Optimal binning

[NicolasC]

Hi there

I am still quite confused using the Transformation Node, especially the optimal binning transform. I have read the other thread regarding the diff. between Interactive Binning and Optimal Binning. It was mentioned that the bins in Optimal binning were created according to the target. Why is it then that when running Interactive Binning one has to set 'Yes' to the target, almost as to mention the target has the role to play in the binning, while it is not the case for Optimal Binning. Seems like Interactive Binning takes into account the target, not Optimal Binning. 

Although, second question, when I edit variables on a node after Optimal Binning has been performed, I always have a few (of course numeric) variables than were not binned....would there be a specific reason for that?

Many thanks

Nicolas

[WendyCzika · Accepted solution]

The Interactive Binning node only does quantile or bucket binning for an interval input, and grouping of rare levels for a nominal input - no optimal binning.  The target is only used for the variable selection based on the Gini statistic.

In the Transform node, the Optimal binning does in fact use the target to do tree-based binning.  I'm guessing the ones not binned do not have a binning that optimizes the relationship with the target.  Do you see messages like this in the SAS log in the Results?

WARNING: Optimal Bin Transformations for Trans did not find any transformations for variable ZZZ

[NicolasC]

Hi Wendy

Thanks for your explanation. I indeed see this warning message for the non-binned variable in the log. I can close this thread. Thanks again. Nicolas