下山单纯形法求函数最小值

上智能优化课的作业，老师布置的作业。无奈上课没听，不知道怎么做。找了一圈也没找到好用的代码，就一个C++的，用起来还麻烦。于是我索性自己写了一个python版的。

import numpy as np

def func(x, y):
    return x ** 2 - 4 * x + y ** 2 - y - x * y

def go(good: np.array, medium:np.array, bad:np.array, f: 'function'):
    alpha = good - bad
    beta = medium - bad
    new = bad + alpha + beta        # 先做一次反射
    if f(*new) < f(*good):              # 反射点比最好点好
        new_ex = new + (alpha + beta) / 2   # 扩张反射点
        if f(*new_ex) < f(*new):            # 扩张点比反射点好
            return [good, medium, new_ex]
        return [good, medium, new]
    elif f(*new) > f(*good):            # 反射点比最好点差
        new_sr = new - (alpha + beta) / 4   # 收缩扩张点
        if f(*new_sr) < f(*bad):            # 收缩点1比最坏点好
            return [good, medium, new_sr]
        bad_sr = bad + (alpha + beta) / 4   # 最坏点收缩
        if f(*bad_sr) < f(*bad):            # 收缩点2比最坏点好
            return [good, medium, bad_sr]
        good2 = bad - alpha / 2             # 压缩
        medium2 = medium - beta / 2
        return [good2, medium2, bad]
    else:
        return [good, medium, new]

def simplex(x0, f, d, n=100):
	'''
	x0: 初试点
	f: 目标函数
	d: 精度（没用上）
	n: 最大迭代次数
	'''
    sol = [np.array(x) for x in x0]
    pre_val = 1e10
    while True:
        sol = sorted(sol, key=lambda x: f(*x))
        if n < 0:       # 防止循环次数过多
            return
        n -= 1
        val = f(*sol[0])
        # if abs(val - pre_val) < d:	# 本来想用差值小于定值作为循环结束条件
        #     return					# 结果可能是我逻辑有问题，有时候一次迭代后函数值不变
        pre_val = val
        print('solution: ', [list(np.round(x, 5)) for x in sol], ', value: ', [f(*x) for x in sol], ', best: ', val)
        sol = go(*sol, f)

if __name__ == '__main__':
    simplex([[0, 0], [1.2, 0], [0, 0.8]], func, 1e-5, 24)

如果代码有问题欢迎大佬指正，我感觉可能有点小问题，不过可以正常求解。

solution:  [[1.2, 0.0], [0.0, 0.8], [0, 0]] , value:  [-3.36, -0.15999999999999992, 0] , best:  -3.36
solution:  [[1.8, 1.2], [1.2, 0.0], [0.0, 0.8]] , value:  [-5.88, -3.36, -0.15999999999999992] , best:  -5.88
solution:  [[1.8, 1.2], [2.25, 0.5], [1.2, 0.0]] , value:  [-5.88, -5.3125, -3.36] , best:  -5.88
solution:  [[2.85, 1.7], [1.8, 1.2], [2.25, 0.5]] , value:  [-6.932499999999999, -5.88, -5.3125] , best:  -6.932499999999999
solution:  [[2.85, 1.7], [2.3625, 1.925], [1.8, 1.2]] , value:  [-6.932499999999999, -6.63578125, -5.88] , best:  -6.932499999999999
solution:  [[3.00937, 2.11875], [2.85, 1.7], [2.3625, 1.925]] , value:  [-6.986923828125, -6.932499999999999, -6.63578125] , best:  -6.986923828125