PyTorch入门教程：手把手教你搭建第一个神经网络模型¶

# CPU版本
pip install torch torchvision

# GPU版本（需预先安装CUDA）
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118/torch_stable.html

import torch
import torch.nn as nn
import torch.optim as optim
from torchvision import datasets, transforms
from torch.utils.data import DataLoader

# 数据变换：将图像转为张量并归一化
transform = transforms.Compose([
    transforms.ToTensor(),  # 转换为张量，像素值从0-255变为0-1
    transforms.Normalize((0.1307,), (0.3081,))  # 标准化（减去均值，除以标准差）
])

# 加载训练集和测试集
train_dataset = datasets.MNIST(
    root='./data', train=True, download=True, transform=transform
)
test_dataset = datasets.MNIST(
    root='./data', train=False, download=True, transform=transform
)

# 创建数据加载器（批量处理数据，打乱顺序）
batch_size = 64  # 每次训练的样本数
train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
test_loader = DataLoader(test_dataset, batch_size=batch_size, shuffle=False)

class SimpleNN(nn.Module):
    def __init__(self):
        super(SimpleNN, self).__init__()
        # 定义全连接层：输入784，输出128
        self.fc1 = nn.Linear(28*28, 128)
        # 定义全连接层：输入128，输出10
        self.fc2 = nn.Linear(128, 10)

    def forward(self, x):
        # 前向传播：先展平图像，再通过全连接层和激活函数
        x = x.view(-1, 28*28)  # 展平图像，-1表示自动计算batch_size
        x = torch.relu(self.fc1(x))  # ReLU激活函数
        x = self.fc2(x)  # 输出层，无激活（CrossEntropyLoss会自动处理）
        return x

# 实例化模型
model = SimpleNN()

criterion = nn.CrossEntropyLoss()  # 交叉熵损失
optimizer = optim.SGD(model.parameters(), lr=0.01)  # SGD优化器

epochs = 5  # 训练轮数

for epoch in range(epochs):
    model.train()  # 模型设为训练模式（启用Dropout/BatchNorm等）
    running_loss = 0.0

    for batch_idx, (data, target) in enumerate(train_loader):
        # 清零梯度
        optimizer.zero_grad()

        # 前向传播：计算模型输出
        outputs = model(data)

        # 计算损失
        loss = criterion(outputs, target)

        # 反向传播：计算梯度
        loss.backward()

        # 更新参数
        optimizer.step()

        # 累计损失
        running_loss += loss.item()

        # 每100个batch打印一次信息
        if batch_idx % 100 == 99:
            print(f'Epoch {epoch+1}, Batch {batch_idx+1}, Loss: {running_loss/100:.4f}')
            running_loss = 0.0

model.eval()  # 模型设为评估模式（禁用Dropout等）
correct = 0
total = 0

with torch.no_grad():  # 测试时不计算梯度，节省内存
    for data, target in test_loader:
        outputs = model(data)
        _, predicted = torch.max(outputs.data, 1)  # 获取预测类别
        total += target.size(0)
        correct += (predicted == target).sum().item()

print(f'Test Accuracy: {100 * correct / total:.2f}%')