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

模型

net = nn.Sequential(nn.Flatten(),
                    nn.Linear(784, 256),
                    nn.ReLU(),  # 与 3.7 节相比多了一层
                    nn.Linear(256, 10))

def init_weights(m):
    if type(m) == nn.Linear:  # 使用正态分布中的随机值初始化权重
        nn.init.normal_(m.weight, std=0.01)

net.apply(init_weights)

Sequential(
  (0): Flatten(start_dim=1, end_dim=-1)
  (1): Linear(in_features=784, out_features=256, bias=True)
  (2): ReLU()
  (3): Linear(in_features=256, out_features=10, bias=True)
)

batch_size, lr, num_epochs = 256, 0.1, 10
loss = nn.CrossEntropyLoss(reduction='none')
trainer = torch.optim.SGD(net.parameters(), lr=lr)

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

练习

（1）尝试添加不同数量的隐藏层（也可以修改学习率），怎样设置效果最好？

net2 = nn.Sequential(nn.Flatten(),
                    nn.Linear(784, 256),
                    nn.ReLU(),
                    nn.Linear(256, 128),
                    nn.ReLU(),
                    nn.Linear(128, 10))

def init_weights(m):
    if type(m) == nn.Linear:  # 使用正态分布中的随机值初始化权重
        nn.init.normal_(m.weight, std=0.01)

net2.apply(init_weights)

batch_size2, lr2, num_epochs2 = 256, 0.3, 10
loss2 = nn.CrossEntropyLoss(reduction='none')
trainer2 = torch.optim.SGD(net2.parameters(), lr=lr2)

train_iter2, test_iter2 = d2l.load_data_fashion_mnist(batch_size2)
d2l.train_ch3(net2, train_iter2, test_iter2, loss2, num_epochs2, trainer2)

（2）尝试不同的激活函数，哪个激活函数效果最好？

net3 = nn.Sequential(nn.Flatten(),
                    nn.Linear(784, 256),
                    nn.Sigmoid(),
                    nn.Linear(256, 10))

net4 = nn.Sequential(nn.Flatten(),
                    nn.Linear(784, 256),
                    nn.Tanh(),
                    nn.Linear(256, 10))

def init_weights(m):
    if type(m) == nn.Linear:
        nn.init.normal_(m.weight, std=0.01)

net3.apply(init_weights)
net4.apply(init_weights)


train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)

batch_size, lr, num_epochs = 256, 0.1, 10
loss = nn.CrossEntropyLoss(reduction='none')
trainer = torch.optim.SGD(net3.parameters(), lr=lr)
d2l.train_ch3(net3, train_iter, test_iter, loss, num_epochs, trainer)

---------------------------------------------------------------------------

AssertionError                            Traceback (most recent call last)

Cell In[5], line 4
      2 loss = nn.CrossEntropyLoss(reduction='none')
      3 trainer = torch.optim.SGD(net3.parameters(), lr=lr)
----> 4 d2l.train_ch3(net3, train_iter, test_iter, loss, num_epochs, trainer)


File c:\Software\Miniconda3\envs\d2l\lib\site-packages\d2l\torch.py:340, in train_ch3(net, train_iter, test_iter, loss, num_epochs, updater)
    338     animator.add(epoch + 1, train_metrics + (test_acc,))
    339 train_loss, train_acc = train_metrics
--> 340 assert train_loss < 0.5, train_loss
    341 assert train_acc <= 1 and train_acc > 0.7, train_acc
    342 assert test_acc <= 1 and test_acc > 0.7, test_acc


AssertionError: 0.5017133202234904

batch_size, lr, num_epochs = 256, 0.1, 10
loss = nn.CrossEntropyLoss(reduction='none')
trainer = torch.optim.SGD(net4.parameters(), lr=lr)
d2l.train_ch3(net4, train_iter, test_iter, loss, num_epochs, trainer)

还是 ReLU 比较奈斯。

（3）尝试不同的方案来初始化权重，什么方案效果最好。

累了，不想试试了。略......