_index, self._value_index)
return shared_name
class TextFileStringTableInitializer(TextFileInitializer):
"""Table initializer for `int64` IDs to string tables from a text file."""
def __init__(self,
filename,
key_column_index=TextFileIndex.LINE_NUMBER,
value_column_index=TextFileIndex.WHOLE_LINE,
vocab_size=None,
delimiter="\t",
name="text_file_string_table_init"):
"""Constructs an initializer for an id-to-string table from a text file.
It populates a table that its key and value types are int64 and string,
respectively. It generates one key-value pair per line.
The content of the key and value are specified by `key_column_index`
and `value_column_index`.
- TextFileIndex.LINE_NUMBER means use the line number starting from zero,
expects data type int64.
- TextFileIndex.WHOLE_LINE means use the whole line content, expects data
type string.
- A value >=0 means use the index (starting at zero) of the split line based
on `delimiter`.
Args:
filename: The filename of the text file to be used for initialization. The
path must be accessible from wherever the graph is initialized (eg.
trainer or eval workers). The filename may be a scalar `Tensor`.
key_column_index: The column index from the text file to get the keys
from. The default is to use the line number, starting from zero.
value_column_index: The column index from the text file to get the values
from. The default is to use the whole line content.
vocab_size: The number of elements in the file, if known.
delimiter: The delimiter to separate fields in a line.
name: Optional name for the op.
Raises:
TypeError: when the filename is empty, or when the table key and value
data types do not match the expected data types.
"""
super(TextFileStringTableInitializer, self).__init__(
filename,
dtypes.int64,
key_column_index,
dtypes.string,
value_column_index,
vocab_size=vocab_size,
delimiter=delimiter,
name=name)
class TextFileIdTableInitializer(TextFileInitializer):
"""Table initializer for string to `int64` IDs tables from a text file."""
def __init__(self,
filename,
key_column_index=TextFileIndex.WHOLE_LINE,
value_column_index=TextFileIndex.LINE_NUMBER,
vocab_size=None,
delimiter="\t",
name="text_file_id_table_init",
key_dtype=dtypes.string):
"""Constructs an initializer for an string-to-id table from a text file.
It populates a table that its key and value types are string and int64,
respectively. It generates one key-value pair per line.
The content of the key and value are specified by the key_index
and value_index.
- TextFileIndex.LINE_NUMBER means use the line number starting from zero,
expects data type int64.
- TextFileIndex.WHOLE_LINE means use the whole line content, expects data
type string.
- A value >=0 means use the index (starting at zero) of the split line based
on `delimiter`.
Args:
filename: The filename of the text file to be used for initialization. The
path must be accessible from wherever the graph is initialized (eg.
trainer or eval workers). The filename may be a scalar `Tensor`.
key_column_index: The column index from the text file to get the `key`
values from. The default is to use the whole line content.
value_column_index: The column index from the text file to get the `value`
values from. The default is to use the line number, starting from zero.
vocab_size: The number of elements in the file, if known.
delimiter: The delimiter to separate fields in a line.
name: Optional name for the op.
key_dtype: The `key` data type.
Raises:
TypeError: when the filename is empty, or when the table key and value
data types do not match the expected data types.
"""
super(TextFileIdTableInitializer, self).__init__(
filename,
key_dtype,
key_column_index,
dtypes.int64,
value_column_index,
vocab_size=vocab_size,
delimiter=delimiter,
name=name)
class HasherSpec(collections.namedtuple("HasherSpec", ["hasher", "key"])):
"""A structure for the spec of the hashing function to use for hash buckets.
`hasher` is the name of the hashing function to use (eg. "fasthash",
"stronghash").
`key` is optional and specify the key to use for the hash function if
supported, currently only used by a strong hash.
Fields:
hasher: The hasher name to use.
key: The key to be used by the hashing function, if required.
"""
__slots__ = ()
FastHashSpec = HasherSpec("fasthash", None) # pylint: disable=invalid-name
class StrongHashSpec(HasherSpec):
"""A structure to specify a key of the strong keyed hash spec.
The strong hash requires a `key`, which is a list of 2 unsigned integer
numbers. These should be non-zero; random numbers generated from random.org
would be a fine choice.
Fields:
key: The key to be used by the keyed hashing function.
"""
__slots__ = ()
def __new__(cls, key):
if len(key) != 2:
raise ValueError("key must have size 2, got %s." % len(key))
if not isinstance(key[0], compat.integral_types) or not isinstance(
key[1], compat.integral_types):
raise TypeError("Invalid key %s. Must be unsigned integer values." % key)
return super(cls, StrongHashSpec).__new__(cls, "stronghash", key)
def _as_string(tensor):
if dtypes.string == tensor.dtype.base_dtype:
return tensor
return string_ops.as_string(tensor)
class IdTableWithHashBuckets(LookupInterface):
"""String to Id table wrapper that assigns out-of-vocabulary keys to buckets.
For example, if an instance of `IdTableWithHashBuckets` is initialized with a
string-to-id table that maps:
* `emerson -> 0`
* `lake -> 1`
* `palmer -> 2`
The `IdTableWithHashBuckets` object will performs the following mapping:
* `emerson -> 0`
* `lake -> 1`
* `palmer -> 2`
* `<other term> -> bucket_id`, where bucket_id will be between `3` and
`3 + num_oov_buckets - 1`, calculated by:
`hash(<term>) % num_oov_buckets + vocab_size`
If input_tensor is `["emerson", "lake", "palmer", "king", "crimson"]`,
the lookup result is `[0, 1, 2, 4, 7]`.
If `table` is None, only out-of-vocabulary buckets are used.
Example usage:
```python
num_oov_buckets = 3
input_tensor = tf.constant(["emerson", "lake", "palmer", "king", "crimnson"])
table = tf.IdTableWithHashBuckets(
tf.StaticHashTable(tf.TextFileIdTableInitializer(filename),
default_value),
num_oov_buckets)
out = table.lookup(input_tensor).
table.init.run()
print(out.eval())
```
The hash function used for generating out-of-vocabulary buckets ID is handled
by `hasher_spec`.
"""
def __init__(self,
table,
num_oov_buckets,
hasher_spec=FastHashSpec,
name=None,
key_dtype=None):
"""Construct a `IdTableWithHashBuckets` object.
Args:
table: Table that maps `tf.string` or `tf.int64` keys to `tf.int64` ids.
num_oov_buckets: Number of buckets to use for out-of-vocabulary keys.
hasher_spec: A `HasherSpec` to specify the hash function to use for
assignation of out-of-vocabulary buckets (optional).
name: A name for the operation (optional).
key_dtype: Data type of keys passed to `lookup`. Defaults to
`table.key_dtype` if `table` is specified, otherwise `tf.string`. Must
be string or integer, and must be castable to `table.key_dtype`.
Raises:
ValueError: when `table` in None and `num_oov_buckets` is not positive.
TypeError: when `hasher_spec` is invalid.
"""
# If a name ends with a '/' it is a "name scope", remove all trailing '/'
# characters to use as table name.
if name:
name = name.rstrip("/")
if table:
if key_dtype is None:
key_dtype = table.key_dtype
supported_table_key_dtypes = (dtypes.int64, dtypes.string)
if table.key_dtype not in supported_table_key_dtypes:
raise TypeError("Invalid key dtype, expected one of %s, but got %s." %
(supported_table_key_dtypes, key_dtype))
if table.key_dtype.is_integer != key_dtype.is_integer:
raise TypeError("Invalid key dtype, expected %s but got %s." %
("integer" if key_dtype.is_integer else "non-integer",
table.key_dtype))
if table.value_dtype != dtypes.int64:
raise TypeError("Invalid value dtype, expected %s but got %s." %
(dtypes.int64, table.value_dtype))
self._table = table
name = name or self._table.name
else:
if num_oov_buckets <= 0:
raise ValueError("oov_buckets must be > 0 if no table is supplied.")
key_dtype = dtypes.string if key_dtype is None else key_dtype
self._table = None
name = name or "hash_bucket"
if (not key_dtype.is_integer) and (dtypes.string != key_dtype):
raise TypeError("Invalid key_dtype, expected integer or string, got %s." %
key_dtype)
self._num_oov_buckets = num_oov_buckets
if not isinstance(hasher_spec, HasherSpec):
raise TypeError("hasher_spec must be of type HasherSpec, got %s" %
hasher_spec)
self._hasher_spec = hasher_spec
if name:
self._table_name = name.split("/")[-1]
else:
self._table_name = None
super(IdTableWithHashBuckets, self).__init__(key_dtype, dtypes.int64)
def _create_resource(self):
if self._table is not None:
return self._table._create_resource() # pylint: disable=protected-access
return None
def _initialize(self):
if self._table is not None:
return self._table._initialize() # pylint: disable=protected-access
with ops.name_scope(None, "init"):
return control_flow_ops.no_op()
@property
def initializer(self):
if self._table is not None:
return self._table._init_op # pylint: disable=protected-access
with ops.name_scope(None, "init"):
return control_flow_ops.no_op()
@property
@deprecated("2018-12-15", "Use `initializer` instead.")
def init(self):
return self.initializer
@property
def resource_handle(self):
if self._table is not None:
return self._table.resource_handle
return None
@property
def name(self):
return self._table_name
def size(self, name=None):
"""Compute the number of elements in this table."""
with ops.name_scope(name, "%s_Size" % self.name):
if self._table:
tsize = self._table.size()
else:
tsize = ops.convert_to_tensor(0, dtype=dtypes.int64)
return tsize + self._num_oov_buckets
def _get_string_to_hash_bucket_fn(self, hasher_spec):
"""Returns the string_to_hash_bucket op to use based on `hasher_spec`."""
if not isinstance(hasher_spec, HasherSpec):
raise TypeError("hasher_spec must be of type HasherSpec %s" % hasher_spec)
if hasher_spec.hasher == "fasthash":
return string_ops.string_to_hash_bucket_fast
if hasher_spec.hasher == "legacy":
return string_ops.string_to_hash_bucket
if hasher_spec.hasher == "stronghash":
return functools.partial(
string_ops.string_to_hash_bucket_strong, key=hasher_spec.key)
raise ValueError("Unknown hasher %s" % hasher_spec.hasher)
def lookup(self, keys, name=None):
"""Looks up `keys` in the table, outputs the corresponding values.
It assigns out-of-vocabulary keys to buckets based in their hashes.
Args:
keys: Keys to look up. May be either a `SparseTensor` or dense `Tensor`.
name: Optional name for the op.
Returns:
A `SparseTensor` if keys are sparse, otherwise a dense `Tensor`.
Raises:
TypeError: when `keys` doesn't match the table key data type.
"""
if keys.dtype.base_dtype != self._key_dtype:
raise TypeError("Signature mismatch. Keys must be dtype %s, got %s." %
(self._key_dtype, keys.dtype))
values = keys
if isinstance(keys, sparse_tensor.SparseTensor):
values = keys.values
if self._table and (self._table.key_dtype.base_dtype == dtypes.int64):
values = math_ops.cast(values, dtypes.int64)
if self._num_oov_buckets == 0:
ids = self._table.lookup(values, name=name)
else:
# TODO(yleon): Consider moving this functionality to its own kernel.
with ops.name_scope(name, "%s_Lookup" % self.name):
str_to_hash_bucket = self._get_string_to_hash_bucket_fn(
self._hasher_spec)
buckets = str_to_hash_bucket(
_as_string(values),
num_buckets=self._num_oov_buckets,
name="hash_bucket")
if self._table:
ids = self._table.lookup(values)
buckets = math_ops.add(buckets, self._table.size())
is_id_non_default = math_ops.not_equal(ids, self._table.default_value)
ids = array_ops.where(is_id_non_default, ids, buckets)
else:
ids = buckets
if isinstance(keys, sparse_tensor.SparseTensor):
return sparse_tensor.SparseTensor(keys.indices, ids, keys.dense_shape)
return ids
@tf_export("lookup.StaticVocabularyTable", v1=[])
class StaticVocabularyTable(LookupInterface):
"""String to Id table wrapper that assigns out-of-vocabulary keys to buckets.
For example, if an instance of `StaticVocabularyTable` is initialized with a
string-to-id initializer that maps:
* `emerson -> 0`
* `lake -> 1`
* `palmer -> 2`
The `Vocabulary` object will performs the following mapping:
* `emerson -> 0`
* `lake -> 1`
* `palmer -> 2`
* `<other term> -> bucket_id`, where bucket_id will be between `3` and
`3 + num_oov_buckets - 1`, calculated by:
`hash(<term>) % num_oov_buckets + vocab_size`
If input_tensor is `["emerson", "lake", "palmer", "king", "crimson"]`,
the lookup result is `[0, 1, 2, 4, 7]`.
If `initializer` is None, only out-of-vocabulary buckets are used.
Example usage:
```python
num_oov_buckets = 3
input_tensor = tf.constant(["emerson", "lake", "palmer", "king", "crimnson"])
table = tf.lookup.StaticVocabularyTable(
tf.TextFileIdTableInitializer(filename), num_oov_buckets)
out = table.lookup(input_tensor).
table.init.run()
print(out.eval())
```
The hash function used for generating out-of-vocabulary buckets ID is
Fingerprint64.
"""
def __init__(self,
initializer,
num_oov_buckets,
lookup_key_dtype=None,
name=None):
"""Construct a `StaticVocabularyTable` object.
Args:
initializer: A TableInitializerBase object that contains the data used to
initialize the table. If None, then we only use out-of-vocab buckets.
num_oov_buckets: Number of buckets to use for out-of-vocabulary keys. Must
be greater than zero.
lookup_key_dtype: Data type of keys passed to `lookup`. Defaults to
`initializer.key_dtype` if `initializer` is specified, otherwise
`tf.string`. Must be string or integer, and must be castable to
`initializer.key_dtype`.
name: A name for the operation (optional).
Raises:
ValueError: when `num_oov_buckets` is not positive.
TypeError: when lookup_key_dtype or initializer.key_dtype are not
integer or string. Also when initializer.value_dtype != int64.
"""
if num_oov_buckets <= 0:
raise ValueError("oov_buckets must be > 0.")
# If a name ends with a '/' it is a "name scope", remove all trailing '/'
# characters to use as table name.
if name:
name = name.rstrip("/")
if initializer:
if lookup_key_dtype is None:
lookup_key_dtype = initializer.key_dtype
supported_table_key_dtypes = (dtypes.int64, dtypes.string)
if initializer.key_dtype not in supported_table_key_dtypes:
raise TypeError("Invalid key dtype, expected one of %s, but got %s." %
(supported_table_key_dtypes, initializer.key_dtype))
if initializer.key_dtype.is_integer != lookup_key_dtype.is_integer:
raise TypeError(
"Invalid key dtype, expected %s but got %s." %
("integer" if lookup_key_dtype.is_integer else "non-integer",
initializer.key_dtype))
if initializer.value_dtype != dtypes.int64:
raise TypeError("Invalid value dtype, expected %s but got %s." %
(dtypes.int64, initializer.value_dtype))
self._table = HashTable(initializer, default_value=-1)
name = name or self._table.name
else:
lookup_key_dtype = dtypes.string
self._table = None
name = name or "hash_bucket"
if (not lookup_key_dtype.is_integer) and (dtypes.string !=
lookup_key_dtype):
raise TypeError("Invalid key_dtype, expected integer or string, got %s." %
lookup_key_dtype)
self._num_oov_buckets = num_oov_buckets
self._table_name = None
if name is not None:
self._table_name = name.split("/")[-1]
super(StaticVocabularyTable, self).__init__(lookup_key_dtype, dtypes.int64)
def _create_resource(self):
if self._table is not None:
return self._table._create_resource() # pylint: disable=protected-access
return None
def _initialize(self):
if self._table is not None:
return self._table._initialize() # pylint: disable=protected-access
with ops.name_scope(None, "init"):
return control_flow_ops.no_op()
@property
def resource_handle(self):
if self._table is not None:
return self._table.resource_handle
return None
@property
def name(self):
return self._table_name
def size(self, name=None):
"""Compute the number of elements in this table."""
with ops.name_scope(name, "%s_Size" % self.name):
if self._table:
tsize = self._table.size()
else:
tsize = ops.convert_to_tensor(0, dtype=dtypes.int64)
return tsize + self._num_oov_buckets
def lookup(self, keys, name=None):
"""Looks up `keys` in the table, outputs the corresponding values.
It assigns out-of-vocabulary keys to buckets based in their hashes.
Args:
keys: Keys to look up. May be either a `SparseTensor` or dense `Tensor`.
name: Optional name for the op.
Returns:
A `SparseTensor` if keys are sparse, otherwise a dense `Tensor`.
Raises:
TypeError: when `keys` doesn't match the table key data type.
"""
if keys.dtype.base_dtype != self._key_dtype:
raise TypeError("Signature mismatch. Keys must be dtype %s, got %s." %
(self._key_dtype, keys.dtype))
values = keys
if isinstance(keys, sparse_tensor.SparseTensor):
values = keys.values
if self._table and (self._table.key_dtype.base_dtype == dtypes.int64):
values = math_ops.cast(values, dtypes.int64)
# TODO(yleon): Consider moving this functionality to its own kernel.
with ops.name_scope(name, "%s_Lookup" % self.name):
buckets = string_ops.string_to_hash_bucket_fast(
_as_string(values),
num_buckets=self._num_oov_buckets,
name="hash_bucket")
if self._table:
ids = self._table.lookup(values)
buckets = math_ops.add(buckets, self._table.size())
is_id_non_default = math_ops.not_equal(ids, self._table.default_value)
ids = array_ops.where(is_id_non_default, ids, buckets)
else:
ids = buckets
if isinstance(keys, sparse_tensor.SparseTensor):
return sparse_tensor.SparseTensor(keys.indices, ids, keys.dense_shape)
return ids
@tf_export(v1=["lookup.StaticVocabularyTable"])
class StaticVocabularyTableV1(StaticVocabularyTable):
@property
def initializer(self):
if self._table is not None:
return self._table._init_op # pylint: disable=protected-access
with ops.name_scope(None, "init"):
return control_flow_ops.no_op()
def index_table_from_file(vocabulary_file=None,
num_oov_buckets=0,
vocab_size=None,
default_value=-1,
hasher_spec=FastHashSpec,
key_dtype=dtypes.string,
name=None,
key_column_index=TextFileIndex.WHOLE_LINE,
value_column_index=TextFileIndex.LINE_NUMBER,
delimiter="\t"):
"""Returns a lookup table that converts a string tensor into int64 IDs.
This operation constructs a lookup table to convert tensor of strings into
int64 IDs. The mapping can be initialized from a vocabulary file specified in
`vocabulary_file`, where the whole line is the key and the zero-based line
number is the ID.
Any lookup of an out-of-vocabulary token will return a bucket ID based on its
hash if `num_oov_buckets` is greater than zero. Otherwise it is assigned the
`default_value`.
The bucket ID range is
`[vocabulary size, vocabulary size + num_oov_buckets - 1]`.
The underlying table must be initialized by calling
`session.run(tf.compat.v1.tables_initializer)` or `session.run(table.init)`
once.
To specify multi-column vocabulary files, use key_column_index and
value_column_index and delimiter.
- TextFileIndex.LINE_NUMBER means use the line number starting from zero,
expects data type int64.
- TextFileIndex.WHOLE_LINE means use the whole line content, expects data
type string.
- A value >=0 means use the index (starting at zero) of the split line based
on `delimiter`.
Sample Usages:
If we have a vocabulary file "test.txt" with the following content:
```
emerson
lake
palmer
```
```python
features = tf.constant(["emerson", "lake", "and", "palmer"])
table = tf.lookup.index_table_from_file(
vocabulary_file="test.txt", num_oov_buckets=1)
ids = table.lookup(features)
...
tf.compat.v1.tables_initializer().run()
ids.eval() ==> [0, 1, 3, 2] # where 3 is the out-of-vocabulary bucket
```
Args:
vocabulary_file: The vocabulary filename, may be a constant scalar `Tensor`.
num_oov_buckets: The number of out-of-vocabulary buckets.
vocab_size: Number of the elements in the vocabulary, if known.
default_value: The value to use for out-of-vocabulary feature values.
Defaults to -1.
hasher_spec: A `HasherSpec` to specify the hash function to use for
assignation of out-of-vocabulary buckets.
key_dtype: The `key` data type.
name: A name for this op (optional).
key_column_index: The column index from the text file to get the `key`
values from. The default is to use the whole line content.
value_column_index: The column index from the text file to get the `value`
values from. The default is to use the line number, starting from zero.
delimiter: The delimiter to separate fields in a line.
Returns:
The lookup table to map a `key_dtype` `Tensor` to index `int64` `Tensor`.
Raises:
ValueError: If `vocabulary_file` is not set.
ValueError: If `num_oov_buckets` is negative or `vocab_size` is not greater
than zero.
"""
if vocabulary_file is None or (isinstance(vocabulary_file, six.string_types)
and not vocabulary_file):
raise ValueError("vocabulary_file must be specified and must not be empty.")
if num_oov_buckets < 0:
raise ValueError(
"num_oov_buckets must be greater or equal than 0, got %d." %
num_oov_buckets)
if vocab_size is not None and vocab_size < 1:
vocab_file_value = vocabulary_file
if isinstance(vocabulary_file, ops.Tensor):
vocab_file_value = tensor_util.constant_value(vocabulary_file) or "?"
raise ValueError("vocab_size must be greater than 0, got %d. "
"vocabulary_file: %s" % (vocab_size, vocab_file_value))
if (not key_dtype.is_integer) and (dtypes.string != key_dtype.base_dtype):
raise TypeError("Only integer and string keys are supported.")
with ops.name_scope(name, "string_to_index"):
table = None
with ops.name_scope(None, "hash_table"):
init = TextFileIdTableInitializer(
vocabulary_file,
vocab_size=vocab_size,
key_dtype=dtypes.int64 if key_dtype.is_integer else key_dtype,
name="table_init",
key_column_index=key_column_index,
value_column_index=value_column_index,
delimiter=delimiter)
table = StaticHashTableV1(init, default_value)
if num_oov_buckets:
table = IdTableWithHashBuckets(
table,
num_oov_buckets=num_oov_buckets,
hasher_spec=hasher_spec,
key_dtype=key_dtype)
return table
def index_table_from_tensor(vocabulary_list,
num_oov_buckets=0,
default_value=-1,
hasher_spec=FastHashSpec,
dtype=dtypes.string,
name=None):
"""Returns a lookup table that converts a string tensor into int64 IDs.
This operation constructs a lookup table to convert tensor of strings into
int64 IDs. The mapping can be initialized from a string `vocabulary_list` 1-D
tensor where each element is a key and corresponding index within the tensor
is the value.
Any lookup of an out-of-vocabulary token will return a bucket ID based on its
hash if `num_oov_buckets` is greater than zero. Otherwise it is assigned the
`default_value`. The bucket ID range is
`[vocabulary list size, vocabulary list size + num_oov_buckets - 1]`.
The underlying table must be initialized by calling
`session.run(tf.compat.v1.tables_initializer)` or `session.run(table.init)`
once.
Elements in `vocabulary_list` cannot have duplicates, otherwise when executing
the table initializer op, it will throw a `FailedPreconditionError`.
Sample Usages:
```python
vocabulary_list = tf.constant(["emerson", "lake", "palmer"])
table = tf.lookup.index_table_from_tensor(
vocabulary_list=vocabulary_list, num_oov_buckets=1, default_value=-1)
features = tf.constant(["emerson", "lake", "and", "palmer"])
ids = table.lookup(features)
...
tf.compat.v1.tables_initializer().run()
ids.eval() ==> [0, 1, 4, 2]
```
Args:
vocabulary_list: A 1-D `Tensor` that specifies the mapping of keys to
indices. The type of this object must be castable to `dtype`.
num_oov_buckets: The number of out-of-vocabulary buckets.
default_value: The value to use for out-of-vocabulary feature values.
Defaults to -1.
hasher_spec: A `HasherSpec` to specify the hash function to use for
assignment of out-of-vocabulary buckets.
dtype: The type of values passed to `lookup`. Only string and integers are
supported.
name: A name for this op (optional).
Returns:
The lookup table to map an input `Tensor` to index `int64` `Tensor`.
Raises:
ValueError: If `vocabulary_list` is invalid.
ValueError: If `num_oov_buckets` is negative.
"""
if vocabulary_list is None:
raise ValueError("vocabulary_list must be specified.")
if num_oov_buckets < 0:
raise ValueError(
"num_oov_buckets must be greater or equal than 0, got %d." %
num_oov_buckets)
if (not dtype.is_integer) and (dtypes.string != dtype.base_dtype):
raise TypeError("Only integer and string keys are supported.")
with ops.name_scope(name, "string_to_index"):
keys = ops.convert_to_tensor(vocabulary_list)
if keys.dtype.is_integer != dtype.is_integer:
raise ValueError(
"Expected %s, got %s." %
("integer" if dtype.is_integer else "non-integer", keys.dtype))
if (not dtype.is_integer) and (keys.dtype.base_dtype != dtype):
raise ValueError("Expected %s, got %s." % (dtype, keys.dtype))
num_elements = array_ops.size(keys)
values = math_ops.cast(math_ops.range(num_elements), dtypes.int64)
with ops.name_scope(None, "hash_table"):
table_keys = math_ops.cast(
keys, dtypes.int64) if keys.dtype.is_integer else keys
init = KeyValueTensorInitializer(
table_keys,
values,
table_keys.dtype.base_dtype,
dtypes.int64,
name="table_init")
table = StaticHashTableV1(init, default_value)
if num_oov_buckets:
table = IdTableWithHashBuckets(
table,
num_oov_buckets=num_oov_buckets,
hasher_spec=hasher_spec,
key_dtype=dtype)
return table
def index_to_string_table_from_file(vocabulary_file,
vocab_size=None,
default_value="UNK",
name=None,
key_column_index=TextFileIndex.LINE_NUMBER,
value_column_index=TextFileIndex.WHOLE_LINE,
delimiter="\t"):
"""Returns a lookup table that maps a `Tensor` of indices into strings.
This operation constructs a lookup table to map int64 indices into string
values. The table is initialized from a vocabulary file specified in
`vocabulary_file`, where the whole line is the value and the
zero-based line number is the index.
Any input which does not have a corresponding index in the vocabulary file
(an out-of-vocabulary entry) is assigned the `default_value`
The underlying table must be initialized by calling
`session.run(tf.compat.v1.tables_initializer)` or `session.run(table.init)`
once.
To specify multi-column vocabulary files, use key_column_index and
value_column_index and delimiter.
- TextFileIndex.LINE_NUMBER means use the line number starting from zero,
expects data type int64.
- TextFileIndex.WHOLE_LINE means use the whole line content, expects data
type string.
- A value >=0 means use the index (starting at zero) of the split line based
on `delimiter`.
Sample Usages:
If we have a vocabulary file "test.txt" with the following content:
```
emerson
lake
palmer
```
```python
indices = tf.constant([1, 5], tf.int64)
table = tf.lookup.index_to_string_table_from_file(
vocabulary_file="test.txt", default_value="UNKNOWN")
values = table.lookup(indices)
...
tf.compat.v1.tables_initializer().run()
values.eval() ==> ["lake", "UNKNOWN"]
```
Args:
vocabulary_file: The vocabulary filename, may be a constant scalar `Tensor`.
vocab_size: Number of the elements in the vocabulary, if known.
default_value: The value to use for out-of-vocabulary indices.
name: A name for this op (optional).
key_column_index: The column index from the text file to get the `key`
values from. The default is to use the line number, starting from zero.
value_column_index: The column index from the text file to get the `value`
values from. The default is to use the whole line content.
delimiter: The delimiter to separate fields in a line.
Returns:
The lookup table to map a string values associated to a given index `int64`
`Tensors`.
Raises:
ValueError: when `vocabulary_file` is empty.
ValueError: when `vocab_size` is invalid.
"""
if vocabulary_file is None or (isinstance(vocabulary_file, six.string_types)
and not vocabulary_file):
raise ValueError("vocabulary_file must be specified and must not be empty.")
if vocab_size is not None and vocab_size < 1:
raise ValueError("vocab_size must be greater than 0, got %d." % vocab_size)
with ops.name_scope(name, "index_to_string"):
init = TextFileStringTableInitializer(
vocabulary_file,
vocab_size=vocab_size,
name="table_init",
key_column_index=key_column_index,
value_column_index=value_column_index,
delimiter=delimiter)
# TODO(yleon): Use a more effienct structure.
return StaticHashTableV1(init, default_value)
def index_to_string_table_from_tensor(vocabulary_list,
default_value="UNK",
name=None):
"""Returns a lookup table that maps a `Tensor` of indices into strings.
This operation constructs a lookup table to map int64 indices into string
values. The mapping is initialized from a string `vocabulary_list` 1-D
`Tensor` where each element is a value and the corresponding index within the
tensor is the key.
Any input which does not have a corresponding index in 'vocabulary_list'
(an out-of-vocabulary entry) is assigned the `default_value`
The underlying table must be initialized by calling
`session.run(tf.compat.v1.tables_initializer)` or `session.run(table.init)`
once.
Elements in `vocabulary_list` cannot have duplicates, otherwise when executing
the table initializer op, it will throw a `FailedPreconditionError`.
Sample Usages:
```python
vocabulary_list = tf.constant(["emerson", "lake", "palmer"])
indices = tf.constant([1, 5], tf.int64)
table = tf.lookup.index_to_string_table_from_tensor(
vocabulary_list, default_value="UNKNOWN")
values = table.lookup(indices)
...
tf.compat.v1.tables_initializer().run()
values.eval() ==> ["lake", "UNKNOWN"]
```
Args:
vocabulary_list: A 1-D string `Tensor` that specifies the strings to map
from indices.
default_value: The value to use for out-of-vocabulary indices.
name: A name for this op (optional).
Returns:
The lookup table to map a string values associated to a given index `int64`
`Tensors`.
Raises:
ValueError: when `vocabulary_list` is not set.
"""
if vocabulary_list is None:
raise ValueError("vocabulary_list must be specified.")
with ops.name_scope(name, "index_to_string"):
vocabulary_list = ops.convert_to_tensor(vocabulary_list, dtypes.string)
num_elements = array_ops.size(vocabulary_list)
keys = math_ops.cast(math_ops.range(num_elements), dtypes.int64)
init = KeyValueTensorInitializer(
keys, vocabulary_list, dtypes.int64, dtypes.string, name="table_init")
# TODO(yleon): Use a more effienct structure.
return StaticHashTableV1(init, default_value)
class MutableHashTable(LookupInterface):
"""A generic mutable hash table implementation.
Data can be inserted by calling the insert method and removed by calling the
remove method. It does not support initialization via the init method.
Example usage:
```python
table = tf.lookup.MutableHashTable(key_dtype=tf.string, value_dtype=tf.int64,
default_value=-1)
sess.run(table.insert(keys, values))
out = table.lookup(query_keys)
print(out.eval())
```
"""
def __init__(self,
key_dtype,
value_dtype,
default_value,
name="MutableHashTable",
checkpoint=True):
"""Creates an empty `MutableHashTable` object.
Creates a table, the type of its keys and values are specified by key_dtype
and value_dtype, respectively.
Args:
key_dtype: the type of the key tensors.
value_dtype: the type of the value tensors.
default_value: The value to use if a key is missing in the table.
name: A name for the operation (optional).
checkpoint: if True, the contents of the table are saved to and restored
from checkpoints. If `shared_name` is empty for a checkpointed table, it
is shared using the table node name.
Returns:
A `MutableHashTable` object.
Raises:
ValueError: If checkpoint is True and no name was specified.
"""
self._default_value = ops.convert_to_tensor(
default_value, dtype=value_dtype)
self._value_shape = self._default_value.get_shape()
self._checkpoint = checkpoint
self._key_dtype = key_dtype
self._value_dtype = value_dtype
self._name = name
self._shared_name = None
if context.executing_eagerly():
# TODO(allenl): This will leak memory due to kernel caching by the
# shared_name attribute value (but is better than the alternative of
# sharing everything by default when executing eagerly; hopefully creating
# tables in a loop is uncommon).
# TODO(rohanj): Use context.shared_name() instead.
self._shared_name = "table_%d" % (ops.uid(),)
super(MutableHashTable, self).__init__(key_dtype, value_dtype)
self._resource_handle = self._create_resource()
if checkpoint:
saveable = MutableHashTable._Saveable(self, name)
if not context.executing_eagerly():
ops.add_to_collection(ops.GraphKeys.SAVEABLE_OBJECTS, saveable)
def _create_resource(self):
# The table must be shared if checkpointing is requested for multi-worker
# training to work correctly. Use the node name if no shared_name has been
# explicitly specified.
use_node_name_sharing = self._checkpoint and self._shared_name is None
if self._default_value.get_shape().ndims == 0:
table_ref = gen_lookup_ops.mutable_hash_table_v2(
shared_name=self._shared_name,
use_node_name_sharing=use_node_name_sharing,
key_dtype=self._key_dtype,
value_dtype=self._value_dtype,
name=self._name)
else:
table_ref = gen_lookup_ops.mutable_hash_table_of_tensors_v2(
shared_name=self._shared_name,
use_node_name_sharing=use_node_name_sharing,
key_dtype=self._key_dtype,
value_dtype=self._value_dtype,
value_shape=self._default_value.get_shape(),
name=self._name)
if context.executing_eagerly():
self._table_name = None
else:
self._table_name = table_ref.op.name.split("/")[-1]
return table_ref
@property
def name(self):
return self._table_name
def size(self, name=None):
"""Compute the number of elements in this table.
Args:
name: A name for the operation (optional).
Returns:
A scalar tensor containing the number of elements in this table.
"""
with ops.name_scope(name, "%s_Size" % self.name, [self.resource_handle]):
with ops.colocate_with(self.resource_handle):
return gen_lookup_ops.lookup_table_size_v2(self.resource_handle)
def remove(self, keys, name=None):
"""Removes `keys` and its associated values from the table.
If a key is not present in the table, it is silently ignored.
Args:
keys: Keys to remove. Can be a tensor of any shape. Must match the table's
key type.
name: A name for the operation (optional).
Returns:
The created Operation.
Raises:
TypeError: when `keys` do not match the table data types.
"""
if keys.dtype != self._key_dtype:
raise TypeError("Signature mismatch. Keys must be dtype %s, got %s." %
(self._key_dtype, keys.dtype))
with ops.name_scope(name, "%s_lookup_table_remove" % self.name,
(self.resource_handle, keys, self._default_value)):
op = gen_lookup_ops.lookup_table_remove_v2(self.resource_handle, keys)
return op
def lookup(self, keys, name=None):
"""Looks up `keys` in a table, outputs the corresponding values.
The `default_value` is used for keys not present in the table.
Args:
keys: Keys to look up. Can be a tensor of any shape. Must match the
table's key_dtype.
name: A name for the operation (optional).
Returns:
A tensor containing the values in the same shape as `keys` using the
table's value type.
Raises:
TypeError: when `keys` do not match the table data types.
"""
with ops.name_scope(name, "%s_lookup_table_find" % self.name,
(self.resource_handle, keys, self._default_value)):
keys = ops.convert_to_tensor(keys, dtype=self._key_dtype, name="keys")
with ops.colocate_with(self.resource_handle):
values = gen_lookup_ops.lookup_table_find_v2(self.resource_handle, keys,
self._default_value)
return values
def insert(self, keys, values, name=None):
"""Associates `keys` with `values`.
Args:
keys: Keys to insert. Can be a tensor of any shape. Must match the table's
key type.
values: Values to be associated with keys. Must be a tensor of the same
shape as `keys` and match the table's value type.
name: A name for the operation (optional).
Returns:
The created Operation.
Raises:
TypeError: when `keys` or `values` doesn't match the table data
types.
"""
with ops.name_scope(name, "%s_lookup_table_insert" % self.name,
[self.resource_handle, keys, values]):
keys = ops.convert_to_tensor(keys, self._key_dtype, name="keys")
values = ops.convert_to_tensor(values, self._value_dtype, name="values")
with ops.colocate_with(self.resource_handle):
# pylint: disable=protected-access
op = gen_lookup_ops.lookup_table_insert_v2(self.resource_handle, keys,
values)
return op
def export(self, name=None):
"""Returns tensors of all keys and values in the table.
Args:
name: A name for the operation (optional).
Returns:
A pair of tensors with the first tensor containing all keys and the
second tensors containing all values in the table.
"""
with ops.name_scope(name, "%s_lookup_table_export_values" % self.name,
[self.resource_handle]):
with ops.colocate_with(self.resource_handle):
exported_keys, exported_values = gen_lookup_ops.lookup_table_export_v2(
self.resource_handle, self._key_dtype, self._value_dtype)
return exported_keys, exported_values
def _gather_saveables_for_checkpoint(self):
"""For object-based checkpointing."""
return {
"table":
functools.partial(
MutableHashTable._Saveable, table=self, name=self._name)
}
class _Saveable(BaseSaverBuilder.SaveableObject):
"""SaveableObject implementation for MutableHashTable."""
def __init__(self, table, name):
tensors = table.export()
specs = [
BaseSaverBuilder.SaveSpec(tensors[0], "", name + "-keys"),
BaseSaverBuilder.SaveSpec(tensors[1], "", name + "-values")
]
# pylint: disable=protected-access
super(MutableHashTable._Saveable, self).__init__(table, specs, name)
def restore(self, restored_tensors, restored_shapes, name=None):
del restored_shapes # unused
# pylint: disable=protected-access
with ops.name_scope(name, "%s_table_restore" % self.name):
with ops.colocate_with(self.op.resource_handle):
return gen_lookup_ops.lookup_table_import_v2(self.op.resource_handle,
restored_tensors[0],
restored_tensors[1])
@tf_export("lookup.experimental.DenseHashTable")
class DenseHashTable(LookupInterface):
"""A generic mutable hash table implementation using tensors as backing store.
Data can be inserted by calling the insert method and removed by calling the
remove method. It does not support initialization via the init method.
It uses "open addressing" with quadratic reprobing to resolve collisions.
Compared to `MutableHashTable` the insert, remove and lookup operations in a
`DenseHashTable` are typically faster, but memory usage can be higher.
However, `DenseHashTable` does not require additional memory for
temporary tensors created during checkpointing and restore operations.
Example usage:
```python
table = tf.lookup.DenseHashTable(key_dtype=tf.int64,
value_dtype=tf.int64,
default_value=-1,
empty_key=0,
deleted_key=-1)
sess.run(table.insert(keys, values))
out = table.lookup(query_keys)
print(out.eval())
```
"""
# TODO(andreasst): consider extracting common code with MutableHashTable into
# a common superclass.
def __init__(self,
key_dtype,
value_dtype,
default_value,
empty_key,
deleted_key,
initial_num_buckets=None,
name="MutableDenseHashTable",
checkpoint=True):
"""Creates an empty `DenseHashTable` object.
Creates a table, the type of its keys and values are specified by key_dtype
and value_dtype, respectively.
Args:
key_dtype: the type of the key tensors.
value_dtype: the type of the value tensors.
default_value: The value to use if a key is missing in the table.
empty_key: the key to use to represent empty buckets internally. Must not
be used in insert, remove or lookup operations.
deleted_key: the key to use to represent deleted buckets internally. Must
not be used in insert, remove or lookup operations and be different from
the empty_key.
initial_num_buckets: the initial number of buckets.
name: A name for the operation (optional).
checkpoint: if True, the contents of the table are saved to and restored
from checkpoints. If `shared_name` is empty for a checkpointed table, it
is shared using the table node name.
Returns:
A `DenseHashTable` object.
Raises:
ValueError: If checkpoint is True and no name was specified.
"""
self._default_value = ops.convert_to_tensor(
default_value, dtype=value_dtype, name="default_value")
self._key_dtype = key_dtype
self._value_dtype = value_dtype
self._initial_num_buckets = initial_num_buckets
self._value_shape = self._default_value.get_shape()
self._checkpoint = checkpoint
self._name = name
self._empty_key = ops.convert_to_tensor(
empty_key, dtype=key_dtype, name="empty_key")
self._deleted_key = ops.convert_to_tensor(
deleted_key, dtype=key_dtype, name="deleted_key")
self._shared_name = None
if context.executing_eagerly():
# TODO(allenl): This will leak memory due to kernel caching by the
# shared_name attribute value (but is better than the alternative of
# sharing everything by default when executing eagerly; hopefully creating
# tables in a loop is uncommon).
# TODO(rohanj): Use context.shared_name() instead.
self._shared_name = "table_%d" % (ops.uid(),)
super(DenseHashTable, self).__init__(key_dtype, value_dtype)
self._resource_handle = self._create_resource()
if checkpoint:
saveable = DenseHashTable._Saveable(self, name)
if not context.executing_eagerly():
ops.add_to_collection(ops.GraphKeys.SAVEABLE_OBJECTS, saveable)
def _create_resource(self):
# The table must be shared if checkpointing is requested for multi-worker
# training to work correctly. Use the node name if no shared_name has been
# explicitly specified.
use_node_name_sharing = self._checkpoint and self._shared_name is None
table_ref = gen_lookup_ops.mutable_dense_hash_table_v2(
empty_key=self._empty_key,
deleted_key=self._deleted_key,
shared_name=self._shared_name,
use_node_name_sharing=use_node_name_sharing,
value_dtype=self._value_dtype,
value_shape=self._value_shape,
initial_num_buckets=self._initial_num_buckets,
name=self._name)
if context.executing_eagerly():
self._table_name = None
else:
self._table_name = table_ref.op.name.split("/")[-1]
return table_ref
@property
def name(self):
return self._table_name
def size(self, name=None):
"""Compute the number of elements in this table.
Args:
name: A name for the operation (optional).
Returns:
A scalar tensor containing the number of elements in this table.
"""
with ops.name_scope(name, "%s_Size" % self.name, [self.resource_handle]):
with ops.colocate_with(self.resource_handle):
return gen_lookup_ops.lookup_table_size_v2(self.resource_handle)
def lookup(self, keys, name=None):
"""Looks up `keys` in a table, outputs the corresponding values.
The `default_value` is used for keys not present in the table.
Args:
keys: Keys to look up. Can be a tensor of any shape. Must match the
table's key_dtype.
name: A name for the operation (optional).
Returns:
A tensor containing the values in the same shape as `keys` using the
table's value type.
Raises:
TypeError: when `keys` do not match the table data types.
"""
with ops.name_scope(name, "%s_lookup_table_find" % self.name,
[self.resource_handle, keys]):
keys = ops.convert_to_tensor(keys, dtype=self._key_dtype, name="keys")
with ops.colocate_with(self.resource_handle):
values = gen_lookup_ops.lookup_table_find_v2(self.resource_handle, keys,
self._default_value)
return values
def insert_or_assign(self, keys, values, name=None):
"""Associates `keys` with `values`.
Args:
keys: Keys to insert. Can be a tensor of any shape. Must match the table's
key type.
values: Values to be associated with keys. Must be a tensor of the same
shape as `keys` and match the table's value type.
name: A name for the operation (optional).
Returns:
The created Operation.
Raises:
TypeError: when `keys` or `values` doesn't match the table data
types.
"""
with ops.name_scope(name, "%s_lookup_table_insert" % self.name,
[self.resource_handle, keys, values]):
keys = ops.convert_to_tensor(keys, dtype=self._key_dtype, name="keys")
values = ops.convert_to_tensor(
values, dtype=self._value_dtype, name="values")
with ops.colocate_with(self.resource_handle):
op = gen_lookup_ops.lookup_table_insert_v2(self.resource_handle, keys,
values)
return op
def insert(self, keys, values, name=None):
"""Associates `keys` with `values`.
Args:
keys: Keys to insert. Can be a tensor of any shape. Must match the table's
key type.
values: Values to be associated with keys. Must be a tensor of the same
shape as `keys` and match the table's value type.
name: A name for the operation (optional).
Returns:
The created Operation.
Raises:
TypeError: when `keys` or `values` doesn't match the table data
types.
"""
return self.insert_or_assign(keys, values, name)
def erase(self, keys, name=None):
"""Removes `keys` and its associated values from the table.
If a key is not present in the table, it is silently ignored.
Args:
keys: Keys to remove. Can be a tensor of any shape. Must match the table's
key type.
name: A name for the operation (optional).
Returns:
The created Operation.
Raises:
TypeError: when `keys` do not match the table data types.
"""
if keys.dtype != self._key_dtype:
raise TypeError("Signature mismatch. Keys must be dtype %s, got %s." %
(self._key_dtype, keys.dtype))
with ops.name_scope(name, "%s_lookup_table_remove" % self.name,
(self.resource_handle, keys, self._default_value)):
# pylint: disable=protected-access
op = gen_lookup_ops.lookup_table_remove_v2(self.resource_handle, keys)
return op
def remove(self, keys, name=None):
"""Removes `keys` and its associated values from the table.
If a key is not present in the table, it is silently ignored.
Args:
keys: Keys to remove. Can be a tensor of any shape. Must match the table's
key type.
name: A name for the operation (optional).
Returns:
The created Operation.
Raises:
TypeError: when `keys` do not match the table data types.
"""
return self.erase(keys, name)
def export(self, name=None):
"""Returns tensors of all keys and values in the table.
Args:
name: A name for the operation (optional).
Returns:
A pair of tensors with the first tensor containing all keys and the
second tensors containing all values in the table.
"""
with ops.name_scope(name, "%s_lookup_table_export_values" % self.name,
[self.resource_handle]):
with ops.colocate_with(self.resource_handle):
exported_keys, exported_values = gen_lookup_ops.lookup_table_export_v2(
self.resource_handle, self._key_dtype, self._value_dtype)
return exported_keys, exported_values
def _gather_saveables_for_checkpoint(self):
"""For object-based checkpointing."""
return {
"table":
functools.partial(
DenseHashTable._Saveable, table=self, name=self._name)
}
class _Saveable(BaseSaverBuilder.SaveableObject):
"""SaveableObject implementation for DenseHashTable."""
def __init__(self, table, name):
tensors = table.export()
specs = [
BaseSaverBuilder.SaveSpec(tensors[0], "", name + "-keys"),
BaseSaverBuilder.SaveSpec(tensors[1], "", name + "-values")
]
# pylint: disable=protected-access
super(DenseHashTable._Saveable, self).__init__(table, specs, name)
def restore(self, restored_tensors, restored_shapes, name=None):
del restored_shapes # unused
# pylint: disable=protected-access
with ops.name_scope(name, "%s_table_restore" % self.name):
with ops.colocate_with(self.op.resource_handle):
return gen_lookup_ops.lookup_table_import_v2(self.op.resource_handle,
restored_tensors[0],
restored_tensors[1])
ops.NotDifferentiable("LookupTableFind")
ops.NotDifferentiable("LookupTableFindV2")
ops.NotDifferentiable("LookupTableInsert")
ops.NotDifferentiable("LookupTableInsertV2")
ops.NotDifferentiable("LookupTableSize")
ops.NotDifferentiable("LookupTableSizeV2")
ops.NotDifferentiable("HashTable")
ops.NotDifferentiable("HashTableV2")
ops.NotDifferentiable("InitializeTable")
ops.NotDifferentiable("InitializeTableV2")
ops.NotDifferentiable("InitializeTableFromTextFile")
ops.NotDifferentiable("InitializeTableFromTextFileV2")
ops.NotDifferentiable("MutableDenseHashTable")
ops.NotDifferentiable("MutableDenseHashTableV2")
ops.NotDifferentiable("MutableHashTable")
ops.NotDifferentiable("MutableHashTableV2")
ops.NotDifferentiable("MutableHashTableOfTensors")
ops.NotDifferentiable("MutableHashTableOfTensorsV2")
|
tensorflow-master
|
tensorflow/python/ops/lookup_ops.py
|
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Benchmark for accumulate_n() in math_ops."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import random
import time
from six.moves import xrange # pylint: disable=redefined-builtin
from tensorflow.python.client import session
from tensorflow.python.framework import ops
from tensorflow.python.framework import tensor_shape
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import data_flow_ops
from tensorflow.python.ops import gen_control_flow_ops
from tensorflow.python.ops import gen_state_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import random_ops
from tensorflow.python.ops import state_ops
from tensorflow.python.platform import test
class AccumulateNBenchmark(test.Benchmark):
def _AccumulateNTemplate(self, inputs, init, shape, validate_shape):
var = gen_state_ops.temporary_variable(
shape=shape, dtype=inputs[0].dtype.base_dtype)
ref = state_ops.assign(var, init, validate_shape=validate_shape)
update_ops = [
state_ops.assign_add(
ref, tensor, use_locking=True).op for tensor in inputs
]
with ops.control_dependencies(update_ops):
return gen_state_ops.destroy_temporary_variable(ref, var_name=var.op.name)
def _AccumulateNInitializedWithFirst(self, inputs):
return self._AccumulateNTemplate(
inputs,
init=array_ops.zeros_like(inputs[0]),
shape=inputs[0].get_shape(),
validate_shape=True)
def _AccumulateNInitializedWithMerge(self, inputs):
return self._AccumulateNTemplate(
inputs,
init=array_ops.zeros_like(gen_control_flow_ops.merge(inputs)[0]),
shape=tensor_shape.vector(0),
validate_shape=False)
def _AccumulateNInitializedWithShape(self, inputs):
return self._AccumulateNTemplate(
inputs,
init=array_ops.zeros(
shape=inputs[0].get_shape(), dtype=inputs[0].dtype.base_dtype),
shape=inputs[0].get_shape(),
validate_shape=True)
def _GenerateUnorderedInputs(self, size, n):
inputs = [random_ops.random_uniform(shape=[size]) for _ in xrange(n)]
random.shuffle(inputs)
return inputs
def _GenerateReplicatedInputs(self, size, n):
return n * self._GenerateUnorderedInputs(size, 1)
def _GenerateOrderedInputs(self, size, n):
inputs = self._GenerateUnorderedInputs(size, 1)
queue = data_flow_ops.FIFOQueue(
capacity=1, dtypes=[inputs[0].dtype], shapes=[inputs[0].get_shape()])
for _ in xrange(n - 1):
op = queue.enqueue(inputs[-1])
with ops.control_dependencies([op]):
inputs.append(math_ops.tanh(1.0 + queue.dequeue()))
return inputs
def _GenerateReversedInputs(self, size, n):
inputs = self._GenerateOrderedInputs(size, n)
inputs.reverse()
return inputs
def _SetupAndRunBenchmark(self, graph, inputs, repeats, format_args):
with graph.as_default():
add_n = math_ops.add_n(inputs)
acc_n_first = self._AccumulateNInitializedWithFirst(inputs)
acc_n_merge = self._AccumulateNInitializedWithMerge(inputs)
acc_n_shape = self._AccumulateNInitializedWithShape(inputs)
test_ops = (("AddN", add_n.op),
("AccNFirst", acc_n_first.op),
("AccNMerge", acc_n_merge.op),
("AccNShape", acc_n_shape.op))
with session.Session(graph=graph):
for tag, op in test_ops:
for _ in xrange(100):
op.run() # Run for warm up.
start = time.time()
for _ in xrange(repeats):
op.run()
duration = time.time() - start
args = format_args + (tag, duration)
print(self._template.format(*args))
def _RunBenchmark(self, tag, input_fn, sizes, ninputs, repeats):
for size in sizes:
for ninput in ninputs:
graph = ops.Graph()
with graph.as_default():
inputs = input_fn(size, ninput)
format_args = (tag, size, ninput, repeats)
self._SetupAndRunBenchmark(graph, inputs, repeats, format_args)
def benchmarkAccumulateN(self):
self._template = "{:<15}" * 6
args = {
"sizes": (128, 128**2),
"ninputs": (1, 10, 100, 300),
"repeats": 100
}
benchmarks = (("Replicated", self._GenerateReplicatedInputs),
("Unordered", self._GenerateUnorderedInputs),
("Ordered", self._GenerateOrderedInputs),
("Reversed", self._GenerateReversedInputs))
print(self._template.format("", "Size", "#Inputs", "#Repeat", "Method",
"Duration"))
print("-" * 90)
for benchmark in benchmarks:
self._RunBenchmark(*benchmark, **args)
if __name__ == "__main__":
test.main()
|
tensorflow-master
|
tensorflow/python/ops/accumulate_n_benchmark.py
|
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for fused_batch_norm related functionality in tensorflow.ops.nn."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.python.compat import compat
from tensorflow.python.framework import constant_op
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import test_util
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gradient_checker
from tensorflow.python.ops import gradients_impl
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import nn_grad
from tensorflow.python.ops import nn_impl
from tensorflow.python.platform import test
class BatchNormalizationTest(test.TestCase):
def _batch_norm(self, x, mean, var, offset, scale, epsilon):
# We compute the batch norm manually in this function because
# nn_impl.batch_normalization does not support float16 yet.
# TODO(reedwm): Add float16 support to nn_impl.batch_normalization.
inv = math_ops.rsqrt(var + epsilon) * scale
y = math_ops.cast(x, scale.dtype) * inv + (offset - mean * inv)
return math_ops.cast(y, x.dtype)
def _inference_ref(self, x, scale, offset, mean, var, epsilon, data_format):
if data_format not in ['NHWC', 'NCHW']:
raise ValueError('data_format must be NCHW or NHWC, '
'got %s.' % data_format)
if data_format == 'NCHW':
x = array_ops.transpose(x, [0, 2, 3, 1])
y = self._batch_norm(x, mean, var, offset, scale, epsilon)
if data_format == 'NCHW':
y = array_ops.transpose(y, [0, 3, 1, 2])
return self.evaluate(y)
def _test_inference(self,
x_shape,
x_dtype,
scale_shape,
scale_dtype,
use_gpu=True,
data_format='NHWC'):
np.random.seed(1)
x_val = np.random.random_sample(x_shape).astype(x_dtype)
scale_val = np.random.random_sample(scale_shape).astype(scale_dtype)
offset_val = np.random.random_sample(scale_shape).astype(scale_dtype)
mean_val = np.random.random_sample(scale_shape).astype(scale_dtype)
var_val = np.random.random_sample(scale_shape).astype(scale_dtype)
with self.cached_session(use_gpu=use_gpu) as sess:
x = constant_op.constant(x_val, name='x')
scale = constant_op.constant(scale_val, name='scale')
offset = constant_op.constant(offset_val, name='offset')
mean = constant_op.constant(mean_val, name='mean')
var = constant_op.constant(var_val, name='variance')
epsilon = 0.001
y, _, _ = nn_impl.fused_batch_norm(
x,
scale,
offset,
mean=mean,
variance=var,
epsilon=epsilon,
data_format=data_format,
is_training=False)
y_val = self.evaluate(y)
y_ref = self._inference_ref(x, scale, offset, mean, var, epsilon,
data_format)
# An atol value of 1e-3 is too small for float16's, because some adjacent
# float16 values that y_val can take are greater than 1e-3 apart, e.g.
# 2.16602 and 2.16797.
atol = 2e-3 if x_dtype == np.float16 else 1e-3
self.assertAllClose(y_ref, y_val, atol=atol)
def _training_ref(self, x, scale, offset, epsilon, data_format):
if data_format not in ['NHWC', 'NCHW']:
raise ValueError('data_format must be NCHW or NHWC, '
'got %s.' % data_format)
if data_format == 'NCHW':
x = array_ops.transpose(x, [0, 2, 3, 1])
mean, var = nn_impl.moments(
math_ops.cast(x, scale.dtype), [0, 1, 2], keep_dims=False)
y = self._batch_norm(x, mean, var, offset, scale, epsilon)
if data_format == 'NCHW':
y = array_ops.transpose(y, [0, 3, 1, 2])
return self.evaluate(y), self.evaluate(mean), self.evaluate(var)
def _test_training(self,
x_shape,
x_dtype,
scale_shape,
scale_dtype,
use_gpu=True,
data_format='NHWC'):
np.random.seed(1)
x_val = np.random.random_sample(x_shape).astype(x_dtype)
scale_val = np.random.random_sample(scale_shape).astype(scale_dtype)
offset_val = np.random.random_sample(scale_shape).astype(scale_dtype)
with self.cached_session(use_gpu=use_gpu) as sess:
x = constant_op.constant(x_val, name='x')
scale = constant_op.constant(scale_val, name='scale')
offset = constant_op.constant(offset_val, name='offset')
epsilon = 0.001
y, mean, var = nn_impl.fused_batch_norm(
x,
scale,
offset,
epsilon=epsilon,
data_format=data_format,
is_training=True)
y_val, mean_val, var_val = self.evaluate([y, mean, var])
y_ref, mean_ref, var_ref = self._training_ref(x, scale, offset, epsilon,
data_format)
y_atol = 2e-3 if x_dtype == np.float16 else 1e-3
self.assertAllClose(y_ref, y_val, atol=y_atol)
self.assertAllClose(mean_ref, mean_val, atol=1e-3)
# This is for Bessel's correction. tf.nn.moments uses n, instead of n-1, as
# the denominator in the formula to calculate variance, while
# tf.compat.v1.nn.fused_batch_norm has Bessel's correction built in.
sample_size = x_val.size / scale_val.size
var_ref = var_ref * sample_size / (max(sample_size - 1.0, 1.0))
self.assertAllClose(var_ref, var_val, atol=1e-3)
def _compute_gradient_error_float16(self, x, x32, x_shape, y, y32, y_shape):
"""Computes the gradient error for float16 inputs and/or outputs.
This returns the same value as gradient_checker.compute_gradient_error. The
difference is that gradient_checker.compute_gradient_error does not
numerically compute the gradients in a numerically stable way for float16
tensors. To fix this, this function requires float32 versions of x and y to
numerically compute the gradients, to compare with the float16 symbolically
computed gradients.
Args:
x: The input tensor.
x32: A float32 version of x.
x_shape: The shape of x.
y: The output tensor.
y32: A float32 version of y. Must be calculated based on x32, not x.
y_shape: The shape of y.
Returns:
The maximum error in between the two Jacobians, as in
gradient_checker.compute_gradient_error.
"""
x_init_val = np.random.random_sample(x_shape).astype(np.float16)
x32_init_val = x_init_val.astype(np.float32)
# TODO(reedwm): Do not perform the unnecessary computations in
# compute_gradient, since they double the computation time of this function.
theoretical_grad, _ = gradient_checker.compute_gradient(
x, x_shape, y, y_shape, delta=1e-3, x_init_value=x_init_val)
_, numerical_grad = gradient_checker.compute_gradient(
x32, x_shape, y32, y_shape, delta=1e-3, x_init_value=x32_init_val)
# If grad is empty, no error.
if theoretical_grad.size == 0 and numerical_grad.size == 0:
return 0
return np.fabs(theoretical_grad - numerical_grad).max()
def _test_gradient(self,
x_shape,
x_dtype,
scale_shape,
scale_dtype,
use_gpu=True,
data_format='NHWC',
is_training=True):
np.random.seed(1)
x_val = np.random.random_sample(x_shape).astype(x_dtype)
scale_val = np.random.random_sample(scale_shape).astype(scale_dtype)
offset_val = np.random.random_sample(scale_shape).astype(scale_dtype)
with self.cached_session(use_gpu=use_gpu):
x = constant_op.constant(x_val, name='x')
scale = constant_op.constant(scale_val, name='scale')
offset = constant_op.constant(offset_val, name='offset')
if is_training:
pop_mean = None
pop_var = None
else:
pop_mean = np.random.random_sample(scale_shape).astype(scale_dtype)
pop_var = np.random.random_sample(scale_shape).astype(scale_dtype)
y, _, _ = nn_impl.fused_batch_norm(
x,
scale,
offset,
mean=pop_mean,
variance=pop_var,
data_format=data_format,
is_training=is_training)
if x_dtype != np.float16:
err_x = gradient_checker.compute_gradient_error(x, x_shape, y, x_shape)
err_scale = gradient_checker.compute_gradient_error(
scale, scale_shape, y, x_shape)
err_offset = gradient_checker.compute_gradient_error(
offset, scale_shape, y, x_shape)
else:
x32 = constant_op.constant(x_val, name='x32', dtype=dtypes.float32)
y32, _, _ = nn_impl.fused_batch_norm(
x32,
scale,
offset,
mean=pop_mean,
variance=pop_var,
data_format=data_format,
is_training=is_training)
err_x = self._compute_gradient_error_float16(x, x32, x_shape, y, y32,
x_shape)
err_scale = self._compute_gradient_error_float16(
scale, scale, scale_shape, y, y32, x_shape)
err_offset = self._compute_gradient_error_float16(
offset, offset, scale_shape, y, y32, x_shape)
x_err_tolerance = 2e-3 if x_dtype == np.float16 else 1e-3
scale_err_tolerance = 1e-3
self.assertLess(err_x, x_err_tolerance)
self.assertLess(err_scale, scale_err_tolerance)
self.assertLess(err_offset, scale_err_tolerance)
def _test_grad_grad(self,
x_shape,
x_dtype,
scale_shape,
scale_dtype,
use_gpu=True,
data_format='NHWC',
is_training=True,
err_tolerance=1e-3):
np.random.seed(1)
x_val = np.random.random_sample(x_shape).astype(x_dtype)
grad_y_val = np.random.random_sample(x_shape).astype(x_dtype)
scale_val = np.random.random_sample(scale_shape).astype(scale_dtype)
offset_val = np.random.random_sample(scale_shape).astype(scale_dtype)
with self.cached_session(use_gpu=use_gpu) as sess:
x = constant_op.constant(x_val, name='x')
grad_y = constant_op.constant(grad_y_val, name='grad_y')
scale = constant_op.constant(scale_val, name='scale')
offset = constant_op.constant(offset_val, name='offset')
if is_training:
pop_mean = None
pop_var = None
else:
pop_mean = np.random.random_sample(scale_shape).astype(scale_dtype)
pop_var = np.random.random_sample(scale_shape).astype(scale_dtype)
y, _, _ = nn_impl.fused_batch_norm(
x,
scale,
offset,
mean=pop_mean,
variance=pop_var,
data_format=data_format,
is_training=is_training)
grad_x, grad_scale, grad_offset = gradients_impl.gradients(
y, [x, scale, offset], grad_y)
if is_training:
epsilon = y.op.get_attr('epsilon')
data_format = y.op.get_attr('data_format')
grad_vals = self.evaluate([grad_x, grad_scale, grad_offset])
grad_internal = nn_grad._BatchNormGrad(grad_y, x, scale, pop_mean,
pop_var, epsilon, data_format)
grad_internal_vals = self.evaluate(list(grad_internal))
for grad_val, grad_internal_val in zip(grad_vals, grad_internal_vals):
self.assertAllClose(grad_val, grad_internal_val, atol=err_tolerance)
if x_dtype != np.float16:
err_grad_grad_y_1 = gradient_checker.compute_gradient_error(
grad_y, x_shape, grad_x, x_shape)
err_grad_grad_y_2 = gradient_checker.compute_gradient_error(
grad_y, x_shape, grad_scale, scale_shape)
err_grad_grad_y_3 = gradient_checker.compute_gradient_error(
grad_y, x_shape, grad_offset, scale_shape)
# In freeze mode, grad_x is not a function of x.
if is_training:
err_grad_x_1 = gradient_checker.compute_gradient_error(
x, x_shape, grad_x, x_shape)
err_grad_x_2 = gradient_checker.compute_gradient_error(
x, x_shape, grad_scale, scale_shape)
err_grad_scale = gradient_checker.compute_gradient_error(
scale, scale_shape, grad_x, x_shape)
else:
x32 = constant_op.constant(x_val, dtype=dtypes.float32, name='x32')
grad_y32 = constant_op.constant(
grad_y_val, dtype=dtypes.float32, name='grad_y32')
y32, _, _ = nn_impl.fused_batch_norm(
x32,
scale,
offset,
mean=pop_mean,
variance=pop_var,
data_format=data_format,
is_training=is_training)
grad_x32, grad_scale32, grad_offset32 = gradients_impl.gradients(
y32, [x32, scale, offset], grad_y32)
err_grad_grad_y_1 = self._compute_gradient_error_float16(
grad_y, grad_y32, x_shape, grad_x, grad_x32, x_shape)
err_grad_grad_y_2 = self._compute_gradient_error_float16(
grad_y, grad_y32, x_shape, grad_scale, grad_scale32, scale_shape)
err_grad_grad_y_3 = self._compute_gradient_error_float16(
grad_y, grad_y32, x_shape, grad_offset, grad_offset32, scale_shape)
# In freeze mode, grad_x is not a function of x.
if is_training:
err_grad_x_1 = self._compute_gradient_error_float16(
x, x32, x_shape, grad_x, grad_x32, x_shape)
err_grad_x_2 = self._compute_gradient_error_float16(
x, x32, x_shape, grad_scale, grad_scale32, scale_shape)
err_grad_scale = self._compute_gradient_error_float16(
scale, scale, scale_shape, grad_x, grad_x32, x_shape)
self.assertLess(err_grad_grad_y_1, err_tolerance)
self.assertLess(err_grad_grad_y_2, err_tolerance)
self.assertLess(err_grad_grad_y_3, err_tolerance)
if is_training:
self.assertLess(err_grad_x_1, err_tolerance)
self.assertLess(err_grad_x_2, err_tolerance)
self.assertLess(err_grad_scale, err_tolerance)
def testInferenceShape1(self):
x_shape = [1, 1, 6, 1]
for dtype in [np.float16, np.float32]:
if test.is_gpu_available(cuda_only=True):
self._test_inference(
x_shape, dtype, [1], np.float32, use_gpu=True, data_format='NHWC')
self._test_inference(
x_shape, dtype, [1], np.float32, use_gpu=True, data_format='NCHW')
self._test_inference(
x_shape, dtype, [1], np.float32, use_gpu=False, data_format='NHWC')
def testInferenceShape2(self):
x_shape = [1, 1, 6, 2]
if test.is_gpu_available(cuda_only=True):
for dtype in [np.float16, np.float32]:
self._test_inference(
x_shape, dtype, [2], np.float32, use_gpu=True, data_format='NHWC')
self._test_inference(
x_shape, dtype, [2], np.float32, use_gpu=False, data_format='NHWC')
def testInferenceShape3(self):
x_shape = [1, 2, 1, 6]
if test.is_gpu_available(cuda_only=True):
for dtype in [np.float16, np.float32]:
self._test_inference(
x_shape, dtype, [2], np.float32, use_gpu=True, data_format='NCHW')
def testInferenceShape4(self):
x_shape = [27, 131, 127, 6]
for dtype in [np.float16, np.float32]:
if test.is_gpu_available(cuda_only=True):
self._test_inference(
x_shape, dtype, [131], np.float32, use_gpu=True, data_format='NCHW')
self._test_inference(
x_shape, dtype, [6], np.float32, use_gpu=True, data_format='NHWC')
self._test_inference(
x_shape, dtype, [6], np.float32, use_gpu=False, data_format='NHWC')
def testInferenceShape5(self):
with compat.forward_compatibility_horizon(2019, 6, 7):
x_shape = [0, 131, 127, 6]
for dtype in [np.float16, np.float32]:
if test.is_gpu_available(cuda_only=True):
self._test_inference(
x_shape,
dtype, [131],
np.float32,
use_gpu=True,
data_format='NCHW')
self._test_inference(
x_shape, dtype, [6], np.float32, use_gpu=True, data_format='NHWC')
self._test_inference(
x_shape, dtype, [6], np.float32, use_gpu=False, data_format='NHWC')
def testTrainingShape1(self):
x_shape = [1, 1, 6, 1]
for dtype in [np.float16, np.float32]:
if test.is_gpu_available(cuda_only=True):
self._test_training(
x_shape, dtype, [1], np.float32, use_gpu=True, data_format='NHWC')
self._test_training(
x_shape, dtype, [1], np.float32, use_gpu=True, data_format='NCHW')
self._test_training(
x_shape, dtype, [1], np.float32, use_gpu=False, data_format='NHWC')
def testTrainingShape2(self):
x_shape = [1, 1, 6, 2]
for dtype in [np.float16, np.float32]:
if test.is_gpu_available(cuda_only=True):
self._test_training(
x_shape, dtype, [2], np.float32, use_gpu=True, data_format='NHWC')
self._test_training(
x_shape, dtype, [2], np.float32, use_gpu=False, data_format='NHWC')
def testTrainingShape3(self):
x_shape = [1, 2, 1, 6]
if test.is_gpu_available(cuda_only=True):
for dtype in [np.float16, np.float32]:
self._test_training(
x_shape, dtype, [2], np.float32, use_gpu=True, data_format='NCHW')
def testTrainingShape4(self):
x_shape = [27, 131, 127, 6]
for dtype in [np.float16, np.float32]:
if test.is_gpu_available(cuda_only=True):
self._test_training(
x_shape, dtype, [131], np.float32, use_gpu=True, data_format='NCHW')
self._test_training(
x_shape, dtype, [6], np.float32, use_gpu=True, data_format='NHWC')
self._test_training(
x_shape, dtype, [6], np.float32, use_gpu=False, data_format='NHWC')
def testTrainingShape5(self):
with compat.forward_compatibility_horizon(2019, 6, 7):
x_shape = [0, 131, 127, 6]
for dtype in [np.float16, np.float32]:
if test.is_gpu_available(cuda_only=True):
self._test_training(
x_shape,
dtype, [131],
np.float32,
use_gpu=True,
data_format='NCHW')
self._test_training(
x_shape, dtype, [6], np.float32, use_gpu=True, data_format='NHWC')
self._test_training(
x_shape, dtype, [6], np.float32, use_gpu=False, data_format='NHWC')
@test_util.run_deprecated_v1
def testBatchNormGradShape1(self):
for is_training in [True, False]:
x_shape = [1, 1, 6, 1]
for dtype in [np.float16, np.float32]:
if test.is_gpu_available(cuda_only=True):
self._test_gradient(
x_shape,
dtype, [1],
np.float32,
use_gpu=True,
data_format='NHWC',
is_training=is_training)
self._test_gradient(
x_shape,
dtype, [1],
np.float32,
use_gpu=True,
data_format='NCHW',
is_training=is_training)
self._test_gradient(
x_shape,
dtype, [1],
np.float32,
use_gpu=False,
data_format='NHWC',
is_training=is_training)
@test_util.run_deprecated_v1
def testBatchNormGradShape2(self):
for is_training in [True, False]:
x_shape = [1, 1, 6, 2]
for dtype in [np.float16, np.float32]:
if test.is_gpu_available(cuda_only=True):
self._test_gradient(
x_shape,
dtype, [2],
np.float32,
use_gpu=True,
data_format='NHWC',
is_training=is_training)
self._test_gradient(
x_shape,
dtype, [2],
np.float32,
use_gpu=False,
data_format='NHWC',
is_training=is_training)
@test_util.run_deprecated_v1
def testBatchNormGradShape3(self):
for is_training in [True, False]:
x_shape = [1, 2, 1, 6]
if test.is_gpu_available(cuda_only=True):
for dtype in [np.float16, np.float32]:
self._test_gradient(
x_shape,
dtype, [2],
np.float32,
use_gpu=True,
data_format='NCHW',
is_training=is_training)
@test_util.run_deprecated_v1
def testBatchNormGradShape4(self):
for is_training in [True, False]:
x_shape = [5, 7, 11, 4]
for dtype in [np.float16, np.float32]:
if test.is_gpu_available(cuda_only=True):
self._test_gradient(
x_shape,
dtype, [7],
np.float32,
use_gpu=True,
data_format='NCHW',
is_training=is_training)
self._test_gradient(
x_shape,
dtype, [4],
np.float32,
use_gpu=True,
data_format='NHWC',
is_training=is_training)
self._test_gradient(
x_shape,
dtype, [4],
np.float32,
use_gpu=False,
data_format='NHWC',
is_training=is_training)
@test_util.run_deprecated_v1
@test_util.disable_xla('This test never passed for XLA')
def testBatchNormGradShape5(self):
with compat.forward_compatibility_horizon(2019, 6, 7):
for is_training in [True, False]:
x_shape = [0, 7, 11, 4]
for dtype in [np.float16, np.float32]:
if test.is_gpu_available(cuda_only=True):
self._test_gradient(
x_shape,
dtype, [7],
np.float32,
use_gpu=True,
data_format='NCHW',
is_training=is_training)
self._test_gradient(
x_shape,
dtype, [4],
np.float32,
use_gpu=True,
data_format='NHWC',
is_training=is_training)
self._test_gradient(
x_shape,
dtype, [4],
np.float32,
use_gpu=False,
data_format='NHWC',
is_training=is_training)
def _testBatchNormGradGrad(self, config):
shape = config['shape']
err_tolerance = config['err_tolerance']
dtype = config['dtype']
for is_training in [True, False]:
if test.is_gpu_available(cuda_only=True):
self._test_grad_grad(
shape,
dtype, [shape[3]],
np.float32,
use_gpu=True,
data_format='NHWC',
is_training=is_training,
err_tolerance=err_tolerance)
self._test_grad_grad(
shape,
dtype, [shape[1]],
np.float32,
use_gpu=True,
data_format='NCHW',
is_training=is_training,
err_tolerance=err_tolerance)
self._test_grad_grad(
shape,
dtype, [shape[3]],
np.float32,
use_gpu=False,
data_format='NHWC',
is_training=is_training,
err_tolerance=err_tolerance)
@test_util.run_deprecated_v1
def testBatchNormGradGradConfig1(self):
config = {
'shape': [2, 3, 4, 5],
'err_tolerance': 1e-2,
'dtype': np.float32,
}
self._testBatchNormGradGrad(config)
@test_util.run_deprecated_v1
def testBatchNormGradGradConfig2(self):
config = {
'shape': [2, 3, 2, 2],
'err_tolerance': 1e-3,
'dtype': np.float32,
}
self._testBatchNormGradGrad(config)
@test_util.run_deprecated_v1
def testBatchNormGradGradConfig3(self):
config = {
'shape': [2, 3, 4, 5],
'err_tolerance': 1e-2,
'dtype': np.float16,
}
self._testBatchNormGradGrad(config)
@test_util.run_deprecated_v1
def testBatchNormGradGradConfig4(self):
config = {
'shape': [2, 3, 2, 2],
'err_tolerance': 2e-3,
'dtype': np.float16,
}
self._testBatchNormGradGrad(config)
if __name__ == '__main__':
test.main()
|
tensorflow-master
|
tensorflow/python/ops/nn_fused_batchnorm_test.py
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Gradients for operators defined in sparse_ops.py."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.framework import sparse_tensor
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import gen_sparse_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import sparse_ops
# TODO(b/31222613): This op may be differentiable, and there may be
# latent bugs here.
ops.NotDifferentiable("SparseAddGrad")
ops.NotDifferentiable("SparseConcat")
@ops.RegisterGradient("SparseReorder")
def _SparseReorderGrad(op, unused_output_indices_grad, output_values_grad):
"""Gradients for the SparseReorder op.
Args:
op: the SparseReorder op
unused_output_indices_grad: the incoming gradients of the output indices
output_values_grad: the incoming gradients of the output values
Returns:
Gradient for each of the 3 input tensors:
(input_indices, input_values, input_shape)
The gradients for input_indices and input_shape is None.
"""
input_indices = op.inputs[0]
input_shape = op.inputs[2]
num_entries = array_ops.shape(input_indices)[0]
entry_indices = math_ops.range(num_entries)
sp_unordered = sparse_tensor.SparseTensor(
input_indices, entry_indices, input_shape)
sp_ordered = sparse_ops.sparse_reorder(sp_unordered)
inverted_permutation = array_ops.invert_permutation(sp_ordered.values)
return (None,
array_ops.gather(output_values_grad, inverted_permutation),
None)
@ops.RegisterGradient("SparseAdd")
def _SparseAddGrad(op, *grads):
"""The backward operator for the SparseAdd op.
The SparseAdd op calculates A + B, where A, B, and the sum are all represented
as `SparseTensor` objects. This op takes in the upstream gradient w.r.t.
non-empty values of the sum, and outputs the gradients w.r.t. the non-empty
values of A and B.
Args:
op: the SparseAdd op
*grads: the incoming gradients, one element per output of `op`
Returns:
Gradient for each of the 6 input tensors of SparseAdd:
(a_indices, a_values, a_shape, b_indices, b_values, b_shape, thresh)
The gradients for the indices, shapes, and the threshold are None.
"""
val_grad = grads[1]
a_indices = op.inputs[0]
b_indices = op.inputs[3]
sum_indices = op.outputs[0]
# NOTE: we do not need to take `thresh` into account, since it simply affects
# the non-zero elements of the sum, and we will peek into `sum_indices` in the
# gradient op.
a_val_grad, b_val_grad = gen_sparse_ops.sparse_add_grad(
val_grad, a_indices, b_indices, sum_indices)
a_val_grad.set_shape(op.inputs[1].get_shape())
b_val_grad.set_shape(op.inputs[4].get_shape())
# (a_indices, a_values, a_shape, b_indices, b_values, b_shape, thresh)
return (None, a_val_grad, None, None, b_val_grad, None, None)
@ops.RegisterGradient("SparseTensorDenseAdd")
def _SparseTensorDenseAddGrad(op, out_grad):
sp_indices = op.inputs[0]
# (sparse_indices, sparse_values, sparse_shape, dense)
return (None, array_ops.gather_nd(out_grad, sp_indices), None, out_grad)
@ops.RegisterGradient("SparseReduceSum")
def _SparseReduceSumGrad(op, out_grad):
"""Similar to gradient for the Sum Op (i.e. tf.reduce_sum())."""
sp_indices = op.inputs[0]
sp_shape = op.inputs[2]
output_shape_kept_dims = math_ops.reduced_shape(sp_shape, op.inputs[3])
out_grad_reshaped = array_ops.reshape(out_grad, output_shape_kept_dims)
scale = sp_shape // math_ops.cast(output_shape_kept_dims, dtypes.int64)
# (sparse_indices, sparse_values, sparse_shape, reduction_axes)
return (None, array_ops.gather_nd(out_grad_reshaped, sp_indices // scale),
None, None)
@ops.RegisterGradient("SparseSlice")
def _SparseSliceGrad(op, *grads):
"""The backward operator for the SparseSlice op.
This op takes in the upstream gradient w.r.t. non-empty values of
the sliced `SparseTensor`, and outputs the gradients w.r.t.
the non-empty values of input `SparseTensor`.
Args:
op: the SparseSlice op
*grads: the incoming gradients, one element per output of `op`
Returns:
Gradient for each of the 5 input tensors of SparseSlice:
(indices, values, shape, start, size)
The gradients for the indices, shape, start and the size are None.
"""
backprop_val_grad = grads[1]
input_indices = op.inputs[0]
input_start = op.inputs[3]
output_indices = op.outputs[0]
val_grad = gen_sparse_ops.sparse_slice_grad(
backprop_val_grad, input_indices, input_start, output_indices)
val_grad.set_shape(op.inputs[1].get_shape())
# (indices, values, shape, start, size)
return (None, val_grad, None, None, None)
@ops.RegisterGradient("SparseTensorDenseMatMul")
def _SparseTensorDenseMatMulGrad(op, grad):
"""Gradients for the dense tensor in the SparseTensorDenseMatMul op.
If either input is complex, no gradient is provided.
Args:
op: the SparseTensorDenseMatMul op
grad: the incoming gradient
Returns:
Gradient for each of the 4 input tensors:
(sparse_indices, sparse_values, sparse_shape, dense_tensor)
The gradients for indices and shape are None.
Raises:
TypeError: When the two operands don't have the same type.
"""
a_indices, a_values, a_shape = op.inputs[:3]
b = op.inputs[3]
adj_a = op.get_attr("adjoint_a")
adj_b = op.get_attr("adjoint_b")
a_type = a_values.dtype.base_dtype
b_type = b.dtype.base_dtype
if a_type != b_type:
raise TypeError("SparseTensorDenseMatMul op received operands with "
"different types: ", a_type, " and ", b_type)
if a_type in (ops.dtypes.complex64, ops.dtypes.complex128):
raise NotImplementedError("SparseTensorDenseMatMul op does not support "
"complex gradients.")
# gradient w.r.t. dense
b_grad = gen_sparse_ops.sparse_tensor_dense_mat_mul(
a_indices, a_values, a_shape, grad, adjoint_a=not adj_a)
if adj_b:
b_grad = array_ops.transpose(b_grad)
# gradient w.r.t. sparse values
rows = a_indices[:, 0]
cols = a_indices[:, 1]
# TODO(zongheng, ebrevdo): add conjugates in the right places when complex
# values are allowed.
# TODO(zongheng): these gather calls could potentially duplicate rows/cols in
# memory. If there is a need, we should look into implementing this more
# intelligently to avoid duplicating data.
parts_a = array_ops.gather(grad, rows if not adj_a else cols)
parts_b = array_ops.gather(b if not adj_b else array_ops.transpose(b),
cols if not adj_a else rows)
a_values_grad = math_ops.reduce_sum(parts_a * parts_b, axis=1)
# gradients w.r.t. (a_indices, a_values, a_shape, b)
return (None, a_values_grad, None, b_grad)
@ops.RegisterGradient("SparseDenseCwiseAdd")
def _SparseDenseCwiseAddGrad(unused_op, unused_grad):
raise NotImplementedError("Gradient for SparseDenseCwiseAdd is currently not"
" implemented yet.")
def _SparseDenseCwiseMulOrDivGrad(op, grad, is_mul):
"""Common code for SparseDenseCwise{Mul,Div} gradients."""
x_indices = op.inputs[0]
x_shape = op.inputs[2]
y = op.inputs[3]
y_shape = math_ops.cast(array_ops.shape(y), dtypes.int64)
num_added_dims = array_ops.expand_dims(
array_ops.size(x_shape) - array_ops.size(y_shape), 0)
augmented_y_shape = array_ops.concat(
[array_ops.ones(num_added_dims, ops.dtypes.int64), y_shape], 0)
scaling = x_shape // augmented_y_shape
scaled_indices = x_indices // scaling
scaled_indices = array_ops.slice(scaled_indices,
array_ops.concat([[0], num_added_dims], 0),
[-1, -1])
dense_vals = array_ops.gather_nd(y, scaled_indices)
if is_mul:
dx = grad * dense_vals
dy_val = grad * op.inputs[1]
else:
dx = grad / dense_vals
dy_val = grad * (-op.inputs[1] / math_ops.square(dense_vals))
# indices can repeat after scaling, so we can't use sparse_to_dense().
dy = sparse_ops.sparse_add(
array_ops.zeros_like(y),
sparse_tensor.SparseTensor(scaled_indices, dy_val, y_shape))
# (sp_indices, sp_vals, sp_shape, dense)
return (None, dx, None, dy)
@ops.RegisterGradient("SparseDenseCwiseMul")
def _SparseDenseCwiseMulGrad(op, grad):
"""Gradients for SparseDenseCwiseMul."""
return _SparseDenseCwiseMulOrDivGrad(op, grad, True)
@ops.RegisterGradient("SparseDenseCwiseDiv")
def _SparseDenseCwiseDivGrad(op, grad):
"""Gradients for SparseDenseCwiseDiv."""
return _SparseDenseCwiseMulOrDivGrad(op, grad, False)
@ops.RegisterGradient("SparseSoftmax")
def _SparseSoftmaxGrad(op, grad):
"""Gradients for SparseSoftmax.
The calculation is the same as SoftmaxGrad:
grad_x = grad_softmax * softmax - sum(grad_softmax * softmax) * softmax
where we now only operate on the non-zero values present in the SparseTensors.
Args:
op: the SparseSoftmax op.
grad: the upstream gradient w.r.t. the non-zero SparseSoftmax output values.
Returns:
Gradients w.r.t. the input (sp_indices, sp_values, sp_shape).
"""
indices, shape = op.inputs[0], op.inputs[2]
out_vals = op.outputs[0]
sp_output = sparse_tensor.SparseTensor(indices, out_vals, shape)
sp_grad = sparse_tensor.SparseTensor(indices, grad, shape)
sp_product = sparse_tensor.SparseTensor(
indices, sp_output.values * sp_grad.values, shape)
# [..., B, 1], dense.
sum_reduced = -sparse_ops.sparse_reduce_sum(sp_product, [-1], keepdims=True)
# sparse [..., B, C] + dense [..., B, 1] with broadcast; outputs sparse.
sp_sum = sparse_ops.sparse_dense_cwise_add(sp_grad, sum_reduced)
grad_x = sp_sum.values * sp_output.values
return [None, grad_x, None]
@ops.RegisterGradient("SparseSparseMaximum")
def _SparseSparseMaximumGrad(unused_op, unused_grad):
raise NotImplementedError("Gradient for SparseSparseMaximum is currently not"
" implemented yet.")
@ops.RegisterGradient("SparseSparseMinimum")
def _SparseSparseMinimumGrad(unused_op, unused_grad):
raise NotImplementedError("Gradient for SparseSparseMinimum is currently not"
" implemented yet.")
@ops.RegisterGradient("SparseFillEmptyRows")
def _SparseFillEmptyRowsGrad(op, unused_grad_output_indices, output_grad_values,
unused_grad_empty_row_indicator,
unused_grad_reverse_index_map):
"""Gradients for SparseFillEmptyRows."""
reverse_index_map = op.outputs[3]
d_values, d_default_value = gen_sparse_ops.sparse_fill_empty_rows_grad(
reverse_index_map=reverse_index_map, grad_values=output_grad_values)
# d_indices, d_values, d_dense_shape, d_default_value.
return [None, d_values, None, d_default_value]
@ops.RegisterGradient("SparseToDense")
def _SparseToDenseGrad(op, grad):
sparse_indices, output_shape, _, _ = op.inputs
sparse_values_grad = array_ops.gather_nd(grad, sparse_indices)
default_value_grad = math_ops.reduce_sum(grad) - math_ops.reduce_sum(
sparse_values_grad)
return [
array_ops.zeros_like(sparse_indices),
array_ops.zeros_like(output_shape), sparse_values_grad, default_value_grad
]
|
tensorflow-master
|
tensorflow/python/ops/sparse_grad.py
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.