### PyTorch 训练过程示例代码模板 #### 1. 导入必要的库 为了确保训练环境的一致性和可重复性，首先需要导入所需的Python库，并固定随机种子。 ```python import random import numpy as np import torch from torch import nn, optim from torch.utils.data import DataLoader, Dataset from torchvision import datasets, transforms import matplotlib.pyplot as plt def set_random_seeds(seed_value=3407): """设置全局随机种子""" random.seed(seed_value) np.random.seed(seed_value) torch.manual_seed(seed_value) if torch.cuda.is_available(): torch.cuda.manual_seed_all(seed_value) set_random_seeds() # 调用函数设定随机数种子[^2] ``` #### 2. 数据预处理与加载 创建自定义的数据集类继承`Dataset`接口，并实现`__len__()`和`__getitem__()`方法来读取样本；通过`DataLoader`构建批量迭代器用于后续训练。 ```python class CustomImageDataset(Dataset): def __init__(self, transform=None): self.transform = transform def __len__(self): pass # 返回数据集中元素的数量 def __getitem__(self, idx): pass # 获取指定索引位置处的数据项及其标签 transform = transforms.Compose([ transforms.ToTensor(), ]) train_dataset = CustomImageDataset(transform=transform) test_dataset = CustomImageDataset(transform=transform) batch_size = 64 train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True) test_loader = DataLoader(test_dataset, batch_size=batch_size * 2) # 测试时可以适当增大批次大小以加快速度[^3] ``` #### 3. 构建神经网络模型 设计一个简单的卷积神经网络作为分类器，在这里仅提供了一个基础结构供参考。实际应用中可根据具体任务需求调整架构细节。 ```python class SimpleCNN(nn.Module): def __init__(self): super(SimpleCNN, self).__init__() self.conv1 = nn.Conv2d(in_channels=1, out_channels=8, kernel_size=(5, 5)) self.pool = nn.MaxPool2d(kernel_size=2, stride=2) self.fc1 = nn.Linear(8*12*12, 10) def forward(self, x): x = self.pool(F.relu(self.conv1(x))) x = x.view(-1, 8*12*12) x = self.fc1(x) return x model = SimpleCNN() if torch.cuda.device_count() > 1: model = nn.DataParallel(model) # 支持多GPU加速计算[^1] device = 'cuda' if torch.cuda.is_available() else 'cpu' model.to(device=device) ``` #### 4. 设置损失函数、优化算法及其他辅助组件 选择合适的损失函数（如交叉熵），初始化Adam优化器，并准备其他可能需要用到的对象或变量。 ```python criterion = nn.CrossEntropyLoss().to(device=device) optimizer = optim.Adam(model.parameters(), lr=0.001) scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=7, gamma=0.1) # 学习率衰减策略 early_stopping_patience = 5 best_accuracy = 0. patience_counter = 0 ``` #### 5. 执行训练循环 编写主训练逻辑，包括前向传播、反向传播更新权重以及周期性的评估性能指标等操作。 ```python num_epochs = 25 for epoch in range(num_epochs): running_loss = 0.0 correct_predictions = 0 total_samples = 0 for inputs, labels in train_loader: optimizer.zero_grad() outputs = model(inputs.to(device)) loss = criterion(outputs, labels.to(device)) _, predicted = torch.max(outputs, dim=-1) correct_predictions += (predicted.cpu() == labels).sum().item() total_samples += len(labels) loss.backward() optimizer.step() running_loss += loss.item() scheduler.step() accuracy = float(correct_predictions)/total_samples print(f"[Epoch {epoch+1}/{num_epochs}] Loss: {running_loss/len(train_loader)}, Accuracy: {accuracy:.4f}") # Early stopping mechanism based on validation performance... if best_accuracy < accuracy: patience_counter = 0 best_accuracy = accuracy elif patience_counter >= early_stopping_patience: break else: patience_counter += 1 ``` #### 6. 结果可视化 最后可以通过绘制图表展示训练过程中的一些重要统计量变化趋势，帮助分析模型表现情况。 ```python plt.plot(range(len(losses)), losses) plt.title('Training Loss') plt.xlabel('Iteration') plt.ylabel('Loss Value') plt.show() ```