在自定义图层中使用输入“图层”的权重

Question

im试图创建自己的图层，该图层接受图层的输入并根据该输入图层的权重计算一些内容。 问题是，我得到的输入是张量，而张量不保存先前操作的权重。

class VarWeightPropagation(Layer):

def __init__(self,
             input_layer_name='Conv2D',
             **kwargs):
    '''

    Arguments:

    '''
    self.input_name = input_layer_name
    super(VarWeightPropagation, self).__init__(**kwargs)

def build(self, input_shape):
    #self.input_spec = [InputSpec(shape=input_shape[0])]
    super(VarWeightPropagation, self).build(input_shape)

def call(self, x, mask=None):
    error = x[0]
    x = x[1]
    if self.input_name == 'Dense':
        # do some calculation based on the weight of the previous layer 
    else:
        weights = x.get_weights()[0] # this is the main problem as tensors dont have get_weight()
        # also do some calculations based on that weights
    return (x,error)

def compute_output_shape(self, input_shape):
    return (input_shape[0], self.output_dim)

是否有一种方法可以在我创建的自定义中使用上一层的权重？ （该层仅在网络的推理状态下处于活动状态） 预先谢谢你！