User splits in binning_fit_params not working properly

[vincentfortin]

I am fitting a scorecard with a BinningProcess and when I use binning_fit_params to specify user_splits, it skips one of the splits.

I have binning_fit_params like {'VAR1':[3000, 5000, 7000, 9000, 11000]}, but when I fit and transform, I get only 5 splits (other than missing/special), 0-3000, 3000-5000, 5000-9000, 9000-11000 and 11000+, where I should get 6.

I have a min IV of 0.05, but this shouldn't change anything since the variable would be removed. I also have max_n_bins=20.

I only tested reducing min_prebin_size to 0.01 which didn't change anything.

Also, between 7000-9000, there are close to 10% of the total observations, same for 5000-7000.

Please let me know if you have any ideas
Thanks !

[guillermo-navas-palencia]

HI @vincentfortin. Did you check if the number of events (assuming a binary target) is > 0 in [5000, 7000] ?