毕设: 协同差分进化算法

def CCDE(pop_size, generation, length, model, image, target_label, eps, batch_size, gradient):
    generation_save = np.zeros((10000,))
    problem = Problem(model, image, target_label, eps, batch_size) # 子种群的问题假定一致, 假定子种群的数量为2, 其中一个作为主种群，另一个作为辅助种群（提供优秀个体）
    subpop = [np.random.uniform(-1, 1, size=(pop_size, length)) for _ in range(2)]
    # pop = np.random.uniform(-1, 1, size=(pop_size, length))
    # for i in range(len(subpop)):
    n_subpop = len(subpop)
    if (not (gradient is None)):
        for i in range(n_subpop):
            subpop[i][0] = np.reshape(np.sign(gradient), (length)) # 初始化每个子种群
    max_eval = pop_size * generation
    eval_count = 0
    fitness, fit = np.zeros((n_subpop,)), np.zeros((n_subpop,))
    for i in range(n_subpop):
        fitness[i], fit[i] = problem.evaluate(subpop[i])
    eval_count += pop_size
    count = 0
    fitmin = np.zeros((n_subpop,))
    for i in range(n_subpop):
        fitmin[i]=np.min(fitness)
    generation_save[count] = fitmin
    F = 0.5
    if (len(np.where(fitness[0] == 0)[0]) == 0):
        while (eval_count < max_eval):
            count += 1
            for i in range(n_subpop):# 每个子种群分别进化变异
                off = mating(subpop[i],F) # 变异
                off_fitness , off_fit = problem.evaluate(off)
                eval_count += pop_size
                subpop[i] ,fitness[i] ,fit[i] = select(subpop[i],fitness[i],fit[i], off, off_fitness, off_fit) # 选择最优适应度的个体
                subpop[i], fitness[i], fit[i] = complement(fit[i], subpop[i], fitness[i], problem)
                fitmin[i] = np.min(fitness[i])
            subpop[0] ,fitness[0] ,fit[0] = select(subpop[0],fitness[0],fit[0],subpop[1],fitness[1],fit[1]) # 选择最优适应度的个体
            subpop[0], fitness[0], fit[0] = complement(fit[0], subpop[0], fitness[0], problem)
            generation_save[count] = np.min(fitmin[0])
            if (len(np.where(fitness[0] == 0)[0]) != 0):
                break
    if (len(np.where(fitness[0] == 0)[0]) != 0):
        return subpop[0][np.where(fitness[0] == 0)[0][0]], eval_count, generation_save[:count + 1]
    else:
        return subpop[0][0], eval_count/100, generation_save[:count + 1]