binning_table on a subset of the original sample used for fitting

[chapmanh]

Hi,
This might not be statistically appropriate to do, nonetheless I'm looking to use OptimalBinning to fit a variable to a binary target, and then look at how a subset of the data performs using identical bins to the originally fitted data.

I'm achieving this by providing a new OptimalBinning object the splits obtained in the original fitting and then 'fitting' the subset to these splits with monotonic_trend set to None to prevent shifting. I'm also interested in how the individual IVs change within each bin depending on the subset which is used.

there are three suggestions off the back of this:
one rather radical;

1. The most radical solution to help me would be some sort of method in binning_table which can ingest a completely different variable array and build a table as if this was the data that created the bins

and two less intense ideas!
2. In instances where the subset does not span the entire range of the original sample, the fitting errors if there are no samples in the bin. Could this default to all zeroes, much like the 'Special' bin already does?
3. Could metric='iv' be an option when using binning_table.plot? This would be really helpful as currently I can only view WoE and event rate within each bin. Thinking about this one, perhaps I just need a crash course on building my own plots from scratch!

Many thanks,
H