How to add trainable variable in Python Tensorflow

Question

I would like to add trainable parameter 'p' into my neural network model coded by Python tensorflow. The parameter 'p' should range between 0 and 1. It can be ranged between 0 and 1, transformed by sigmoid function (e.g. p = 1/(1 + exp(-pi)) and in this case, values obtained from sigmoid function should be normally distributed. Please let me know if you have any ideas to solve this problem. Thank you.

Answer 1

You probably want to do tensorflow layers subclassing and add your p trainable variable through self.add_weight in your subclass.

Applying a sigmoid is one way to have your value between 0 and 1. Another way is through the constraint parameter of self.add_weight (It will clip your value between 0 and 1).