KBinsDiscretizer doesn't handle NAs

If I try to pass some input to KBinsDiscretizer that would contain NaNs, it will throw an error instead of ignoring the NAs:

import numpy as np
from sklearn.preprocessing import KBinsDiscretizer

X = np.arange(10).reshape((-1,1))
X[2] = np.nan
kb = KBinsDiscretizer(encode="ordinal", strategy="quantile")
kb.fit(X)

ValueError: cannot convert float NaN to integer

The bins could still be calculated if one ignores the NAs and outputs them as np.nan in the transformation.

[david-cortes]

Perhaps there could be an option about whether to treat NAs as a separate bin or to ignore them (i.e. outputting NaN for ordinal, and all-zeros for one-hot).

[glemaitre]

I think that our OneHotEncoder and OrdinalEncoder are letting pass the NaN nowadays (I am at least sure regarding the former).

On Mon, 19 Apr 2021 at 19:08, david-cortes @.***> wrote:

Perhaps there could be an option about whether to treat NAs as a separate bin or to ignore them (i.e. outputting NaN for ordinal, and all-zeros for one-hot).

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/scikit-learn/scikit-learn/issues/19920#issuecomment-822632024, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABY32P4IMMFW3JBMYC3OXF3TJRPSTANCNFSM43EUKAAQ .

– Guillaume Lemaitre Scikit-learn @ Inria Foundation https://glemaitre.github.io/