I have a custom train_step and test_step function in my class. After specifying the keras tuner and running tuner.search, I get an error specifying incorrect size, which to me indicates that the default fit function is used.
It does work when i use the model.fit of my class, but with the tuner it doesn't. The only thing I have changed is that I have copied my class and made it a HyperModel instead of a regular Model.
Is there a way to have it use the custom train en test functions without having a fully custom training loop?
Edit: one other difference is that you have to return the model in the build function, which could mean it uses the model's fit instead of the class' fit.