@疏锦行

知识点回顾：

1.  过拟合的判断：测试集和训练集同步打印指标

2.  模型的保存和加载

a.  仅保存权重

b.  保存权重和模型

c.  保存全部信息checkpoint，还包含训练状态

3.  早停策略

作业：对信贷数据集训练后保存权重，加载权重后继续训练50轮，并采取早停策略

# 保存模型权重
torch.save(model.state_dict(), 'credit_model_weights.pth')# 加载模型权重
model.load_state_dict(torch.load('credit_model_weights.pth'))# 设置继续训练的轮数
additional_epochs = 50for epoch in range(additional_epochs):# 前向传播outputs = model(X_train_tensor)loss = criterion(outputs, y_train_tensor)# 反向传播和优化optimizer.zero_grad()loss.backward()optimizer.step()if (epoch + 1) % 10 == 0:print(f'Epoch [{epoch+1}/{additional_epochs}], Loss: {loss.item():.4f}')# 保存继续训练后的模型权重
torch.save(model.state_dict(), 'credit_model_weights_continued.pth')
# 早停策略参数
patience = 10  # 容忍验证集损失不下降的最大轮数
best_val_loss = float('inf')
counter = 0for epoch in range(num_epochs):# 训练代码model.train()outputs = model(X_train_tensor)train_loss = criterion(outputs, y_train_tensor)optimizer.zero_grad()train_loss.backward()optimizer.step()# 验证代码model.eval()with torch.no_grad():val_outputs = model(X_val_tensor)val_loss = criterion(val_outputs, y_val_tensor)print(f'Epoch [{epoch+1}/{num_epochs}], Train Loss: {train_loss.item():.4f}, Val Loss: {val_loss.item():.4f}')# 早停策略逻辑if val_loss < best_val_loss:best_val_loss = val_losscounter = 0# 保存最佳模型权重torch.save(model.state_dict(), 'best_credit_model_weights.pth')else:counter += 1if counter >= patience:print('Early stopping!')break