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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
| import os
import torch
import torch.nn as nn
import torch.optim as optim
import matplotlib.pyplot as plt
import torch.nn.functional as F
from torch.utils.data import Dataset, DataLoader
from torchvision import datasets, transforms
from time import time
import argparse
from tqdm import tqdm
import wandb
def compute_accuracy(preds: torch.Tensor, target: torch.Tensor, from_logits: bool = True) -> float:
"""
计算准确率。
参数:
- preds: 模型输出，形状为 [batch_size, num_classes] 或 [batch_size]（如果已经是类标）
- target: 真实标签，形状为 [batch_size]
- from_logits: 若为 True，表示 preds 是 logits，需要取 argmax；若为 False，表示 preds 已是预测标签
返回:
- 准确率（float）
"""
if from_logits:
pred_labels = torch.argmax(preds, dim=1)
else:
pred_labels = preds
correct = (pred_labels == target).sum().item()
total = target.numel()
return correct / total
class ConvNet(nn.Module):
def __init__(self):
super(ConvNet, self).__init__()
self.features = nn.Sequential(
nn.Conv2d(1, 32, 3, 1),
nn.ReLU(),
nn.Conv2d(32, 64, 3, 1),
nn.ReLU(),
nn.MaxPool2d(2),
nn.Dropout(0.25)
)
self.classifier = nn.Sequential(
nn.Linear(9216, 128),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(128, 10)
)
def forward(self, x):
x = self.features(x)
x = torch.flatten(x, 1)
x = self.classifier(x)
output = F.log_softmax(x, dim=1)
return output
class MyDataset(Dataset):
def __init__(self, data):
self.data = data
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
image, label = self.data[idx]
return image, label
class Trainer:
def __init__(
self,
model: torch.nn.Module,
optimizer: torch.optim.Optimizer,
gpu_id: int=0,
) -> None:
self.gpu_id = gpu_id
self.model = model.to(gpu_id)
self.optimizer = optimizer
def save_checkpoint(self):
ckp = self.model.state_dict()
PATH = "checkpoint.pt"
torch.save(ckp, PATH)
print(f"Training checkpoint saved at {PATH}")
def _train_batch(self, source, targets):
self.model.train()
self.optimizer.zero_grad()
source = source.to(self.gpu_id)
targets = targets.to(self.gpu_id)
output = self.model(source)
loss = F.cross_entropy(output, targets)
loss.backward()
self.optimizer.step()
batch_loss = loss.detach().item()
return batch_loss
def _test_batch(self, source, targets):
self.model.eval()
with torch.no_grad():
source = source.to(self.gpu_id)
targets = targets.to(self.gpu_id)
output = self.model(source)
loss = F.cross_entropy(output, targets)
batch_loss = loss.detach().item()
accuracy = compute_accuracy(output, targets)
return batch_loss, accuracy
def train_epoch(self, train_dataloader):
total_loss = 0.0
num_batches = 0
for source, targets in tqdm(train_dataloader, desc="Training", mininterval=1):
loss = self._train_batch(source, targets)
total_loss += loss
num_batches += 1
avg_loss = total_loss / num_batches
return avg_loss
def test_epoch(self, test_dataloader):
total_loss = 0.0
num_batches = 0
total_accuracy = 0.0
for source, targets in tqdm(test_dataloader, desc="Testing", mininterval=1):
loss, accuracy = self._test_batch(source, targets)
total_loss += loss
total_accuracy += accuracy
num_batches += 1
avg_loss = total_loss / num_batches
avg_accuracy = total_accuracy / num_batches
return avg_loss, avg_accuracy
def prepare_dataset():
transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.3081,))
])
train_data = datasets.MNIST(
root = './mnist',
train=True,
transform = transform,
)
train_set = MyDataset(train_data)
test_data = datasets.MNIST(
root = './mnist',
train=False,
transform = transform,
)
test_set = MyDataset(test_data)
return train_set, test_set
def prepare_dataloader(dataset: Dataset, batch_size: int):
return DataLoader(
dataset,
batch_size=batch_size,
pin_memory=True,
shuffle=False,
)
def arg_parser():
parser = argparse.ArgumentParser(description="MNIST Training Script")
parser.add_argument("--epochs", type=int, default=5, help="Number of training epochs")
parser.add_argument("--batch_size", type=int, default=512, help="Batch size for training")
parser.add_argument("--lr", type=float, default=0.0005, help="Learning rate")
parser.add_argument("--lr_decay_step_num", type=int, default=1, help="Step size for learning rate decay")
parser.add_argument("--lr_decay_factor", type=float, default=0.5, help="Factor by which learning rate is decayed")
parser.add_argument("--cuda_id", type=int, default=0, help="CUDA device ID to use")
parser.add_argument('--save_every', type=int, default=1, help='How often to save a snapshot')
return parser.parse_args()
if __name__ == "__main__":
args = arg_parser()
print(f"Training arguments: {args}")
EPOCHS = args.epochs
BATCH_SIZE = args.batch_size
LR = args.lr
LR_DECAY_STEP_NUM = args.lr_decay_step_num
LR_DECAY_FACTOR = args.lr_decay_factor
CUDA_ID = args.cuda_id
DEVICE = torch.device(f"cuda:{CUDA_ID}")
SAVE_EVERY = args.save_every
wandb_run = wandb.init(
project="mnist",
config={
"epochs": EPOCHS,
"batch_size": BATCH_SIZE,
"lr": LR,
"lr_decay_step_num": LR_DECAY_STEP_NUM,
"lr_decay_factor": LR_DECAY_FACTOR,
}
)
train_set, test_set = prepare_dataset()
print(f"Train dataset size: {len(train_set)}")
print(f"Test dataset size: {len(test_set)}")
train_dataloader = prepare_dataloader(dataset=train_set, batch_size=BATCH_SIZE)
test_dataloader = prepare_dataloader(dataset=test_set, batch_size=BATCH_SIZE)
model = ConvNet()
optimizer = optim.Adam(model.parameters(), lr=1e-3)
scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=LR_DECAY_STEP_NUM, gamma=LR_DECAY_FACTOR)
trainer = Trainer(model, optimizer, CUDA_ID)
for epoch in range(EPOCHS):
print(f"Epoch {epoch+1}/{EPOCHS}")
print(f'Learning Rate: {scheduler.get_last_lr()[0]}')
start_time = time()
train_loss = trainer.train_epoch(train_dataloader)
test_loss, test_accuracy = trainer.test_epoch(test_dataloader)
print(f"Train Loss: {train_loss:.4f}, Test Loss: {test_loss:.4f}, Test Accuracy: {test_accuracy:.4f}")
epoch_time = time() - start_time
current_lr = scheduler.get_last_lr()[0]
wandb_run.log({
"epoch": epoch + 1,
"train_loss": train_loss,
"test_loss": test_loss,
"test_acc": test_accuracy,
"lr": current_lr,
"epoch_time": epoch_time
})
scheduler.step()
if (epoch + 1) % SAVE_EVERY == 0:
trainer.save_checkpoint()
wandb_run.finish()
|
