python - how to update a parameter (i.e. dropout rate or units) at each epoch within a keras model

Question

I am trying to add another parameter to Keras's implementation of deep learning architecture which changes at each or after a number of epochs.

Assume in new architecture (CNN, RNN, etc.), a parameter 'alpha1' is added, and I want to initialize it with a value for example 16,

Now, at the time of training, at each epoch, I want to update the alpha1. Suppose, at each epoch, alpha1 = alpha1 * somevalue.

Since in keras/../recurrent.py, step(.) function where the computations are made is only called one time (not at every epoch), I could not add the update of a parameter in here. Is there any way of updating a parameter in keras model during training?

question from:https://stackoverflow.com/questions/65909353/how-to-update-a-parameter-i-e-dropout-rate-or-units-at-each-epoch-within-a-ke

Answer 1

A simple solution is to loop the training as followed.

for epoch in range(epochs):
  model.fit(X=... , epochs=1)
  alpha1 = alpha1 * somevalue