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

import torch
from torch import nn


class SmallCNN(nn.Module):
    def __init__(self, num_classes: int = 10):
        super().__init__()
        self.features = nn.Sequential(
            nn.Conv2d(3, 32, 3, padding=1, bias=False),
            nn.BatchNorm2d(32),
            nn.ReLU(inplace=True),

            nn.Conv2d(32, 64, 3, stride=2, padding=1, bias=False),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),

            nn.Conv2d(64, 128, 3, stride=2, padding=1, bias=False),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),
        )
        self.pool = nn.AdaptiveAvgPool2d(1)
        self.classifier = nn.Linear(128, num_classes)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.features(x)
        x = self.pool(x)
        x = torch.flatten(x, 1)
        return self.classifier(x)  # logits，不要在这里 softmax


model = SmallCNN()
dummy = torch.randn(8, 3, 32, 32)
assert model(dummy).shape == (8, 10)

  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

"""A compact, reproducible CIFAR-10 CNN baseline used by the companion blog.

Run:
    python cnn_cifar10.py --epochs 20 --data-dir ./data

The script downloads CIFAR-10 when it is not already present.
"""

from __future__ import annotations

import argparse
import random
from pathlib import Path

import numpy as np
import torch
from torch import nn
from torch.utils.data import DataLoader, Subset
from torchvision import datasets, transforms


CIFAR10_MEAN = (0.4914, 0.4822, 0.4465)
CIFAR10_STD = (0.2470, 0.2435, 0.2616)


class SmallCNN(nn.Module):
    def __init__(self, num_classes: int = 10) -> None:
        super().__init__()
        self.features = nn.Sequential(
            nn.Conv2d(3, 32, 3, padding=1, bias=False),
            nn.BatchNorm2d(32),
            nn.ReLU(inplace=True),
            nn.Conv2d(32, 64, 3, stride=2, padding=1, bias=False),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),
            nn.Conv2d(64, 128, 3, stride=2, padding=1, bias=False),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),
        )
        self.pool = nn.AdaptiveAvgPool2d(1)
        self.classifier = nn.Linear(128, num_classes)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.features(x)
        x = self.pool(x)
        x = torch.flatten(x, 1)
        return self.classifier(x)


def seed_everything(seed: int) -> None:
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    if torch.cuda.is_available():
        torch.cuda.manual_seed_all(seed)


def make_loaders(
    data_dir: Path,
    batch_size: int,
    workers: int,
    seed: int,
    validation_size: int,
) -> tuple[DataLoader, DataLoader, DataLoader]:
    train_transform = transforms.Compose(
        [
            transforms.RandomCrop(32, padding=4),
            transforms.RandomHorizontalFlip(),
            transforms.ToTensor(),
            transforms.Normalize(CIFAR10_MEAN, CIFAR10_STD),
        ]
    )
    test_transform = transforms.Compose(
        [
            transforms.ToTensor(),
            transforms.Normalize(CIFAR10_MEAN, CIFAR10_STD),
        ]
    )
    augmented_train_set = datasets.CIFAR10(
        data_dir, train=True, download=True, transform=train_transform
    )
    evaluation_train_set = datasets.CIFAR10(
        data_dir, train=True, download=False, transform=test_transform
    )
    test_set = datasets.CIFAR10(data_dir, train=False, download=True, transform=test_transform)
    if not 1 <= validation_size < len(augmented_train_set):
        raise ValueError("validation-size must be between 1 and the training-set size")
    generator = torch.Generator().manual_seed(seed)
    indices = torch.randperm(len(augmented_train_set), generator=generator).tolist()
    validation_indices = indices[:validation_size]
    training_indices = indices[validation_size:]
    train_set = Subset(augmented_train_set, training_indices)
    validation_set = Subset(evaluation_train_set, validation_indices)
    pin_memory = torch.cuda.is_available()
    train_loader = DataLoader(
        train_set,
        batch_size=batch_size,
        shuffle=True,
        num_workers=workers,
        pin_memory=pin_memory,
        persistent_workers=workers > 0,
    )
    test_loader = DataLoader(
        test_set,
        batch_size=batch_size,
        shuffle=False,
        num_workers=workers,
        pin_memory=pin_memory,
        persistent_workers=workers > 0,
    )
    validation_loader = DataLoader(
        validation_set,
        batch_size=batch_size,
        shuffle=False,
        num_workers=workers,
        pin_memory=pin_memory,
        persistent_workers=workers > 0,
    )
    return train_loader, validation_loader, test_loader


def train_one_epoch(
    model: nn.Module,
    loader: DataLoader,
    loss_fn: nn.Module,
    optimizer: torch.optim.Optimizer,
    device: torch.device,
) -> tuple[float, float]:
    model.train()
    loss_sum = 0.0
    correct = 0
    count = 0
    for images, labels in loader:
        images = images.to(device, non_blocking=True)
        labels = labels.to(device, non_blocking=True)

        optimizer.zero_grad(set_to_none=True)
        logits = model(images)
        loss = loss_fn(logits, labels)
        loss.backward()
        optimizer.step()

        batch_size = labels.size(0)
        loss_sum += loss.item() * batch_size
        correct += (logits.argmax(dim=1) == labels).sum().item()
        count += batch_size
    return loss_sum / count, correct / count


@torch.inference_mode()
def evaluate(
    model: nn.Module,
    loader: DataLoader,
    loss_fn: nn.Module,
    device: torch.device,
) -> tuple[float, float]:
    model.eval()
    loss_sum = 0.0
    correct = 0
    count = 0
    for images, labels in loader:
        images = images.to(device, non_blocking=True)
        labels = labels.to(device, non_blocking=True)
        logits = model(images)
        loss = loss_fn(logits, labels)
        batch_size = labels.size(0)
        loss_sum += loss.item() * batch_size
        correct += (logits.argmax(dim=1) == labels).sum().item()
        count += batch_size
    return loss_sum / count, correct / count


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser(description="Train a small CNN on CIFAR-10")
    parser.add_argument("--data-dir", type=Path, default=Path("data"))
    parser.add_argument("--epochs", type=int, default=20)
    parser.add_argument("--batch-size", type=int, default=128)
    parser.add_argument("--learning-rate", type=float, default=3e-3)
    parser.add_argument("--weight-decay", type=float, default=5e-4)
    parser.add_argument("--workers", type=int, default=2)
    parser.add_argument("--validation-size", type=int, default=5_000)
    parser.add_argument("--seed", type=int, default=42)
    parser.add_argument("--output", type=Path, default=Path("small_cnn_best.pt"))
    return parser.parse_args()


def main() -> None:
    args = parse_args()
    if args.epochs < 1 or args.batch_size < 1 or args.workers < 0:
        raise ValueError("epochs and batch-size must be positive; workers cannot be negative")

    seed_everything(args.seed)
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    train_loader, validation_loader, test_loader = make_loaders(
        args.data_dir,
        args.batch_size,
        args.workers,
        args.seed,
        args.validation_size,
    )
    model = SmallCNN().to(device)
    loss_fn = nn.CrossEntropyLoss()
    optimizer = torch.optim.AdamW(
        model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay
    )
    scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=args.epochs)

    with torch.inference_mode():
        shape = model(torch.zeros(2, 3, 32, 32, device=device)).shape
    assert shape == (2, 10), f"Unexpected output shape: {shape}"

    best_accuracy = -1.0
    args.output.parent.mkdir(parents=True, exist_ok=True)
    print(f"device={device}; parameters={sum(p.numel() for p in model.parameters()):,}")
    for epoch in range(1, args.epochs + 1):
        train_loss, train_accuracy = train_one_epoch(model, train_loader, loss_fn, optimizer, device)
        validation_loss, validation_accuracy = evaluate(
            model, validation_loader, loss_fn, device
        )
        scheduler.step()
        print(
            f"epoch={epoch:02d} "
            f"train_loss={train_loss:.4f} train_acc={train_accuracy:.2%} "
            f"val_loss={validation_loss:.4f} val_acc={validation_accuracy:.2%}"
        )
        if validation_accuracy > best_accuracy:
            best_accuracy = validation_accuracy
            torch.save(
                {
                    "model_state": model.state_dict(),
                    "validation_accuracy": validation_accuracy,
                    "epoch": epoch,
                    "classes": test_loader.dataset.classes,
                },
                args.output,
            )

    checkpoint = torch.load(args.output, map_location=device)
    model.load_state_dict(checkpoint["model_state"])
    test_loss, test_accuracy = evaluate(model, test_loader, loss_fn, device)
    print(
        f"best_val_acc={best_accuracy:.2%}; "
        f"test_loss={test_loss:.4f}; test_acc={test_accuracy:.2%}; "
        f"checkpoint={args.output}"
    )


if __name__ == "__main__":
    main()