Woe won't work with NaNs

[rcg-uab]

Thanks for putting this package together - it's great to have a scorecard package in Python.

However, it appears the weight of evidence binning algorithm only works with complete data, even though it should factor in missing data.

As a simple test, I added a column to the germancredit.csv file and included some NaN data in the new column and reran the example code. The woebin function breaks as described in other threads (e.g. #78 ). Is this on the radar for a fix?

Cheers,
Ryan

[ShichenXie]

I cant reproduce your issue. The package should be able to handle missing values. Please upgrade your package to the latest version on the Github and try again.

[theinexorable]

I was able to trace this back to the line 126 (in 1.9.2 available in Pypi it was 116) in woebin.py.
The specific code is:
dtm = dtm[~dtm.index.isin(dtm_sv.index)].reset_index() if len(dtm_sv.index) < len(dtm.index) else None
which deletes the rows from dtm (the dataset for the final table) which are the same as the one from a list of missing values (i.e. from 0 up to w/e number of missing you have). I can't make sense of this and the reason it's still included, but this is where the rows with missing values are deleted.
As a workaround for now, I can just comment or delete this line and everything works perfectly.