Repository navigation
Expand file tree
/
Copy pathtrain.py
More file actions
170 lines (127 loc) · 5.71 KB
/
Copy pathtrain.py
File metadata and controls
170 lines (127 loc) · 5.71 KB
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
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
'''Train CIFAR10 with PyTorch.'''
from __future__ import print_function
import torch
import torch.optim as optim
from torch import nn
import utils
import numpy as np
import os
import argparse
from torch.autograd import Variable
import shutil
from time import time
parser = argparse.ArgumentParser(description='Trin DNN with SBN modeules')
parser.add_argument('--lr', default=0.001, type=float, help='Initial learning rate')
parser.add_argument('--epochs', default=200, type=int, help='Number of epochs')
parser.add_argument('--bs', default=16, type=int, help='Batch size')
parser.add_argument('--test_bs', default=200, type=int, help='Batch size for test dataloader')
parser.add_argument('--decrease_from', default=1, type=int, help='Epoch to decrease learning_rate linear to 0 from')
parser.add_argument('--log_dir', help='Directory for logging')
parser.add_argument('--augmentation', dest='augmentation', action='store_true',
help='Eigther use or not data augmentation for cifar')
parser.add_argument('--no-augmentation', dest='augmentation', action='store_false')
parser.set_defaults(augmentation=True)
parser.add_argument('--model', '-m', default='ResNet18', help='Model to train')
parser.add_argument('--k', '-k', default=1, type=float, help='Model size for VGG')
parser.add_argument('--dropout', type=float, nargs='+', default=None,
help='Dropout rate, include dropout layer if > 0.')
parser.add_argument('--data', default='cifar', help='Dataset, one of: cifar, cifar5, mnist, not-mnist')
parser.add_argument('--seed', default=42, type=int)
parser.add_argument('--epoch_tune_bn', default=5, type=int,
help='Number of epochs to tune SBN approximation parameters')
args = parser.parse_args()
args.script = os.path.basename(__file__)
def save_checkpoint(state, is_best, epoch=None):
if epoch:
torch.save(state, '{}/model-{}'.format(args.log_dir, epoch))
if is_best:
shutil.copyfile('{}/model-{}'.format(args.log_dir, epoch), '{}/best_model'.format(args.log_dir))
else:
torch.save(state, '{}/model'.format(args.log_dir))
if is_best:
shutil.copyfile('{}/model'.format(args.log_dir), '{}/best_model'.format(args.log_dir))
def adjust_learning_rate(optimizer, lr):
"""Sets the learning rate to the initial LR decayed by 10 every 30 epochs"""
for param_group in optimizer.param_groups:
param_group['lr'] = lr
def lr_linear(epoch):
return max(0, (args.lr * np.minimum((args.decrease_from - epoch) * 1. / (args.epochs - args.decrease_from) + 1, 1.)))
use_cuda = torch.cuda.is_available()
best_acc = 0 # best test accuracy
start_epoch = 0 # start from epoch 0 or last checkpoint epoch
torch.cuda.manual_seed_all(args.seed)
torch.manual_seed(args.seed)
np.random.seed(args.seed)
NCLASSES = 10
if args.data == 'cifar5':
NCLASSES = 5
print('==> Preparing data..')
trainloader, testloader = utils.get_dataloader(data=args.data, train_bs=args.bs, test_bs=args.test_bs,
augmentation=args.augmentation)
print('==> Building model..')
net = utils.get_model(n_classes=NCLASSES, **vars(args))
if use_cuda:
net.cuda()
net = torch.nn.DataParallel(net, device_ids=range(torch.cuda.device_count()))
criterion = nn.CrossEntropyLoss().cuda()
optimizer = optim.Adam(net.parameters(), lr=args.lr)
with open('{}/log'.format(args.log_dir), 'w') as f:
f.write('#{}\n'.format(utils.make_description(args)))
f.write('epoch,loss,train_acc,test_acc\n')
best_test_acc = 0
counter = utils.AccCounter()
model_args = vars(args)
model_args['n_classes'] = NCLASSES
print(net)
for epoch in range(args.epochs):
counter.flush()
t0 = time()
lr = lr_linear(epoch)
adjust_learning_rate(optimizer, lr)
net.train()
utils.set_strategy(net, 'batch')
training_loss = 0
for i, (inputs, labels) in enumerate(trainloader):
inputs, labels = Variable(inputs.cuda(async=True)), Variable(labels.cuda(async=True))
optimizer.zero_grad()
outputs = net(inputs)
counter.add(outputs.data.cpu().numpy(), labels.data.cpu().numpy())
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
training_loss += loss.cpu().data.numpy()[0] * float(inputs.size(0))
train_acc = counter.acc()
counter.flush()
test_loss = 0
net.eval()
utils.set_strategy(net, 'running')
for _, (inputs, labels) in enumerate(testloader):
inputs, labels = Variable(inputs.cuda(async=True)), Variable(labels.cuda(async=True))
outputs = net(inputs)
loss = criterion(outputs, labels)
test_loss += utils.to_np(loss) * float(inputs.size(0))
counter.add(utils.to_np(outputs), utils.to_np(labels))
print(' -- Epoch %d | time: %.4f | loss: %.4f | training acc: %.4f validation accuracy: %.4f | lr %.6f --' %
(epoch, time() - t0, training_loss, train_acc, counter.acc(), lr))
save_checkpoint({
'epoch': epoch + 1,
'state_dict': net.module.state_dict() if use_cuda else net.state_dict(),
'test_accuracy': counter.acc(),
'optimizer': optimizer.state_dict(),
'name': args.model,
'model_args': model_args,
'script_args': vars(args)
}, best_test_acc < counter.acc())
with open('{}/log'.format(args.log_dir), 'a') as f:
f.write('{},{},{},{}\n'.format(epoch, training_loss, train_acc, counter.acc()))
if best_test_acc < counter.acc():
best_test_acc = counter.acc()
print('==> Finish Training')
print('==> Tune SBN parameters')
net.train()
utils.set_strategy(net, 'batch')
for _ in range(args.epoch_tune_bn):
for inputs, _ in trainloader:
inputs = Variable(inputs.cuda(async=True))
net(inputs)
print('==> Done.')