handle_unknown about category_encoders HOT 3 CLOSED

On Feb 11, 2018, at 5:56 AM, Gavin Smith ***@***.***> wrote: Yes, I can confirm this error (same code) v1.2.6. This looks like a bug. The statement block in question is: if X[~X['D'].isin([str(x[1]) for x in switch.get('mapping')])].shape[0] > 0: raise ValueError('Unexpected categories found in %s' % (switch.get('col'),)) At this point the code is looping over each column and converting the string representation to the ordinal one - if there is no match in the previously learnt mapping then it is mapped to NaN. Then we look to impute and/or check which, assuming we've picked this option, is when the above block is run. The above check is a generic though, and just checking if the (newly mapped) ordinal encoding values in the column X['D'] are valid. Of course since we are looping over columns in X (and in any case have an arbitrary DataFrame) the column 'D' almost certainly doesn't exist and this is a bug. The current column is given by switch.get('col'). As such the corrected code is: if X[~X[switch.get('col')].isin([str(x[1]) for x in switch.get('mapping')])].shape[0] > 0: raise ValueError('Unexpected categories found in %s' % (switch.get('col'),)) — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.