)
def forward(self, img):
feature = self.conv(img)
output = self.fc(feature.view(img.shape[0], -1))
return output
net = LeNet()
net,
'''测试'''
with st.echo():
batch_size = st.slider(label='批量', min_value=256, max_value=2560, value=256, step=256)
lr = st.slider(label='学习率', min_value=0.001, max_value=1.0, value=0.001, step=0.001)
num_epochs = st.slider(label='迭代周期', min_value=5, max_value=100, value=5, step=5)
train_iter, test_iter = load_data_fashion_mnist(batch_size)
# 因为卷积神经网络计算比多层感知机要复杂，建议使用GPU来加速计算。因此，我们对3.6节（softmax回归的从零开始实现）中描述的evaluate_accuracy函数略作修改，使其支持GPU计算。
def train_ch5(net, train_iter, test_iter, batch_size, optimizer, device, num_epochs):
net = net.to(device)
device,
loss = torch.nn.CrossEntropyLoss()
batch_count = 0
for epoch in range(num_epochs):
train_l_sum, train_acc_sum, n, start = 0.0, 0.0, 0, time.time()
for X, y in train_iter:
X = X.to(device)
y = y.to(device)
y_hat = net(X)
l = loss(y_hat, y)
optimizer.zero_grad()
net,
'''
### 读取数据
虽然论文中AlexNet使用ImageNet数据集，但因为ImageNet数据集训练时间较长，我们仍用前面的Fashion-MNIST数据集来演示AlexNet。读取数据的时候我们额外做了一步将图像高和宽扩大到AlexNet使用的图像高和宽224。这个可以通过torchvision.transforms.Resize实例来实现。也就是说，我们在ToTensor实例前使用Resize实例，然后使用Compose实例来将这两个变换串联以方便调用。
'''
with st.echo():
batch_size = st.slider(label='批量',
min_value=1,
max_value=2560,
value=128,
step=128)
lr = st.slider(label='学习率',
min_value=0.001,
max_value=1.0,
value=0.001,
step=0.001)
num_epochs = st.slider(label='迭代周期',
min_value=1,
max_value=100,
value=5,
step=5)
# 如出现“out of memory”的报错信息，可减小batch_size或resize
train_iter, test_iter = load_data_fashion_mnist(batch_size,
resize=224,
num_workers=0)
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
train_ch5(net, train_iter, test_iter, batch_size, optimizer, device,
num_epochs)