无法从keras中的自定义图层获取渐变

Question

我编写以下图层来模拟喀拉拉邦的辍学层：

class ShakeoutDense(Dense):

def __init__(self, p, **kwargs):
    self.p = p
    super(ShakeoutDense, self).__init__(**kwargs)

def generate_bernoulli_matrix(self, shape):
    r_matrix = tf.random.uniform(shape=shape, maxval=1)
    b = tf.math.greater(self.p, r_matrix)
    f = tf.cast(b, dtype=tf.float32)
    f = Lambda(lambda x: K.stop_gradient(x))(f)
    return f

def build(self, input_shape):
    super(ShakeoutDense, self).build(input_shape)

def call(self, inputs, **kwargs):
    input_dim = inputs.get_shape().as_list()[-1]
    kernel_shape = [input_dim, self.units]
    mask = self.generate_bernoulli_matrix(kernel_shape)
    weight = self.kernel * mask
    output = K.dot(inputs, weight)
    if self.use_bias:
        output = K.bias_add(output, self.bias, data_format='channels_last')
    if self.activation is not None:
        output = self.activation(output)
    return output

我收到以下错误：

ValueError: An operation has `None` for gradient. Please make sure that all of your ops have a gradient defined (i.e. are differentiable). Common ops without gradient: K.argmax, K.round, K.eval.

我放了Lambda(lambda x: K.stop_gradient(x))来防止传播，但这是行不通的！我该怎么办？ ：-\