I am writing a custom layer in Keras and I wonder if there is a way to access the target values inside the layer.
class Custom(Layer):
def __init__(self, **kwargs):
super(Custom, self).__init__(**kwargs)
def build(self, input_shape):
super(Custom, self).build(input_shape)
def call(self, x):
result = K.dot(x, self.kernel)
targets = ???
return result
def compute_output_shape(self, input_shape):
return (input_shape[0], self.output_dim)
Now what I have done previously is to use a placeholder to store and update the target values at the begining of each batch. But now I am using a generator and it seems that inside a generator I cannot call the set_value function of Keras so I cant update my "target placeholder"
Is there any neat way to access the target value in the custom layer?