python实现梯度下降

import numpy as np

x1 = np.array([2.104, 1.600, 2.400, 1.416, 3.000])  # 数值过大会报错
x2 = np.array([3, 3, 3, 2, 4])
y = np.array([4.00, 3.30, 3.69, 2.32, 5.40])
alpha = 0.01  # 学习率0.01
# 赋初始值都为0
th0 = 0
th1 = 0
th2 = 0


def sum(x1, x2):
    return th0 + th1 * x1 + th2 * x2  # 求和函数


for i in range(1000):    # 迭代1000次
        th0 = th0 - alpha * (sum(x1[i % 5], x2[i % 5]) - y[i % 5])  # 梯度下降公式
        th1 = th1 - alpha * (sum(x1[i % 5], x2[i % 5]) - y[i % 5]) * x1[i % 5]
        th2 = th2 - alpha * (sum(x1[i % 5], x2[i % 5]) - y[i % 5]) * x2[i % 5]

        gred0 = (sum(x1[i % 5], x2[i % 5]) - y[i % 5])              # 梯度公式
        gred1 = (sum(x1[i % 5], x2[i % 5]) - y[i % 5]) * x1[i % 5]
        gred2 = (sum(x1[i % 5], x2[i % 5]) - y[i % 5]) * x2[i % 5]

        # 条件判断，偏导趋于零
        if (abs(gred0) < 0.01 and abs(gred1) < 0.01 and abs(gred2) < 0.01):
            print(th0, th1, th2)
            print(gred0, gred1, gred2)
            break