PyTorch入門教程：手把手教你搭建第一個神經網絡模型¶

# CPU版本
pip install torch torchvision

# GPU版本（需預先安裝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

# 數據變換：將圖像轉爲張量並歸一化
transform = transforms.Compose([
    transforms.ToTensor(),  # 轉換爲張量，像素值從0-255變爲0-1
    transforms.Normalize((0.1307,), (0.3081,))  # 標準化（減去均值，除以標準差）
])

# 加載訓練集和測試集
train_dataset = datasets.MNIST(
    root='./data', train=True, download=True, transform=transform
)
test_dataset = datasets.MNIST(
    root='./data', train=False, download=True, transform=transform
)

# 創建數據加載器（批量處理數據，打亂順序）
batch_size = 64  # 每次訓練的樣本數
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__()
        # 定義全連接層：輸入784，輸出128
        self.fc1 = nn.Linear(28*28, 128)
        # 定義全連接層：輸入128，輸出10
        self.fc2 = nn.Linear(128, 10)

    def forward(self, x):
        # 前向傳播：先展平圖像，再通過全連接層和激活函數
        x = x.view(-1, 28*28)  # 展平圖像，-1表示自動計算batch_size
        x = torch.relu(self.fc1(x))  # ReLU激活函數
        x = self.fc2(x)  # 輸出層，無激活（CrossEntropyLoss會自動處理）
        return x

# 實例化模型
model = SimpleNN()

criterion = nn.CrossEntropyLoss()  # 交叉熵損失
optimizer = optim.SGD(model.parameters(), lr=0.01)  # SGD優化器

epochs = 5  # 訓練輪數

for epoch in range(epochs):
    model.train()  # 模型設爲訓練模式（啓用Dropout/BatchNorm等）
    running_loss = 0.0

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

        # 前向傳播：計算模型輸出
        outputs = model(data)

        # 計算損失
        loss = criterion(outputs, target)

        # 反向傳播：計算梯度
        loss.backward()

        # 更新參數
        optimizer.step()

        # 累計損失
        running_loss += loss.item()

        # 每100個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()  # 模型設爲評估模式（禁用Dropout等）
correct = 0
total = 0

with torch.no_grad():  # 測試時不計算梯度，節省內存
    for data, target in test_loader:
        outputs = model(data)
        _, predicted = torch.max(outputs.data, 1)  # 獲取預測類別
        total += target.size(0)
        correct += (predicted == target).sum().item()

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