Thanks for the answer! Gradient boosting is good because it allows you to get good baselines, without going into the nature of the data. Therefore, intuition dictates that it is best to break into Bins by quantiles. Therefore, the question to you, as to the developer h20: why you have implemented so many options and for what purposes they can be more preferable (or how AUTO works, schematically)? h2o-release.s3.amazonaws.com/h2o/rel-turnbull/1/docs-website‌​/…