-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathminist.py
186 lines (160 loc) · 7.73 KB
/
minist.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
178
179
180
181
182
183
184
185
186
import argparse
import os
import logging
import random
import shutil
import numpy as np
from distutils.version import LooseVersion
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torch.backends.cudnn as cudnn
from torchvision import datasets, transforms
from utils.util import set_logging
from nets.simpleNet import Net
# Training settings
parser = argparse.ArgumentParser(description='PyTorch MNIST Example')
parser.add_argument('--base_dir', type=str, default='RES-train-minist',
help='location of the log path')
parser.add_argument('--batch_size', type=int, default=512, metavar='N',
help='input batch size for training (default: 64)')
parser.add_argument('--test_batch_size', type=int, default=150, metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs', type=int, default=10, metavar='N',
help='number of epochs to train (default: 10)')
parser.add_argument('--lr', type=float, default=0.01, metavar='LR',
help='learning rate (default: 0.01)')
parser.add_argument('--momentum', type=float, default=0.5, metavar='M',
help='SGD momentum (default: 0.5)')
parser.add_argument('--seed', type=int, default=42, metavar='S',
help='random seed (default: 42)')
parser.add_argument('--log_interval', type=int, default=10, metavar='N',
help='how many batches to wait before logging training status')
parser.add_argument('--use_mixed_precision', action='store_true', default=True,
help='use mixed precision for training')
parser.add_argument('--data_dir', type=str, default='./data',
help='location of the training dataset in the local filesystem (will be downloaded if needed)')
def main(args):
seed = args.seed
set_random_seed(seed)
logging.info(f'set random seed: {seed}')
if (args.use_mixed_precision and LooseVersion(torch.__version__)
< LooseVersion('1.6.0')):
raise ValueError("""Mixed precision is using torch.cuda.amp.autocast(),
which requires torch >= 1.6.0""")
data_dir = args.data_dir
train_dataset = \
datasets.MNIST(data_dir, train=True, download=True,
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.3081,))
]))
def worker_init_fn(worker_id):#为每一个worker设置固定的seed
set_random_seed(seed + worker_id)
train_loader = torch.utils.data.DataLoader(
train_dataset, batch_size=args.batch_size, shuffle=True, num_workers=16, pin_memory=True, drop_last=True, worker_init_fn=worker_init_fn)
test_dataset = \
datasets.MNIST(data_dir, train=False, transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.3081,))
]))
test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=args.test_batch_size, shuffle=False, num_workers=16, pin_memory=True)
model = Net()
# model = nn.DataParallel(model)
# Move model to GPU.
model.to(device=args.device)
optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=args.momentum)
if args.use_mixed_precision:
# Initialize scaler in global scale
scaler = torch.cuda.amp.GradScaler()
for epoch in range(1, args.epochs + 1):
if args.use_mixed_precision:
train_mixed_precision(epoch, scaler, model, optimizer, train_loader, train_dataset)
else:
train_epoch(epoch, model, optimizer, train_loader, train_dataset)
# Keep test in full precision since computation is relatively light.
test_acc = test(model, test_loader, test_dataset)
def train_mixed_precision(epoch, scaler, model, optimizer, train_loader, train_dataset):
model.train()
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.cuda(), target.cuda()
optimizer.zero_grad()
with torch.cuda.amp.autocast():
output = model(data)
loss = F.nll_loss(output, target)
scaler.scale(loss).backward()
scaler.step(optimizer)
scaler.update()
if batch_idx % args.log_interval == 0:
logging.info('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}\tLoss Scale: {}'.format(
epoch, batch_idx * len(data), len(train_dataset),
100. * batch_idx / len(train_loader), loss.item(), scaler.get_scale()))
def train_epoch(epoch, model, optimizer, train_loader, train_dataset):
model.train()
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.cuda(), target.cuda()
optimizer.zero_grad()
output = model(data)
loss = F.nll_loss(output, target)
loss.backward()
optimizer.step()
if batch_idx % args.log_interval == 0:
logging.info('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
epoch, batch_idx * len(data), len(train_dataset),
100. * batch_idx / len(train_loader), loss.item()))
def test(model, test_loader, test_dataset):
model.eval()
test_loss = 0.
test_accuracy = 0.
for data, target in test_loader:
data, target = data.cuda(), target.cuda()
output = model(data)
# sum up batch loss
test_loss += F.nll_loss(output, target, size_average=False).item()
# get the index of the max log-probability
pred = output.data.max(1, keepdim=True)[1]
test_accuracy += pred.eq(target.data.view_as(pred)).cpu().float().sum()
test_loss /= len(test_dataset)
test_accuracy /= len(test_dataset)
test_accuracy = test_accuracy.item()
logging.info('\nTest set: Average loss: {:.4f}, Accuracy: {:.2f}%\n'.format(
test_loss, 100. * test_accuracy))
return test_accuracy
def set_random_seed(seed_num):
cudnn.benchmark = False #False：不进行最优卷积搜索，以控制CUDNN种子
cudnn.deterministic = True #True ：调用相同的CuDNN的卷积操作，以控制CUDNN种子
random.seed(seed_num) #为python设置随机种子
np.random.seed(seed_num) #为numpy设置随机种子
torch.manual_seed(seed_num) # 为CPU设置随机种子
torch.cuda.manual_seed(seed_num) # 为当前GPU设置随机种子 torch.cuda.manual_seed_all(seed) # 为所有GPU设置随机种子
def backup_code(base_dir):
###备份当前train代码文件及dataset代码文件
code_path = os.path.join(base_dir, 'code')
if not os.path.exists(code_path):
os.makedirs(code_path)
train_name = os.path.basename(__file__)
net_name = 'simpleNet.py'
shutil.copy('nets/' + net_name, code_path + '/' + net_name)
shutil.copy(train_name, code_path + '/' + train_name)
if __name__ == '__main__':
args = parser.parse_args()
if not os.path.exists(args.base_dir):
os.makedirs(args.base_dir)
backup_code(args.base_dir)
log_path = os.path.join(args.base_dir, 'training.log')
set_logging(log_path=log_path)
"""选择GPU ID"""
gpu_list = [0] #[0,1]
gpu_list_str = ','.join(map(str, gpu_list))
os.environ.setdefault("CUDA_VISIBLE_DEVICES", gpu_list_str)
args.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
logging.info(f'Using device : {args.device}\n'
f'\tGPU ID is [{os.environ["CUDA_VISIBLE_DEVICES"]}],using {torch.cuda.device_count()} device\n'
f'\tdevice name:{torch.cuda.get_device_name(0)}')
try:
logging.info(args)
main(args)
except Exception as exception:
logging.exception(exception)
raise