Quantile-binning
Quantiles are specific values or cut-points that help in partitioning the continuous-valued distribution of a specific field into contiguous intervals.
Why it doesn’t really make sense
Classification (softmax) in bins doesn’t really make sense in general - you are really creating two problems (how to bin, and classification) where there used to only be one (regression).
However, it is dependent on how the target distribution is structured. If it is unimodal, then using the L1 or L2 norm should work well, but if the target is beyond unimodal, then the model might not work as well. 
— uuid: 202004250332 date: Apr 25, 2020 tags: #raise