Put variable importance in context

[kescobo]

From @andrewGhazi:

It's kind of hard to understand the scale of RF importance (probably moreso for readers without ML background). If you were to repeatedly run the RF while including a pure noise covariate, could you show the distribution of the importance scores for that noise variable as a way of grounding these scores against a sort of baseline?

[kescobo]

I think this is good - yes?