i modeled some regression test cases with TF/Keras (in a Python environment) and tried to train the resulting NNs with “KerasGA” from PyGad. But I was not successful:
For any test case, KerasGA converged (fast) to something like the mean value of the desired outputs, but there was no tendency for the individual output / example to be modelled…
So my questions are:
What might be the reason for this insufficient GA-based training?
Where should I look at to identify my mistake?
Are there other Python -based packages (for the GA-based training of Keras models) which I should try?
Any hints is welcome…