L1正则化代码pytorch

在PyTorch中，L1正则化（也称为Lasso回归）通常是在模型的损失函数中添加一个权重衰减项，用来防止过拟合。这种正则化方法鼓励模型学习到稀疏的权重，即一些权重接近于0。以下是使用PyTorch实现L1正则化的简单示例：

import torch
import torch.nn as nn

class MyModel(nn.Module):
    def __init__(self, input_size, output_size):
        super(MyModel, self).__init__()
        # 在模型定义中添加参数
        self.linear = nn.Linear(input_size, output_size)

    def forward(self, x):
        return self.linear(x)

    # 添加L1正则化
    def l1_regularizer(self, weight_decay=0.01):
        # weight_decay是L1惩罚的强度
        return torch.sum(torch.abs(self.linear.weight))

# 实例化模型并训练
model = MyModel(input_size, output_size)
optimizer = torch.optim.SGD(model.parameters(), lr=0.01, weight_decay=0.01)  # 注意weight_decay参数

loss_fn = nn.CrossEntropyLoss()
for epoch in range(num_epochs):
    # 训练步骤
    optimizer.zero_grad()
    outputs = model(x_train)
    loss = loss_fn(outputs, y_train) + weight_decay * model.l1_regularizer()
    loss.backward()
    optimizer.step()