Adding types on the public API surface would allow us to do some runtime type checking later on and would allow user's IDE to have more info for static analysis.

The functions/signatures to type are the ones listed here https://github.com/keras-team/autokeras/blob/master/autokeras/__init__.py

For the context, see #856 where I add some type information on a ImageClassifier.

This issue can

It will be very valuable to have a best practices documentation on how one should scale a problem on large-scale systems with DeepHyper. What scale sizes should one evaluate, such as 8, 16, 32 nodes first? Next, when should one be ready to scale out and the parameters involved here.