一起深度学习（AlexNet网络）

代码实现：

import torch
from torch import nn
from d2l import torch as d2l

net = nn.Sequential(
    # 采用了11*11的卷积核来捕捉对象，因为原始输入数据比较大
    #步幅为4 ，可减少输出的高度核宽度。
    #输出通道为96，远大于Lenet
    #卷积之后大小的计算：H_2 = (H_1 + 2 * padding - kernel_size) / stride + 1,向下取整
    # 因为这里我们用的数据是fahsion_mnist ，输入通道就为1
    # [1,224,224] => [1,96，54,54]
    nn.Conv2d(in_channels=1,out_channels=96,kernel_size=11,stride=4,padding=1),nn.ReLU(),
    #池化后的大小计算公式：H_2 = (H1 + 2 * padding - kernel_size) / stride + 1
    # [1,96，54,54] => [1,96,26，26]
    nn.MaxPool2d(kernel_size=3,stride=2),
    #减小卷积窗口大小，使用padding = 2 来保证输入与输出的宽高一致，且增大输出通道。
    # [1,96,26,26] => [1,256,26,26]  H2=(26 + 2*2 -5)/1 +1
    nn.Conv2d(in_channels=96,out_channels=256,kernel_size=5,padding=2),nn.ReLU(),
    #[1,256,26,26] => [1,256,12,12]
    nn.MaxPool2d(kernel_size=3,stride=2),
    #三个卷积层
    # [1,256,12,12] => [1,384,12,12]
    nn.Conv2d(256,384, kernel_size=3, padding=1),nn.ReLU(),
    # [1,384,12,12]  =>  [1,384,12,12]
    nn.Conv2d(384,384, kernel_size=3, padding=1),nn.ReLU(),
    # [1,384,12,12] => [1,,256,12,12]
    nn.Conv2d(384,256, kernel_size=3, padding=1),nn.ReLU(),
    #[1,,256,12,12] => [1,256,5,5] 因为 H2 = (12 -3 )/2 +1 = 5
    nn.MaxPool2d(kernel_size=3, stride=2),
    #输入数据打平，传送给全连接层.
    nn.Flatten(),  # 256 * 5 * 5 = 6400
    # 由于输入的数据过大，远远大于leNet网络，为了避免过拟合，可采用 dropout
    nn.Linear(6400,4096),nn.ReLU(),
    nn.Dropout(p=0.5),
    nn.Linear(4096,4096),nn.ReLU(),
    nn.Dropout(p=0.5),
    nn.Linear(4096,10)
)

#加载数据集
batch_size = 128
train_iter,test_iter = d2l.load_data_fashion_mnist(batch_size,resize=224)

#训练模型：

#开始训练
lr,num_epochs = 0.01,10
d2l.train_ch6(net,train_iter,test_iter,num_epochs,lr,d2l.try_gpu())

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