You could go through the tests to see if there are any failures that I missed. LMK if you need any help. This is almost done, great work!

@fkiraly @achieveordie can you please review test results? it is failing for assertion error. Thank you

Regarding AssertionError in test_fit_does_not_overwrite_hyper_params:

An excellent way to see what went wrong is to check what the error is, here it says:

AssertionError: Estimator ResNetClassifier should not change or mutate the parameter optimizer from None to <keras.optimizers.optimizer_experimental.adam.Adam object at 0x7fbc8d5f2470> during fit.

which makes sense since we don't want to change parameters passed by the user. For that purpose, we actually use self.optimizer_ (and not self.optimizer) as the parameter we set in case the user doesn't provide any value. You can check out any of the accepted estimators to compare your own model since the only change thing that changes is the actual implementation of the estimator, the mechanics remain similar, if not the same.

Regarding AssertionError in test_fit_idempotent:

This is a known problem in DL estimators that sktime currently faces - check out issue #3616 for more details. All existing DL estimators, therefore, skips this test. This can be done by adding this estimator in sktime/tests/_config.py as hyperlinked.

(While we are still changing things, another easy win is renaming lr to learning_rate in the same line to avoid this warning.)

(Also, I would highly recommend adding yourself as a contributor by editing .all-contributorsrc). Check out this subsection.