线性分类代码示例

以下是一个基于Python语言的线性分类代码示例，主要用到pytorch,numpy,matplotlib，

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

# 生成随机数据
x_ = np.linspace(1, 0.1, 100).reshape(100, 1)
y_ = 3 * x_ + np.random.uniform(0, 1, (100, 1))

# 转成tensor
x = torch.from_numpy(x_).float()
y = torch.from_numpy(y_).float()

linear = torch.nn.Linear(1, 1)  # 线性分类
optimizer = torch.optim.SGD(linear.parameters(), lr=0.1)  # 优化器
loss_func = torch.nn.MSELoss()  # 损失函数

EPOCH_NUM = 200
for epoch in range(EPOCH_NUM):
    plt.scatter(x, y, s=5)
    y_pred = linear(x)  # 线性分类
    loss = loss_func(y_pred, y)  # 计算损失
    print(loss)
    optimizer.zero_grad()  # 梯度清零
    loss.backward()  # 损失回传
    optimizer.step()  # 进行一次优化，更新参数
    # 线性分类表达式，f(x,W)=Wx+b
    f = linear.state_dict()['weight'].numpy()[0][0] * x_ + linear.state_dict()['bias'].numpy()[0]
    plt.plot(x_, f, 'r-', linewidth=1, label='f(x)')
    plt.pause(0.001)
    plt.clf()
print(linear.state_dict())
f = linear.state_dict()['weight'].numpy()[0][0] * x_ + linear.state_dict()['bias'].numpy()[0]
plt.scatter(x_, y_, s=5)
plt.plot(x_, f, 'r-', linewidth=1, label='f(x)')
plt.show()
# torch.save(linear.state_dict(), 'linear.pt')  # 保存模型

对线性拟合过程做了可视化如下：