Add 2D matrices support for GapEncoder

[LilianBoulard]

Fixes #165

[GaelVaroquaux]

Thanks for the super fast implementation of this feature. I did a first full review (sorry, not as thorough as I would like, but it's late). I added some inline comments, and here are a few:

We are changing the public interface of GapEncoder in a backward incompatible way: before it would accept 1D array, and now it will throw an error. While I think that this is absolutely the right thing to do, we need to indicate this in a very clear way in our changelog, for instance by writing in bold in the beginning "Backward incompatible change to GapEncoder" and describing the change.

I think that we need to test all the public methods with 2D inputs: score, get_feature_names, transform. Just to make sure that we have no hidden problem.

Finally, I would suggest to change the default of SuperVectorizer to now use the GapEncoder (which is the motivation for this change). It will stress-test a lot this implementation, and will help us convince ourselves that it is functional.

[alexis-cvetkov]

I have made a pull request on your branch @LilianBoulard with many updates on the PR:

• docstrings
• I have made the tests more exhaustive, and replaced all 1D test inputs by 2D inputs with several columns.
• implemented the score method
• updated the get_feature_names method to add column names manually/automatically as prefixes before the labels.
• changed the SuperVectorizer to use the GapEncoder as default encoder (instead of SimilarityEncoder).
• slightly changed the tests for the SuperVectorizer to make it work with GapEncoder
• checked that the example 03 with the SuperVectorizer works well (we get similar conclusions from the features importance)

[GaelVaroquaux]

Absolutely awesome. @LilianBoulard , when you have time, can you merge @alexis-cvetkov 's branch in yours.

We'll need the note in the CHANGES.rst (unless Alexis's branch implements it), and we might be good to go.

[alexis-cvetkov]

Looks good to me !

[GaelVaroquaux]

OK, I commited a few changes directly in the PR and will merge once CI has ran.

Thanks a lot!!

[GaelVaroquaux]

Merging! Thanks!!

[GaelVaroquaux]

Here it should be (n_samples, ) because we are in the GapEncoderColoumn class. The GapEncoder expects 2D arrays, but the GapEncoderColumn expect 1D arrays.

Thank you for your review. I will fix my mistake