New Split Criteria Seem to Outperform Existing Criteria in Some Simulations

[morgsmss7]

My plan for the semester will consist of 3 main tasks:

1. PR existing split criteria comparison into scikit-learn
2. Clean up axis and oblique projection split criteria from last semester.

• Write more tests (random state, memory efficiency, shared weights)
• Implement shared weights (existing code for this exists, but I abandoned it last semester because I was getting errors which turned out to be unrelated. I would simply like to revert to this version, fix what ended up being the error after all, and make sure it works)

3. Write a conference paper about these new splitters and how they compare against existing splitters in different simulations.

[morgsmss7]

Part 1 will be my official Sprint 1 goal, but I'm hoping I will be able to finish 2 as a reach goal for this semester.