Speed ups, and bug fix

[aikramer2]

There is currently a bug where if training_labels are not passed, but the datamanager is being asked to include y anyway in generate_sample.

https://github.com/datascienceinc/Skater/blob/master/skater/core/global_interpretation/feature_importance.py#L98

I've also sped up data access by removing support for user defined indexes. Instead, a datamanager's index is a property that returns a range(0, len(data)). This way we no longer have to make calls to:

[datamanager.index(i) for i in indices]

[pramitchoudhary]

@aikramer2 can you check on the travis build? Also, for these changes, you synced from master right?

[aikramer2]

@pramitchoudhary yeah, freshly fetched; and ill resolve the build errors this weekend.

[pramitchoudhary]

@aikramer2 any updates on this? Do you think we can get this in the next couple of weeks?

[pramitchoudhary]

@aikramer2 any update here? Was wondering if when can we resolve this? Thoughts?

[pramitchoudhary]

@aikramer2 will you have time soon, to check on this PR?