pytorch: 学习笔记11, pytorch权重衰减

1, pytorch权重衰减逐步实现

代码：

# weight_decay.py
import torch
import torch.nn as nn
import numpy as np
from matplotlib import pyplot as plt


def init_params():
    w = torch.randn((num_inputs, 1), requires_grad=True)
    b = torch.zeros(1, requires_grad=True)
    return [w, b]

def l2_penalty(w):
    return (w**2).sum() / 2

def squared_loss(y_hat, y):
    # 注意这里返回的是向量, 另外, pytorch里的MSELoss并没有除以 2
    return ((y_hat - y.view(y_hat.size())) ** 2) / 2

def linreg(X, w, b):
    return torch.mm(X, w) + b

def sgd(params, lr, batch_size):
    # 为了和原书保持一致，这里除以了batch_size，但是应该是不用除的，因为一般用PyTorch计算loss时就默认已经
    # 沿batch维求了平均了。
    for param in params:
        param.data -= lr * param.grad / batch_size

def semilogy_draw(x_vals, y_vals, x_label, y_label, x2_vals=None, y2_vals=None,
             legend=None, figsize=(3.5, 2.5)):
    # plt.figure(1, figsize)
    plt.xlabel(x_label)
    plt.ylabel(y_label)
    plt.semilogy(x_vals, y_vals)
    if x2_vals and y2_vals:
        plt.semilogy(x2_vals, y2_vals, linestyle=':')
        plt.legend(legend)
    plt.savefig('weight_decay.png')
    plt.show()

def fit_and_plot(lambd):
    w, b = init_params()
    train_ls, test_ls = [], []
    for _ in range(num_epochs):
        for X, y in train_iter:
            # 添加了L2范数惩罚项
            l = loss(net(X, w, b), y) + lambd * l2_penalty(w)  # lambd为0时，权重衰减项不起作用
            l = l.sum()

            if w.grad is not None:
                w.grad.data.zero_()
                b.grad.data.zero_()
            l.backward()
            sgd([w, b], lr, batch_size)
        train_ls.append(loss(net(train_features, w, b), train_labels).mean().item())
        test_ls.append(loss(net(test_features, w, b), test_labels).mean().item())
    semilogy_draw(range(1, num_epochs + 1), train_ls, 'epochs', 'loss',
                 range(1, num_epochs + 1), test_ls, ['train', 'test'])
    print('L2 norm of w:', w.norm().item())

def fit_and_plot_pytorch(wd):
    # 对权重参数衰减。权重名称一般是以weight结尾
    net = nn.Linear(num_inputs, 1)
    nn.init.normal_(net.weight, mean=0, std=1)
    nn.init.normal_(net.bias, mean=0, std=1)
    optimizer_w = torch.optim.SGD(params=[net.weight], lr=lr, weight_decay=wd) # 对权重参数衰减
    optimizer_b = torch.optim.SGD(params=[net.bias], lr=lr)  # 不对偏差参数衰减

    train_ls, test_ls = [], []
    for _ in range(num_epochs):
        for X, y in train_iter:
            l = loss(net(X), y).mean()
            optimizer_w.zero_grad()
            optimizer_b.zero_grad()

            l.backward()

            # 对两个optimizer实例分别调用step函数，从而分别更新权重和偏差
            optimizer_w.step()
            optimizer_b.step()
        train_ls.append(loss(net(train_features), train_labels).mean().item())
        test_ls.append(loss(net(test_features), test_labels).mean().item())
    semilogy_draw(range(1, num_epochs + 1), train_ls, 'epochs', 'loss',
                 range(1, num_epochs + 1), test_ls, ['train', 'test'])
    print('L2 norm of w:', net.weight.data.norm().item())


if __name__ == '__main__':
    n_train, n_test, num_inputs = 20, 100, 200
    true_w, true_b = torch.ones(num_inputs, 1) * 0.01, 0.05  # torch.Size([200, 1])

    features = torch.randn((n_train + n_test, num_inputs))  # torch.Size([120, 200])
    labels = torch.matmul(features, true_w) + true_b  # torch.Size([120, 1])
    labels += torch.tensor(np.random.normal(0, 0.01, size=labels.size()), dtype=torch.float)
    train_features, test_features = features[:n_train, :], features[n_train:, :]  # torch.Size([20, 200])  torch.Size([100, 200])
    train_labels, test_labels = labels[:n_train], labels[n_train:]  # torch.Size([20, 1])  torch.Size([100, 1])

    batch_size, num_epochs, lr = 2, 100, 0.003
    net, loss = linreg, squared_loss

    dataset = torch.utils.data.TensorDataset(train_features, train_labels)
    train_iter = torch.utils.data.DataLoader(dataset, batch_size, shuffle=True)

    fit_and_plot(lambd=3)  # lambd为0时，权重衰减项不起作用

结果：
a，当 fit_and_plot(lambd=0)参数lambd=0时，在训练集表现良好,损失逐渐下降，而在测试集loss降不下来，过拟合

b，当 fit_and_plot(lambd=3)参数lambd=3时，在训练集与测试集表现良好，一定程度上防止了过拟合。

2, pytorch权重衰减简单实现

代码：

import torch
import torch.nn as nn
import numpy as np
from matplotlib import pyplot as plt


def init_params():
    w = torch.randn((num_inputs, 1), requires_grad=True)
    b = torch.zeros(1, requires_grad=True)
    return [w, b]

def l2_penalty(w):
    return (w**2).sum() / 2

def squared_loss(y_hat, y):
    # 注意这里返回的是向量, 另外, pytorch里的MSELoss并没有除以 2
    return ((y_hat - y.view(y_hat.size())) ** 2) / 2

def sgd(params, lr, batch_size):
    # 为了和原书保持一致，这里除以了batch_size，但是应该是不用除的，因为一般用PyTorch计算loss时就默认已经
    # 沿batch维求了平均了。
    for param in params:
        param.data -= lr * param.grad / batch_size

def semilogy_draw(x_vals, y_vals, x_label, y_label, x2_vals=None, y2_vals=None,
             legend=None, figsize=(3.5, 2.5)):
    # plt.figure(1, figsize)
    plt.xlabel(x_label)
    plt.ylabel(y_label)
    plt.semilogy(x_vals, y_vals)
    if x2_vals and y2_vals:
        plt.semilogy(x2_vals, y2_vals, linestyle=':')
        plt.legend(legend)
    plt.savefig('weight_decay_simple.png')
    plt.show()

def fit_and_plot_pytorch(wd=0):
    # 定义网络并初始化网络参数w b
    net = nn.Linear(num_inputs, 1)
    nn.init.normal_(net.weight, mean=0, std=1)
    nn.init.normal_(net.bias, mean=0, std=1)
    optimizer_w = torch.optim.SGD(params=[net.weight], lr=lr, weight_decay=wd) # 在SGD中对权重参数衰减
    optimizer_b = torch.optim.SGD(params=[net.bias], lr=lr)  # 不对偏差参数衰减

    train_ls, test_ls = [], []
    for _ in range(num_epochs):
        for X, y in train_iter:
            l = loss(net(X), y).mean()
            optimizer_w.zero_grad()
            optimizer_b.zero_grad()

            l.backward()

            # 对两个optimizer实例分别调用step函数，从而分别更新权重和偏差
            optimizer_w.step()
            optimizer_b.step()
        train_ls.append(loss(net(train_features), train_labels).mean().item())
        test_ls.append(loss(net(test_features), test_labels).mean().item())
    semilogy_draw(range(1, num_epochs + 1), train_ls, 'epochs', 'loss',
                 range(1, num_epochs + 1), test_ls, ['train', 'test'])
    print('L2 norm of w:', net.weight.data.norm().item())


if __name__ == '__main__':
    n_train, n_test, num_inputs = 20, 100, 200
    true_w, true_b = torch.ones(num_inputs, 1) * 0.01, 0.05  # torch.Size([200, 1])

    features = torch.randn((n_train + n_test, num_inputs))  # torch.Size([120, 200])
    labels = torch.matmul(features, true_w) + true_b  # torch.Size([120, 1])
    labels += torch.tensor(np.random.normal(0, 0.01, size=labels.size()), dtype=torch.float)
    train_features, test_features = features[:n_train, :], features[n_train:, :]  # torch.Size([20, 200])  torch.Size([100, 200])
    train_labels, test_labels = labels[:n_train], labels[n_train:]  # torch.Size([20, 1])  torch.Size([100, 1])

    batch_size, num_epochs, lr = 2, 100, 0.003
    loss = squared_loss

    dataset = torch.utils.data.TensorDataset(train_features, train_labels)
    train_iter = torch.utils.data.DataLoader(dataset, batch_size, shuffle=True)

    fit_and_plot_pytorch(wd=0)  # lambd为0时，权重衰减项不起作用

结果：
a，当 fit_and_plot_pytorch参数wd=0时，在训练集表现良好,损失逐渐下降，而在测试集loss降不下来，过拟合。

b，当 fit_and_plot_pytorch参数wd=3时，在训练集与测试集表现良好，一定程度上防止了过拟合。

代码并非自己实现，参考并摘自：
https://tangshusen.me/Dive-into-DL-PyTorch/#/chapter03_DL-basics/3.12_weight-decay