Describe the bug

test_deep_estimator_full[keras-adamax] is failing due to a type check in keras. See, for example

https://github.com/sktime/sktime/actions/runs/3495913658/jobs/5855642783

"ValueError: optimizer of type <class 'keras.optimizers.optimizer_v2.adamax.Adamax'> cannot be serialized, it should either be absent/None/str/tf.keras.optimizers.Optimizer object"

it could be that the wrong type of Optimizer is being used, output states

Optimizer = <class 'keras.optimizers.optimizer_experimental.optimizer.Optimizer'>
whereas it is expecting
tf.keras.optimizers.Optimizer

I don't want to hack around in the testing suite, I'll let someone who understands it better do that,