Specifying the order of encoding in Ordinal Encoder

Question

I'm using OrdinalEncoder, and I cannot find how to specify the encoding order. I mean that I have categories like "bad", "average", "good" which naturally have an order. But I want to specify that order, since the encoder cannot know itself the meaning of categories. Indeed, with categories='auto', some categories are encoded in wrong direction with respect to some others and I do not want this because I know, at least for some of them, if the correlation is positive or negative.

But specifying the categories results in an error during fitting:

'OrdinalEncoder' object has no attribute 'handle_unknown'.

If I do not specify the categories, fitting process goes well, and I do not understand why (the attribute "categories_", after fitting, shows me the same categories I enter by hand when I try to specify them).

I specify the categories as a list of lists. Here what happens without specifying categories.

import pandas as pd
from sklearn.preprocessing import OrdinalEncoder

df = pd.DataFrame(np.array([['a','a','a'], ['b','c','c']]).transpose())
oE = OrdinalEncoder(categories='auto')
oE.fit(df)

print(oE.categories_)

Resulting in: [array(['a'], dtype=object), array(['b', 'c'], dtype=object)]

Specifying categories explicitely:

df = pd.DataFrame(np.array([['a','a','a'], ['b','c','c']]).transpose())
oE = OrdinalEncoder(categories=[['a'], ['b', 'c']])
oE.fit(df)

The result is this error:

Traceback (most recent call last):

File "", line 3, in oE.fit(df)

File "/home/alessio/anaconda3/lib/python3.6/site-packages/sklearn/preprocessing/_encoders.py", line 774, in fit self._fit(X)

File "/home/alessio/anaconda3/lib/python3.6/site-packages/sklearn/preprocessing/_encoders.py", line 85, in _fit if self.handle_unknown == 'error':

AttributeError: 'OrdinalEncoder' object has no attribute 'handle_unknown'

Answer 1

I had the same problem. This is bug in scikit-learn, already fixed and added to version 0.20.1, which is still not released. https://github.com/scikit-learn/scikit-learn/issues/12365

I solved it temporarily by copying fixed _encoders.py to my project and using.

from _encoders import OrdinalEncoder