File size: 2,831 Bytes
226675b |
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 |
import torch
import torch.optim as optim
from catalyst.contrib.nn import Lookahead
from catalyst import utils
import math
class lambdax:
def __init__(self, cfg):
self.cfg = cfg
@staticmethod
def lambda_epoch(self, epoch):
return math.pow(1 - epoch / self.cfg.max_epoch, self.cfg.poly_exp)
def get_optimizer(cfg, net):
if cfg.lr_mode == 'multi':
layerwise_params = {"backbone.*": dict(lr=cfg.backbone_lr, weight_decay=cfg.backbone_weight_decay)}
net_params = utils.process_model_params(net, layerwise_params=layerwise_params)
else:
net_params = net.parameters()
if cfg.type == "AdamW":
optimizer = optim.AdamW(net_params, lr=cfg.lr, weight_decay=cfg.weight_decay)
# optimizer = Lookahead(optimizer)
elif cfg.type == "SGD":
optimizer = optim.SGD(net_params, lr=cfg.lr, weight_decay=cfg.weight_decay, momentum=cfg.momentum,
nesterov=False)
else:
raise KeyError("The optimizer type ( %s ) doesn't exist!!!" % cfg.type)
return optimizer
def get_scheduler(cfg, optimizer):
if cfg.type == 'Poly':
lambda1 = lambda epoch: math.pow(1 - epoch / cfg.max_epoch, cfg.poly_exp)
scheduler = optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=lambda1)
elif cfg.type == 'CosineAnnealingLR':
scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=cfg.max_epoch, eta_min=1e-6)
elif cfg.type == 'linear':
def lambda_rule(epoch):
lr_l = 1.0 - epoch / float(cfg.max_epoch + 1)
return lr_l
scheduler = optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=lambda_rule)
elif cfg.type == 'step':
scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=cfg.step_size, gamma=cfg.gamma)
elif cfg.type == 'multistep':
scheduler = optim.lr_scheduler.MultiStepLR(optimizer, milestones=cfg.milestones, gamma=cfg.gamma)
elif cfg.type == 'reduce':
scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, 'max', patience=cfg.patience, factor=cfg.factor)
else:
raise KeyError("The scheduler type ( %s ) doesn't exist!!!" % cfg.type)
return scheduler
def build_optimizer(cfg, net):
optimizer = get_optimizer(cfg.optimizer, net)
scheduler = get_scheduler(cfg.scheduler, optimizer)
# if cfg.type == 'Poly':
# lambda1 = lambda epoch: math.pow(1 - epoch / cfg.max_epoch, cfg.poly_exp)
# scheduler = optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=lambda1)
# elif cfg.type == 'CosineAnnealingLR':
# scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=cfg.max_epoch, eta_min=1e-6)
# else:
# raise KeyError("The scheduler type ( %s ) doesn't exist!!!" % cfg.type)
return optimizer, scheduler
|