sa.py

import numpy as np
import math
import random
import matplotlib.pyplot as plt
from copy import deepcopy
from tqdm import tqdm

def schaffer(x):
    # x[0] in [-100, 100], x[1] in [-100, 100]
    r = x[0] ** 2 + x[1] ** 2
    f = 4.5 - (np.power(np.sin(np.sqrt(r)), 2) - 0.5) / np.power((1 + 0.001*r), 2)
    return f

def schaffer_std(x):
    return 5 - schaffer(x)


def simulated_annealing_2d(
        func,
        bounds = [(-100, 100), (-100, 100)],
        T_init = 1000,
        T_final = 1e-4,
        alpha = 0.998,
        step_size = 0.3,
        max_iter = 1000,
    ):

    x_c = np.array([random.uniform(bounds[0][0], bounds[0][1]), random.uniform(bounds[1][0], bounds[1][1])])
    f_c = func(x_c)

    x_best = deepcopy(x_c)
    f_best = func(x_best)

    search_history = []

    t_c = T_init

    while t_c > T_final:
        for _ in range(max_iter):
            x_new = x_c + np.random.uniform(-step_size, step_size, size=2)
            x_new[0] = np.clip(x_new[0], bounds[0][0], bounds[0][1])
            x_new[1] = np.clip(x_new[1], bounds[1][0], bounds[1][1]) 

            f_new = func(x_new)
            delta = f_new - f_c

            if delta < 0 or random.random() < math.exp(-delta / t_c):
                x_c = x_new
                f_c = f_new
                if f_c < f_best:
                    x_best = deepcopy(x_c)
                    f_best = f_c

                    search_history.append(x_best.copy())

        
        t_c *= alpha

    return x_best, f_best, search_history


if __name__ == "__main__":

    np.random.seed(42)
    random.seed(42)

    x_result, f_result, search_history = simulated_annealing_2d(schaffer_std)

    f_right = schaffer(x_result)

    print(f"x = {x_result}, f = {f_right}")

    x = [p[0] for p in search_history]
    y = [p[1] for p in search_history]

    plt.plot(x,y)
    plt.show()