LeNet实现

导入包

import torch
from torch import nn
from d2l import torch as d2l #李沐-动手深度学习课程中的一些程序打包的库
from matplotlib import pyplot as plt

保持形状一致

class Reshape(torch.nn.Module):
    def forward(self, x):
        return x.view(-1, 1, 28, 28)  # 把x变成批量数不变，通道数变成1，大小变成28*28

LeNet模型

net = torch.nn.Sequential(
    Reshape(),
    nn.Conv2d(1, 6, kernel_size=5, padding=2),  # 通道数为输入为1，输出为6，卷积核大小5×5，上下左右填充2行两列，变成32×32 、、卷积
    nn.Sigmoid(),  # 、、 激活
    nn.AvgPool2d(kernel_size=2, stride=2),  # 、、平均池化
    nn.Conv2d(6, 16, kernel_size=5),  # 通道数为输入为6，输出为16
    nn.Sigmoid(),
    nn.AvgPool2d(kernel_size=2, stride=2),
    nn.Flatten(),
    nn.Linear(16 * 5 * 5, 120),
    nn.Sigmoid(),
    nn.Linear(120, 84),
    nn.Sigmoid(),
    nn.Linear(84, 10))

查看网络结构

X = torch.rand(size=(1,1,28,28),dtype = torch.float32)
for layer in net:
    X = layer(X)
    print(layer.__class__.__name__,'output shape: \t',X.shape)

批量加载数据

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

GPU上计算Acc

def evaluate_accuracy_gpu(net, data_iter, device=None): #@save
    """使用GPU计算模型在数据集上的精度"""
    if isinstance(net, nn.Module):
        net.eval()  # 设置为评估模式
        if not device:
            device = next(iter(net.parameters())).device # device没给定的话，从网络的第一个参数里面拿出来，看他的device在哪里
    # 正确预测的数量，总预测的数量
    metric = d2l.Accumulator(2)
    with torch.no_grad():
        for X, y in data_iter:
            if isinstance(X, list):
                # BERT微调所需的（之后将介绍）
                X = [x.to(device) for x in X]  #对每一个x，先移动到device上去，是列表的话，每一个移动一下
            else:
                X = X.to(device) # 张量直接移动
            y = y.to(device)
            metric.add(d2l.accuracy(net(X), y), y.numel())
    return metric[0] / metric[1]

定义训练函数

def train_ch6(net, train_iter, test_iter, num_epochs, lr, device):
    """用GPU训练模型(在第六章定义)"""
    def init_weights(m): #初始化权重
        if type(m) == nn.Linear or type(m) == nn.Conv2d:
            nn.init.xavier_uniform_(m.weight) #如果是全连接层或者是卷积层，调用定义好的初始化操作，xavier_uniform_通过网络层时，输入和输出的方差相同，防止初始时模型爆炸或者为0
    net.apply(init_weights) #网络应用初始化
    print('training on', device) #查看是在cpu还是GPU上训练
    net.to(device) #模型送入GPU
    optimizer = torch.optim.SGD(net.parameters(), lr=lr) #SGD优化器
    loss = nn.CrossEntropyLoss() #交叉熵损失函数
    animator = d2l.Animator(xlabel='epoch', xlim=[1, num_epochs],
                            legend=['train loss', 'train acc', 'test acc']) #画出误差和Acc曲线
    timer, num_batches = d2l.Timer(), len(train_iter)
    for epoch in range(num_epochs):
        # 训练损失之和，训练准确率之和，样本数
        metric = d2l.Accumulator(3)
        net.train()
        for i, (X, y) in enumerate(train_iter): #每次数据迭代中拿一个batch出来
            timer.start()
            optimizer.zero_grad() #梯度置0
            X, y = X.to(device), y.to(device) #输入输出移到GPU
            y_hat = net(X) #前向操作
            l = loss(y_hat, y) #计算损失
            l.backward() # 计算梯度
            optimizer.step() #迭代
            with torch.no_grad():
                metric.add(l * X.shape[0], d2l.accuracy(y_hat, y), X.shape[0])
            timer.stop()
            train_l = metric[0] / metric[2]
            train_acc = metric[1] / metric[2]
            if (i + 1) % (num_batches // 5) == 0 or i == num_batches - 1:
                animator.add(epoch + (i + 1) / num_batches,
                             (train_l, train_acc, None))
        test_acc = evaluate_accuracy_gpu(net, test_iter)
        animator.add(epoch + 1, (None, None, test_acc))
    print(f'loss {train_l:.3f}, train acc {train_acc:.3f}, 'f'test acc {test_acc:.3f}')
    print(f'{metric[2] * num_epochs / timer.sum():.1f} examples/sec ' f'on {str(device)}')

初始化参数，训练

lr, num_epochs = 0.9, 30
train_ch6(net, train_iter, test_iter, num_epochs, lr, d2l.try_gpu())
plt.savefig('lenet.jpg',dpi=100)
plt.show()