public class GuidedSplitPoints
extends java.lang.Object
Implements a method for finding new histogram bins split-points based on a result of previous binning.
Idea:
We take non-empty bins and look at the squared error they have. Based on the target bin number, we discard
the empty bins and used the frees-up space to refine the non-non empty bins. Splitting of non-empty bins
is guided by Squared Error accumulated in the bin. Bins with higher SE are split more than the bins with lower SE.
Sub-bins (bins created from a single original bin) are refined uniformly.
If uniform splitting fails in this iteration (= the distribution of values is significantly skewed), next iteration
will attempt correct the issue by repeating the procedure with new bins (we are recursively refining the promising
bins as we get deeper in the tree).