-
Notifications
You must be signed in to change notification settings - Fork 124
/
train_trades_cifar10.py
177 lines (153 loc) · 7.05 KB
/
train_trades_cifar10.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 os
import argparse
import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision
import torch.optim as optim
from torchvision import datasets, transforms
from models.wideresnet import *
from models.resnet import *
from trades import trades_loss
parser = argparse.ArgumentParser(description='PyTorch CIFAR TRADES Adversarial Training')
parser.add_argument('--batch-size', type=int, default=128, metavar='N',
help='input batch size for training (default: 128)')
parser.add_argument('--test-batch-size', type=int, default=128, metavar='N',
help='input batch size for testing (default: 128)')
parser.add_argument('--epochs', type=int, default=76, metavar='N',
help='number of epochs to train')
parser.add_argument('--weight-decay', '--wd', default=2e-4,
type=float, metavar='W')
parser.add_argument('--lr', type=float, default=0.1, metavar='LR',
help='learning rate')
parser.add_argument('--momentum', type=float, default=0.9, metavar='M',
help='SGD momentum')
parser.add_argument('--no-cuda', action='store_true', default=False,
help='disables CUDA training')
parser.add_argument('--epsilon', default=0.031,
help='perturbation')
parser.add_argument('--num-steps', default=10,
help='perturb number of steps')
parser.add_argument('--step-size', default=0.007,
help='perturb step size')
parser.add_argument('--beta', default=6.0,
help='regularization, i.e., 1/lambda in TRADES')
parser.add_argument('--seed', type=int, default=1, metavar='S',
help='random seed (default: 1)')
parser.add_argument('--log-interval', type=int, default=100, metavar='N',
help='how many batches to wait before logging training status')
parser.add_argument('--model-dir', default='./model-cifar-wideResNet',
help='directory of model for saving checkpoint')
parser.add_argument('--save-freq', '-s', default=1, type=int, metavar='N',
help='save frequency')
args = parser.parse_args()
# settings
model_dir = args.model_dir
if not os.path.exists(model_dir):
os.makedirs(model_dir)
use_cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
device = torch.device("cuda" if use_cuda else "cpu")
kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}
# setup data loader
transform_train = transforms.Compose([
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
])
transform_test = transforms.Compose([
transforms.ToTensor(),
])
trainset = torchvision.datasets.CIFAR10(root='../data', train=True, download=True, transform=transform_train)
train_loader = torch.utils.data.DataLoader(trainset, batch_size=args.batch_size, shuffle=True, **kwargs)
testset = torchvision.datasets.CIFAR10(root='../data', train=False, download=True, transform=transform_test)
test_loader = torch.utils.data.DataLoader(testset, batch_size=args.test_batch_size, shuffle=False, **kwargs)
def train(args, model, device, train_loader, optimizer, epoch):
model.train()
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.to(device), target.to(device)
optimizer.zero_grad()
# calculate robust loss
loss = trades_loss(model=model,
x_natural=data,
y=target,
optimizer=optimizer,
step_size=args.step_size,
epsilon=args.epsilon,
perturb_steps=args.num_steps,
beta=args.beta)
loss.backward()
optimizer.step()
# print progress
if batch_idx % args.log_interval == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
epoch, batch_idx * len(data), len(train_loader.dataset),
100. * batch_idx / len(train_loader), loss.item()))
def eval_train(model, device, train_loader):
model.eval()
train_loss = 0
correct = 0
with torch.no_grad():
for data, target in train_loader:
data, target = data.to(device), target.to(device)
output = model(data)
train_loss += F.cross_entropy(output, target, size_average=False).item()
pred = output.max(1, keepdim=True)[1]
correct += pred.eq(target.view_as(pred)).sum().item()
train_loss /= len(train_loader.dataset)
print('Training: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)'.format(
train_loss, correct, len(train_loader.dataset),
100. * correct / len(train_loader.dataset)))
training_accuracy = correct / len(train_loader.dataset)
return train_loss, training_accuracy
def eval_test(model, device, test_loader):
model.eval()
test_loss = 0
correct = 0
with torch.no_grad():
for data, target in test_loader:
data, target = data.to(device), target.to(device)
output = model(data)
test_loss += F.cross_entropy(output, target, size_average=False).item()
pred = output.max(1, keepdim=True)[1]
correct += pred.eq(target.view_as(pred)).sum().item()
test_loss /= len(test_loader.dataset)
print('Test: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)'.format(
test_loss, correct, len(test_loader.dataset),
100. * correct / len(test_loader.dataset)))
test_accuracy = correct / len(test_loader.dataset)
return test_loss, test_accuracy
def adjust_learning_rate(optimizer, epoch):
"""decrease the learning rate"""
lr = args.lr
if epoch >= 75:
lr = args.lr * 0.1
if epoch >= 90:
lr = args.lr * 0.01
if epoch >= 100:
lr = args.lr * 0.001
for param_group in optimizer.param_groups:
param_group['lr'] = lr
def main():
# init model, ResNet18() can be also used here for training
model = WideResNet().to(device)
optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=args.momentum, weight_decay=args.weight_decay)
for epoch in range(1, args.epochs + 1):
# adjust learning rate for SGD
adjust_learning_rate(optimizer, epoch)
# adversarial training
train(args, model, device, train_loader, optimizer, epoch)
# evaluation on natural examples
print('================================================================')
eval_train(model, device, train_loader)
eval_test(model, device, test_loader)
print('================================================================')
# save checkpoint
if epoch % args.save_freq == 0:
torch.save(model.state_dict(),
os.path.join(model_dir, 'model-wideres-epoch{}.pt'.format(epoch)))
torch.save(optimizer.state_dict(),
os.path.join(model_dir, 'opt-wideres-checkpoint_epoch{}.tar'.format(epoch)))
if __name__ == '__main__':
main()