如何为遗传算法编写交叉函数

Question

def crossover(parents):
    result = copy.deepcopy(parents[0].brain)
    for param in result.parameters():
        print((param.shape[0] / 2))
        if len(param.shape) == 2:  # weights of linear layer
            for i0 in range(param.shape[0]):
                for i1 in range(param.shape[1]):
                    if i0 < (param.shape[0] / 2) and i1 < (param.shape[1]):
                        param[i0][i1] = [param[i0][i1] for param in parents[0].brain.parameters() if
                                         len(param.shape) == 2]
                        # print("x is", x)
                        # print(f"param{i0}{i1}", param[i0][i1])
                        # param[i0][i1] = 2
                    else:
                        param[i0][i1] = [param[i0][i1] for param in parents[1].brain.parameters() if
                                         len(param.shape) == 2]

我正在使用pytorch库，并尝试为我的项目编写一个交叉函数。最初，为了获得交叉结果，我只是为了轻松而复制了父母之一的大脑。然后我从中间切入，并合并其中两个父母的大脑。问题是此“ param [i0] [i1]的parents [1] .brain.parameters（）中的param的输出”出来的是三个张量的列表，不应该这样。输出

x is [tensor(0.0775), tensor(0.0372), tensor(0.2628)]
param07 tensor(0.0775)

我该如何解决？

Answer 1

我试图在此处使用列表理解功能时出现了问题，我认为我仍然还不完全了解。所以我为此写了一个合适的循环，它起作用了。

def crossover(parents):
    result = copy.deepcopy(parents[0].brain)
    for param in result.parameters():
        if len(param.shape) == 2:  # weights of linear layer
            for i0 in range(param.shape[0]):
                for i1 in range(param.shape[1]):
                    if i0 < (param.shape[0] / 2) and i1 < (param.shape[1]):
                        for p in parents[0].brain.parameters():
                            if len(p.shape) == 2 and param.shape == p.shape:
                                param[i0][i1] = p[i0][i1]
                    else:
                        for p in parents[1].brain.parameters():
                            if len(p.shape) == 2 and param.shape == p.shape:
                                param[i0][i1] = p[i0][i1]