-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
70 lines (58 loc) · 3.27 KB
/
utils.py
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
import torch
from torch import optim as optim
from torch.optim import lr_scheduler
import os
def get_Optimizer(opt, model):
if opt.optimizer.lower() == 'sgd':
optimizer = optim.SGD(model.parameters(), lr=opt.lr, momentum=opt.momentum, weight_decay=opt.weight_decay)
elif opt.optimizer.lower() == 'adam':
optimizer = optim.Adam([{'params': model.parameters(), 'initial_lr': opt.lr}], lr=opt.lr, betas=(opt.beta1, opt.beta2), eps=opt.eps)
elif opt.optimizer.lower() == 'adamw':
optimizer = optim.AdamW([{'params': model.parameters(), 'initial_lr': opt.lr}], lr=opt.lr, betas=(opt.beta1, opt.beta2), eps=opt.eps, weight_decay=opt.weight_decay)
else:
raise Exception('Unexpected Optimizer of {}'.format(opt.optimizer))
return optimizer
def get_Scheduler(opt, optimizer):
if opt.scheduler == 'step':
scheduler = lr_scheduler.StepLR(optimizer, step_size=opt.lr_step, gamma=opt.gamma, last_epoch=-1 if opt.start_epoch==0 else opt.start_epoch, verbose=True)
elif opt.scheduler == 'multistep':
scheduler = lr_scheduler.MultiStepLR(optimizer, milestones=[100, 200], gamma=opt.gamma, last_epoch=-1 if opt.start_epoch==0 else opt.start_epoch, verbose=True)
elif opt.scheduler == 'cosine':
scheduler = lr_scheduler.CosineAnnealingLR(optimizer, T_max=50, eta_min=1e-5, last_epoch=-1 if opt.start_epoch==0 else opt.start_epoch, verbose=True)
elif opt.scheduler == 'cycle':
scheduler = torch.optim.lr_scheduler.CyclicLR(optimizer, base_lr=1e-6, max_lr=opt.lr, step_size_up=50, step_size_down=50,
mode='triangular', gamma=1.0, scale_fn=None, scale_mode='cycle',
cycle_momentum=False, base_momentum=0.8, max_momentum=0.9, last_epoch=-1 if opt.start_epoch==0 else opt.start_epoch, verbose=True)
else:
raise Exception('Unexpected Scheduler of {}'.format(opt.scheduler))
return scheduler
class Parser:
def __init__(self, parser):
self.__parser = parser
self.__args = parser.parse_args()
def get_parser(self):
return self.__parser
def get_arguments(self):
return self.__args
def write_args(self, log_name):
params_dict = vars(self.__args)
log_dir = os.path.join('./%s' % (params_dict['log_dir']))
if not os.path.exists(log_dir):
os.makedirs(log_dir)
args_name = os.path.join(log_dir, '%s_args.txt' % (log_name))
with open(args_name, 'wt') as args_fid:
args_fid.write('----' * 10 + '\n')
args_fid.write('{0:^40}'.format('PARAMETER TABLES') + '\n')
args_fid.write('----' * 10 + '\n')
for k, v in sorted(params_dict.items()):
args_fid.write('{}'.format(str(k)) + ' : ' + ('{0:>%d}' % (35 - len(str(k)))).format(str(v)) + '\n')
args_fid.write('----' * 10 + '\n')
def print_args(self, name='PARAMETER TABLES'):
params_dict = vars(self.__args)
print('----' * 10)
print('{0:^40}'.format(name))
print('----' * 10)
for k, v in sorted(params_dict.items()):
if '__' not in str(k):
print('{}'.format(str(k)) + ' : ' + ('{0:>%d}' % (35 - len(str(k)))).format(str(v)))
print('----' * 10)