-
Notifications
You must be signed in to change notification settings - Fork 3
/
train_cifar_mean.py
177 lines (145 loc) · 6.24 KB
/
train_cifar_mean.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
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
169
170
171
172
173
174
175
176
177
from __future__ import print_function
import sys
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import torch.backends.cudnn as cudnn
import torchvision
import torchvision.transforms as transforms
import pytorch_cifar.models.resnet_unmodified as resnet
import os
import argparse
from prog_bar import progress_bar
import utils_color_mean as utils
parser = argparse.ArgumentParser(description='PyTorch CIFAR Training')
parser.add_argument('--lr', default=0.01, type=float, help='learning rate')
parser.add_argument('--keep', default=75, type=int, help='pixels to keep')
parser.add_argument('--regularization', default=0.0005, type=float, help='weight decay')
parser.add_argument('--model', default='resnet18', type=str, help='model')
parser.add_argument('--resume', help='resume from checkpoint')
parser.add_argument('--seed', default=0, type=int, help='random seed')
parser.add_argument('--testsamples', default=1000, type=int, help='samples for test')
args = parser.parse_args()
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
checkpoint_dir = 'checkpoints'
if not os.path.exists('./checkpoints'):
os.makedirs('./checkpoints')
checkpoint_file = f'./{checkpoint_dir}/cifar_mean_lr_{args.lr}_regularization_{args.regularization}_model_{args.model}_keep_{args.keep}_epoch_{{}}.pth'
print("==> Checkpoint directory", checkpoint_dir)
print("==> Saving to", checkpoint_file)
device = 'cuda' if torch.cuda.is_available() else 'cpu'
start_epoch = 0 # start from epoch 0 or last checkpoint epoch
# Data
print('==> Preparing data..')
transform_train = transforms.Compose([
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor()
# transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
transform_test = transforms.Compose([
transforms.ToTensor(),
# transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
trainset = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transform_train)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=128, shuffle=True, num_workers=2)
testset = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transform_test)
testloader = torch.utils.data.DataLoader(testset, batch_size=2, shuffle=False, num_workers=2)
nomtestloader = torch.utils.data.DataLoader(testset, batch_size=512, shuffle=False, num_workers=2)
# Model
print('==> Building model..')
if (args.model == 'resnet50'):
net = resnet.ResNet50()
elif (args.model == 'resnet18'):
net = resnet.ResNet18()
net = net.to(device)
if device == 'cuda':
net = torch.nn.DataParallel(net)
cudnn.benchmark = True
criterion = nn.CrossEntropyLoss()
if args.resume:
# Load checkpoint.
print('==> Resuming from checkpoint..')
assert os.path.isdir(checkpoint_dir), 'Error: no checkpoint directory found!'
resume_file = '{}/{}'.format(checkpoint_dir, args.resume)
assert os.path.isfile(resume_file)
checkpoint = torch.load(resume_file)
net.load_state_dict(checkpoint['net'])
start_epoch = checkpoint['epoch']+1
checkpoint_file = './{}/cifar_mean_lr_{}_regularization_{}_model_{}_keep_{}_epoch_{}_resume_{}.pth'.format(checkpoint_dir, args.lr,args.regularization,args.model, args.keep, '{}', args.resume)
optimizer = optim.SGD(net.parameters(), lr=args.lr, momentum=0.9, weight_decay=args.regularization)
# Training
def train(epoch):
print('\nEpoch: %d' % epoch)
net.train()
train_loss = 0
correct = 0
total_epsilon = 0
total = 0
for batch_idx, (inputs, targets) in enumerate(trainloader):
inputs, targets = inputs.to(device), targets.to(device)
optimizer.zero_grad()
outputs = net(utils.random_mask_batch_one_sample(inputs, args.keep, reuse_noise=True))
_, predicted = outputs.max(1)
correct += predicted.eq(targets).sum().item()
loss = criterion(outputs, targets)
loss.backward()
optimizer.step()
total += targets.size(0)
train_loss += loss.item()
progress_bar(batch_idx, len(trainloader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
% (train_loss/(batch_idx+1), 100.*correct/total, correct, total))
def test_nominal(epoch):
print('\nEpoch: %d' % epoch)
net.eval()
test_loss = 0
correct = 0
total = 0
for batch_idx, (inputs, targets) in enumerate(nomtestloader):
inputs, targets = inputs.to(device), targets.to(device)
with torch.no_grad():
#breakpoint()
outputs = net(utils.random_mask_batch_one_sample(inputs, args.keep, reuse_noise=True))
loss = criterion(outputs, targets)
_, predicted = outputs.max(1)
correct += predicted.eq(targets).sum().item()
test_loss += loss.item()
total += targets.size(0)
progress_bar(batch_idx, len(nomtestloader), 'Nominal Test Loss: %.3f | Acc: %.3f%% (%d/%d)'
% (test_loss/(batch_idx+1), 100.*correct/total, correct, total))
def test(epoch):
net.eval()
test_loss = 0
correct = 0
total = 0
for batch_idx, (inputs, targets) in enumerate(testloader):
inputs, targets = inputs.to(device), targets.to(device)
with torch.no_grad():
#breakpoint()
outputs = utils.avg_hard_forward(inputs, net, args.testsamples, args.keep)
loss = criterion(outputs, targets)
_, predicted = outputs.max(1)
correct += predicted.eq(targets).sum().item()
test_loss += loss.item()
total += targets.size(0)
progress_bar(batch_idx, len(testloader), 'Test Loss: %.3f | Acc: %.3f%% (%d/%d)'
% (test_loss/(batch_idx+1), 100.*correct/total, correct, total))
# Save checkpoint.
if (epoch == 199 or epoch == 399):
acc = 100.*correct/total
print('Saving..')
state = {
'net': net.state_dict(),
'acc': acc,
'epoch': epoch
}
if not os.path.isdir(checkpoint_dir):
os.mkdir(checkpoint_dir)
torch.save(state, checkpoint_file.format(epoch))
for epoch in range(start_epoch, start_epoch+200):
train(epoch)
test_nominal(epoch)
if (epoch == 199 or epoch == 399):
test(epoch)