File size: 3,565 Bytes
cd5fcb4 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 |
# Copyright (c) 2017-present, Facebook, Inc.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
#
'''
TREC question-type classification
'''
from __future__ import absolute_import, division, unicode_literals
import os
import io
import logging
import numpy as np
from senteval.tools.validation import KFoldClassifier
class TRECEval(object):
def __init__(self, task_path, seed=1111):
logging.info('***** Transfer task : TREC *****\n\n')
self.seed = seed
self.train = self.loadFile(os.path.join(task_path, 'train_5500.label'))
self.test = self.loadFile(os.path.join(task_path, 'TREC_10.label'))
def do_prepare(self, params, prepare):
samples = self.train['X'] + self.test['X']
return prepare(params, samples)
def loadFile(self, fpath):
trec_data = {'X': [], 'y': []}
tgt2idx = {'ABBR': 0, 'DESC': 1, 'ENTY': 2,
'HUM': 3, 'LOC': 4, 'NUM': 5}
with io.open(fpath, 'r', encoding='latin-1') as f:
for line in f:
target, sample = line.strip().split(':', 1)
sample = sample.split(' ', 1)[1].split()
assert target in tgt2idx, target
trec_data['X'].append(sample)
trec_data['y'].append(tgt2idx[target])
return trec_data
def run(self, params, batcher):
train_embeddings, test_embeddings = [], []
# Sort to reduce padding
sorted_corpus_train = sorted(zip(self.train['X'], self.train['y']),
key=lambda z: (len(z[0]), z[1]))
train_samples = [x for (x, y) in sorted_corpus_train]
train_labels = [y for (x, y) in sorted_corpus_train]
sorted_corpus_test = sorted(zip(self.test['X'], self.test['y']),
key=lambda z: (len(z[0]), z[1]))
test_samples = [x for (x, y) in sorted_corpus_test]
test_labels = [y for (x, y) in sorted_corpus_test]
# Get train embeddings
for ii in range(0, len(train_labels), params.batch_size):
batch = train_samples[ii:ii + params.batch_size]
embeddings = batcher(params, batch)
train_embeddings.append(embeddings)
train_embeddings = np.vstack(train_embeddings)
logging.info('Computed train embeddings')
# Get test embeddings
for ii in range(0, len(test_labels), params.batch_size):
batch = test_samples[ii:ii + params.batch_size]
embeddings = batcher(params, batch)
test_embeddings.append(embeddings)
test_embeddings = np.vstack(test_embeddings)
logging.info('Computed test embeddings')
config_classifier = {'nclasses': 6, 'seed': self.seed,
'usepytorch': params.usepytorch,
'classifier': params.classifier,
'kfold': params.kfold}
clf = KFoldClassifier({'X': train_embeddings,
'y': np.array(train_labels)},
{'X': test_embeddings,
'y': np.array(test_labels)},
config_classifier)
devacc, testacc, _ = clf.run()
logging.debug('\nDev acc : {0} Test acc : {1} \
for TREC\n'.format(devacc, testacc))
return {'devacc': devacc, 'acc': testacc,
'ndev': len(self.train['X']), 'ntest': len(self.test['X'])}
|