MLKR + KNRegressor outperforms RFR in materials dataset

[angelotc]

Hello,

Just wanted to share my implementation of MLKR + Bagged KNRegressor. Through my team's tests, it was able to outperform RFR and performs similar to a Bagged LASSO when testing on "unseen" data. We developed our own splitter for this "unseen" data test, which can be seen from our slides (Slide 3) . Still trying to understand the math still on this, but I thought I should share.

Code example:
https://github.com/leschultz/materials_application_domain/blob/angelo_dev/examples/MLKR_KNN.py

MLKR KNN Implementation :
https://github.com/leschultz/materials_application_domain/blob/angelo_dev/mad/ml/MLKR_KNN.py

Slides:
https://docs.google.com/presentation/d/1FvytXnCkNnd-S-pfTHZeczEzPV5q6JrxVv9GtP3CsPM/edit#slide=id.g106dd184839_0_8