core.matrix support

[mikera]

Thanks for creating a very interesting library, I'm interested to use it for some of my own machine learning projects.

One thing that would be really useful is core.matrix support. core.matrix already provides protocols for sparse table based datasets and it should be fairly simple to implement these for BTables.

This would give a number of advantages:

• You could use core.matrix functions directly on BTables (rows, labels etc.)
• Better interop with Incanter / other core.matrix based libraries
• You would get all the other core.matrix functionality essentially for free through the default implementations
• People using core.matrix could switch to BTables very easily when it meets their use cases
• Help with interoperability in the Clojure numerical computing ecosystem

Any interest in implementing this? Or getting a PR?

[andrewberls]

Hi Mike - yes I think core.matrix support would be very useful! I likely don't have time to tackle an integration immediately, but I'd be happy to review a PR anytime.

I'm not deeply familiar with the core.matrix protocols - I'd imagine that BTables are only suitable for a subset of operations, as they're not really designed for efficient columnar access or anything like that and default impls would suffer I think. Is it possible to only define implementations for row-wise operations?