1
import torch
2
from torch import nn
3
from d2l import torch as d2l
4

5
batch_size = 256
6
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)

1
num_inputs, num_outputs, num_hiddens = 784, 10, 256
2

3
W1 = nn.Parameter(torch.randn(
4
    num_inputs, num_hiddens, requires_grad=True) * 0.01)
5
b1 = nn.Parameter(torch.zeros(num_hiddens, requires_grad=True))
6
W2 = nn.Parameter(torch.randn(
7
    num_hiddens, num_outputs, requires_grad=True) * 0.01)
8
b2 = nn.Parameter(torch.zeros(num_outputs, requires_grad=True))
9

10
params = [W1, b1, W2, b2]

1
def relu(X):
2
    a = torch.zeros_like(X)
3
    return torch.max(X, a)

1
def net(X):
2
    X = X.reshape(-1, num_inputs)    # X形状为(256, 784)
3
    H = relu(X @ W1 + b1)    # W1形状为(784, 256)，H形状为(256, 256)
4
    return (H @ W2 + b2)    # W2形状为(256, 10)
5
    # “@”代表矩阵乘法

1
num_epochs, lr = 10, 0.1
2
updater = torch.optim.SGD(params, lr=lr)
3
d2l.train_ch3(net, train_iter, test_iter, loss, num_epochs, updater)
4
#

1
import torch
2
from torch import nn
3
from d2l import torch as d2l
4

5
net = nn.Sequential(nn.Flatten(),
6
                   nn.Linear(784, 256),
7
                   nn.ReLU(),
8
                   nn.Linear(256, 10))
9

10
def init_weights(m):
11
    if type(m) == nn.Linear:
12
        nn.init.normal_(m.weight, std=0.01);
13

14
net.apply(init_weights)
15

16
batch_size, lr, num_epochs = 256, 0.1, 10
17
loss = nn.CrossEntropyLoss()
18
trainer = torch.optim.SGD(net.parameters(), lr=lr)
19

20
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
21
d2l.train_ch3(net, train_iter, test_iter, loss, num_epochs, trainer)