Attempt at adding cosine similarity metric

[jlevy44]

Hopefully I'm not too far off. Let's merge when we're both feeling confident about the changes! Thanks!

[jlevy44]

#26

I did not test this implementation. Been swamped with some other works, but I can test when the code has been shaped a bit more.

[rusty1s]

Looks good so far :)

I suggest the following changes:

• dotProduct and calc_norm should actually take in scalar_t values using templates. And maybe rename them to dot and norm?
• You said that the sklearn implementation is slower, so I think we should branch here dependent on which metric is used.
• sklearn needs to be a dependency in setup.py.

[jlevy44]

Ok, just updated a bit, still need your help to test and finish. Thanks!

[rusty1s]

Great, will try to merge tomorrow.

[jlevy44]

Awesome let me know how it goes.

[rusty1s]

Well, I cleaned up the code. It was kinda buggy, didn't you test it? In addition, there is no sklearn kNN implementation for the cosine distance, so currently I raise a NotImplementedError for the cpu version.