future ideas:

[ClimbsRocks]

1. voluntary data gathering to run ML on our own library
   once trained, ask users if they'd be ok sharing some summary info about their data and the parameters of their models (totally anonymized) with us
   we could then figure out what combinations of params work best for data of a certain shape (num features, standard dev, min, max, range, num data points, type of predicted output, num of categories being predicted, etc.).
   offer a "Quickstart" option- default to whatever we predict will be the best params for your data set, and train algos with those- don't traverse the param space trying all combos of params.
   1. create a UI to make this even more accessible (and create pretty visualizations while training!), and collect that summarized data by default (with very obvious notifications about this and a way to turn it off)
   2. create an automated deployment script for aws so people can run this on a big box in the cloud super easily