Keras Custom Scaling Layer

Question

For my work, I need to make a layer, that has only single weight, that will multiply the data in the current layer by some trained value. Is there a way to do this?

Or change the merge layer, which will be able to make a weighted average of input layers. Thanks

Answer 1

Try Lambda layer

model.add(Lambda(lambda x: x *MyValue))

https://keras.io/layers/core/#lambda

Answer 2

Extending @user375348's answer you can create a learned variable and use it within the Lambda :

initial_value = 1
learned = tf.Variable(initial_value, name='learned_scalar')
Lambda(lambda x: x * learned)