深度学习入门 FashionMNIST数据集训练和测试（30层神经网路）_fmnist是几通道的数据

使用pytorch框架。模型包含13层卷积层、2层池化层、15层全连接层。为什么叠这么多层？就是玩。

FashionMNIST数据集包含训练集6w张图片，测试集1w张图片，每张图片是单通道、大小28×28。

import argparse
import torch
import torch.nn as nn  # 指定torch.nn别名nn
import torch.optim as optim
import torchvision  # 一些加载数据的函数及常用的数据集接口
import torchvision.transforms as transforms


class CNN(nn.Module):
    def __init__(self):
        super(CNN, self).__init__()
        self.conv = nn.Sequential(
            nn.Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=32),
            nn.ReLU(inplace=True),
            nn.Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=64),
            nn.ReLU(inplace=True),
            nn.Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=64),
            nn.ReLU(inplace=True),
            nn.Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=64),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=2, stride=2),  # out 14

            nn.Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
            nn.BatchNorm2d(num_features=128),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=2, stride=2),  # out 7
        )
        self.linear = nn.Sequential(
            nn.Linear(128 * 7 * 7, 128),  # 全连接层
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 128),
            nn.BatchNorm1d(num_features=128),
            nn.ReLU(inplace=True),
            nn.Linear(128, 10),
        )

    def forward(self, inputs):
        out = self.conv(inputs)
        out = out.view(out.size(0), -1)
        logits = self.linear(out)
        return logits


def train(model, device, train_loader, criterion, optimizer):
    model.train()
    total_loss, total_num = 0, 0
    for data, target in train_loader:
        data, target = data.to(device), target.to(device)
        optimizer.zero_grad()
        logits = model(data)
        loss = criterion(logits, target)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        total_loss += loss.item()
        total_num += 1

    return total_loss / total_num


def test(model, device, test_loader):
    model.eval()
    t_correct, t_sum = 0, 0
    with torch.no_grad():
        for data, target in test_loader:
            data, target = data.to(device), target.to(device)
            logits = model(data)
            pred = logits.argmax(dim=1)
            t_correct += torch.eq(pred, target).float().sum().item()
            t_sum += data.size(0)
        acc = t_correct / t_sum
        return acc


def main():
    fashionmnist_train = torchvision.datasets.FashionMNIST(
        root='F:/dataset/minist', train=True, download=True,
        transform=transforms.ToTensor(),
    )
    fashionmnist_test = torchvision.datasets.FashionMNIST(
        root='F:/dataset/minist', train=False, download=True,
        transform=transforms.ToTensor(),
    )
    print('训练集和测试集大小：', len(fashionmnist_train), len(fashionmnist_test))

    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    train_iter = torch.utils.data.DataLoader(fashionmnist_train, batch_size=args.batch_size, shuffle=True)
    test_iter = torch.utils.data.DataLoader(fashionmnist_test, batch_size=args.test_batch_size)

    model = CNN().to(device)  # 类转换到cuda上
    criterion = nn.CrossEntropyLoss().to(device)

    optimizer = optim.Adam(model.parameters(), lr=args.lr, weight_decay=0.001)  # model.parameters() 网络里面的参数
    lr_sch = torch.optim.lr_scheduler.ExponentialLR(optimizer, 0.9)

    for epoch in range(args.epochs):
        train_loss = train(model, device, train_iter, criterion, optimizer)
        acc = test(model, device, test_iter)
        print(f'epoch {epoch:2d}: loss = {train_loss:.6f}; acc={acc:.4f}; lr：{lr_sch.get_last_lr()[0]:.8f}')
        lr_sch.step()  # 执行一次学习率衰减   lr = lr * 衰减率

    if args.save_model:
        torch.save(model.state_dict(), "cnn.pt")


if __name__ == '__main__':
    parser = argparse.ArgumentParser(description='test')
    parser.add_argument('--batch-size', type=int, default=64, metavar='N',
                        help='input batch size for training (default: 64)')
    parser.add_argument('--test-batch-size', type=int, default=256, metavar='N',
                        help='input batch size for testing (default: 1000)')
    parser.add_argument('--epochs', type=int, default=20, metavar='N',
                        help='number of epochs to train (default: 10)')
    parser.add_argument('--lr', type=float, default=0.001, metavar='LR',
                        help='learning rate (default: 0.01)')
    parser.add_argument('--save-model', action='store_true', default=False,
                        help='For Saving the current Model')

    # 解析参数
    args = parser.parse_args()
    main()

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

运行结果：

训练集和测试集大小： 60000 10000
epoch  0: loss = 0.680793; acc=0.8512; lr：0.00100000
epoch  1: loss = 0.413065; acc=0.8831; lr：0.00090000
epoch  2: loss = 0.375292; acc=0.8931; lr：0.00081000
epoch  3: loss = 0.352440; acc=0.8852; lr：0.00072900
epoch  4: loss = 0.342234; acc=0.8936; lr：0.00065610
epoch  5: loss = 0.316583; acc=0.8845; lr：0.00059049
epoch  6: loss = 0.304564; acc=0.8570; lr：0.00053144
epoch  7: loss = 0.282753; acc=0.9003; lr：0.00047830
epoch  8: loss = 0.259815; acc=0.9098; lr：0.00043047
epoch  9: loss = 0.237412; acc=0.9196; lr：0.00038742
epoch 10: loss = 0.218256; acc=0.9124; lr：0.00034868
epoch 11: loss = 0.205545; acc=0.9228; lr：0.00031381
epoch 12: loss = 0.187289; acc=0.9243; lr：0.00028243
epoch 13: loss = 0.173118; acc=0.9268; lr：0.00025419
epoch 14: loss = 0.156642; acc=0.9301; lr：0.00022877
epoch 15: loss = 0.140609; acc=0.9210; lr：0.00020589
epoch 16: loss = 0.128453; acc=0.9296; lr：0.00018530
epoch 17: loss = 0.116559; acc=0.9321; lr：0.00016677
epoch 18: loss = 0.104093; acc=0.9324; lr：0.00015009
epoch 19: loss = 0.089764; acc=0.9308; lr：0.00013509
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21