write all trained algo's parameters to file

[ClimbsRocks]

then, ideally, we can run a regression against them to see which ones matter for your dataset and which ones do not.

in a perfect world, we'd then use that to inform future searchCV searches

realistically, those are optimizations the user is going to have to do by hand for now.

see if we can get the results from EVERY trained algo, even the ones in searchCV that end up not being the best ones.