gradient descent 梯度下降法

import matplotlib.pyplot as plt
import numpy as np


# gradient descent
# 全部变量的偏导数汇总而成的向量称为梯度
def num_diff(f, x):  # 求单变量数值微分（偏导数），差分求导方式
    h = 1e-4
    grad = np.zeros_like(x)

    for i in range(x.size):
        tmp = x[i]

        # 差分求导 == 求微分 计算公式：  f(x+h) - f(x-h) / 2 * h
        x[i] = tmp + h
        fxh1 = f(x)

        x[i] = tmp - h
        fxh2 = f(x)

        grad[i] = (fxh1 - fxh2) / (2 * h)
        x[i] = tmp  # 将值还原

    return grad


# 梯度法
def grad_des(f, init_x, lr=0.1, step=100):
    x = init_x

    for i in range(step):
        grad = num_diff(f, x)
        x -= lr * grad
    return x


# model
def model(x):
    return x[0]**2 + x[1]**2


init_x = np.array([-3.0, 4.0])
res = grad_des(model, init_x)
print(res)