Understand our data copying

In @bchen1116 's recent analysis of our ensembling algorithm, it's come to light that we spend a lot of time copying data. I've noticed throughout the code a lot of copying of dataframes and think we should understand all these copies and whether they're necessary or just impediments to performance.

Acceptance criteria for this issue is a benchmarking and profiling of our unit test suite and visualization with snakeviz. @bchen1116 if you could also link or copy/paste your results you showed earlier today, that would be helpful to highlight the problem.