Question

情况：我正在写一个张量流程序。我试图创建一个损失函数来比较克矩阵。计算克数矩阵似乎进展顺利 - 或者至少程序没有崩溃。然而，在我计算克数矩阵后，我想做元素减法，元素方形平方，然后总结所有元素。当我尝试进行元素减法时，程序似乎崩溃了。这是（我认为）相关的代码：

def _gram_matrix(self, F, N, M):
    print("in gram")
    F = np.reshape(F,(M,N))
    F_trans = np.transpose(F)


    return np.dot(F,F_trans)

def _single_style_loss(self, a, g):
    print("in single style loss")
    print("a shape ",a.shape)
    N = a.shape[3]
    M = a.shape[1]*a.shape[2]
    gram_a = self._gram_matrix(a,N,M)
    gram_g = self._gram_matrix(g,N,M)
    print("out of gram")
    multiplier = 1/(4*N*N*M*M)
    print("leaving single style loss")
    return_val = np.subtract(gram_a,gram_g)
    print("subtract was fine")
    return_val = np.square(return_val)
    print("square was fine")
    array_sum = np.sum(return_val)
    print("sum was fine")
    array_sum = multiplier*array_sum
    print("multiplier was fine")
    print("going to return ",array_sum)
    return array_sum

def _style_loss(self, A):
    print("in style loss")
    loss = 0;
    with tf.Session() as sess:
        sess.run(self.input_img.assign(self.initial_img))
        generated_layers = sess.run([getattr(self.vgg,layer) for layer in self.style_layers])
        sess.run(self.input_img.assign(self.style_img))#reset input_img to style img
        for i in range(len(self.style_layer_w)):
            print("len ",self.style_layer_w)
            print("len generated layers",len(generated_layers))
            print("len style layers",len(A))
            loss += self.style_layer_w[i]*self._single_style_loss(A[i],generated_layers[i])
            print("loss ",loss)
        self.style_loss = loss
    print("out of style loss")

所以我得到的错误是：＆＃34;留下单一风格的损失被杀：9＆＃34;这就是为什么我总结下一行（np.subtract行）是问题。可悲的是，我不知道为什么这条线是一个问题，我会非常感谢任何人提供的洞察力。谢谢！

为什么np.subtract导致＆＃34;被杀9＆＃34;

0 个答案: