[WIP] kwargs flow through for Image Similarity -> Nearest Neighbors

[nishantnath]

This will allow users to use methods for nearest neighbor model apart from 'auto' along with other params like distance & nearest neighbor's kwargs

Nearest neighbor already checks for validity of kwargs downstream.

[TobyRoseman]

@nishantnath - thanks for the pull request. I like this idea and the change looks good. Would it be possible to add a unit test for this new functionality?

[nishantnath]

@TobyRoseman I did ponder over a unit test but considering kwargs would be passing through to downstream functions and each downstream function would (or should) have its own checks for params & unit tests - seemed redundant.

Downstream methods:
feature_extractor.extract_features (doesn't support kwargs)
nearest_neighbors.create (supports kwargs): has own unit test + kwargs param check

Happy to add an unit test if you can provide some pointers on the direction you are envisioning for the unit test.

[TobyRoseman]

Please change tc.nearest_neighbors.create() to :py:func:turicreate.nearest_neighbors.create()`. That will cause the text, in our API docs, to be linked to the appropriate documentation page for that function.

[TobyRoseman]

@TobyRoseman I did ponder over a unit test but considering kwargs would be passing through to downstream functions and each downstream function would (or should) have its own checks for params & unit tests - seemed redundant.

Downstream methods:
feature_extractor.extract_features (doesn't support kwargs)
nearest_neighbors.create (supports kwargs): has own unit test + kwargs param check

Happy to add an unit test if you can provide some pointers on the direction you are envisioning for the unit test.

@nishantnath - I agree a unit test here is largely redundant. However I still think it's worth doing since it will ensure we never accidentally remove this functionality.

If I was going to add a unit test for this, I would probably add a new add a new testing class to our image similarity unit test. The setUpClass method could create a model with a small dataset and pass it a kwarg value.
Something like this:

self.model = tc.image_similarity.create(data.head(), distance='squared_euclidean')

Then we could have a single test method that just checks the similarity model has that distance. I think it would be something like:

self.assertEqual(
    self.model.similarity_model.distance[0][1],
    'squared_euclidean'
)

[TobyRoseman]

@nishantnath - thanks for adding a unit test.

[TobyRoseman]

I don't think you need the majority of the code in this setUpClass. It looks like the only member variable that's actually getting used is self.model. I think everything else should be removed.

[TobyRoseman]

What are you trying to accomplish by this change?