pytorch 线性回归代码

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

# 样本数：1000
num_examples = 1000

# 设置权重
true_w = 3
# 设置偏置
true_b = 20

# 1000个特征数为2的样本
delta_x = torch.randn(num_examples, dtype=torch.float32)
meta_x = torch.arange(0, num_examples, dtype=torch.float32)
meta_x = delta_x + (0.01 * meta_x)

# 计算出对应的标签
meta_y = true_w * meta_x + true_b
# 给标签加噪音
meta_y += torch.tensor(np.random.normal(0, 15, size=meta_y.size()), dtype=torch.float32)

meta_point = np.dstack((meta_x.numpy(), meta_y.numpy()))[0]
# 打乱
np.random.shuffle(meta_point)

points = torch.from_numpy(meta_point)

plt.scatter(points[:, 0].numpy(), points[:, 1].numpy(), c='', edgecolors='b', s=15)
plt.show()


# y = wx + b
# 真实值 与 预测值 的方差函数
def compute_error_for_line_given_points(b, w, points):
    """
        计算给定超参数[w,b]的误差值
    """
    totalError = 0
    for i in range(0, len(points)):
        x = points[i, 0]
        y = points[i, 1]
        totalError += (y - (w * x + b)) ** 2
    return totalError / float(len(points))


# 求导函数
def step_gradient(b, w, points, lr):
    b_gradient = 0
    w_gradient = 0
    N = float(len(points))
    for i in range(len(points)):
        x = points[i, 0]
        y = points[i, 1]
        b_gradient += -(2 / N) * (y - ((w * x) + b))
        w_gradient += -(2 / N) * x * (y - ((w * x) + b))
    b_new = b - (lr * b_gradient)
    w_new = w - (lr * w_gradient)
    return [b_new, w_new]


# 梯度下降
def gradient_descent_runner(points, b, w, lr, iterations):
    for i in range(iterations):
        b, w = step_gradient(b, w, np.array(points), lr)

    return [b, w]


def run():
    lr = 0.01
    initial_b = 0
    initial_w = 0
    iterations = 2000
    print(
        f"Starting project descent at b = {initial_b}, w = {initial_w},error = {compute_error_for_line_given_points(initial_b, initial_w, points)}")
    print('\nRunning...')
    [b, w] = gradient_descent_runner(points, initial_b, initial_w, lr, iterations)
    print(f"\nAfter project descent at b = {b}, w = {w},error = {compute_error_for_line_given_points(b, w, points)}")
    print('\nb:{}，w:{}'.format(b, w))
    paint_line(b, w)


# 画出直线
def paint_line(b, w):
    x = points[:, 0]
    y = w * x + b
    plt.scatter(points[:, 0], points[:, 1], c='r', edgecolors='b', s=15, label='orginal')
    plt.plot(x, y, c='red', label='predict', linewidth=2.0, linestyle=':')
    plt.legend()
    plt.show()


if __name__ == '__main__':
    run()