如何在Keras训练期间记录张量的值？

Question

I have a customized layer in Keras, and I want to monitor a specific tensor during the training (ie, evaluate this tensor per epoch, and record the value to a list). What should I do to achieve this?

A piece of code as an example:

class customizedLayer(Layer):
    def __init__():
        ...
    def build():
        ...
    def call():
        tensor_1 = ...
        tensor_2 = ...
        return ...

output = customizedLayer(input)
model = Model(input=X, output=Y)
model.compile(...)
model.fit(...)

what should I do to get the values of tensor_1 or tensor_2 shown above?

Many thanks.

Answer 1

You may implement a Callback object to save that values to a file each epoch I think. If I've got something more specific I'm telling you.