`ToDatetime` does not implement `get_feature_names_out`

[auguste-probabl]

Describe the bug

I believe this is the case for other single-column transformers, as discussed here.

Context is probabl-ai/skore#2042: get_feature_names_out is used to get the feature names before the last step of a sklearn Pipeline.

Steps/Code to Reproduce

import pandas as pd
from sklearn.utils.validation import check_is_fitted
from skrub import ToDatetime

login = pd.Series(["2024-05-13T12:05:36", None, "2024-05-15T13:46:02"], name="login")
encoder = ToDatetime()
encoder.fit_transform(login)
encoder.get_feature_names_out()

Expected Results

["login"]

Actual Results

AttributeError: 'ToDatetime' object has no attribute 'get_feature_names_out'

Versions

System:
    python: 3.12.8 (main, Dec  3 2024, 18:42:41) [GCC 13.3.0]
executable: /home/auguste/Desktop/work/skore/.venv/bin/python3
   machine: Linux-6.12.29-x86_64-with-glibc2.40

Python dependencies:
      sklearn: 1.7.2
          pip: 24.3.1
   setuptools: 80.7.1
        numpy: 2.1.3
        scipy: 1.15.3
       Cython: None
       pandas: 2.2.3
   matplotlib: 3.10.3
       joblib: 1.5.2
threadpoolctl: 3.6.0

Built with OpenMP: True

threadpoolctl info:
       user_api: blas
   internal_api: openblas
    num_threads: 12
         prefix: libscipy_openblas
       filepath: /home/auguste/Desktop/work/skore/.venv/lib/python3.12/site-packages/numpy.libs/libscipy_openblas64_-ff651d7f.so
        version: 0.3.27
threading_layer: pthreads
   architecture: Haswell

       user_api: blas
   internal_api: openblas
    num_threads: 12
         prefix: libscipy_openblas
       filepath: /home/auguste/Desktop/work/skore/.venv/lib/python3.12/site-packages/scipy.libs/libscipy_openblas-68440149.so
        version: 0.3.28
threading_layer: pthreads
   architecture: Haswell

       user_api: openmp
   internal_api: openmp
    num_threads: 12
         prefix: libgomp
       filepath: /home/auguste/Desktop/work/skore/.venv/lib/python3.12/site-packages/scikit_learn.libs/libgomp-a34b3233.so.1.0.0
        version: None
0.6.2

[rcap107]

As it turns out, #1622 did not fix this bug

Thanks for the issue @auguste-probabl, I'll work on a fix

[rcap107]

@GaelVaroquaux this is the third issue that was opened because of this problem, I want to deal with it properly now.

The problem stems from the fact that like most other transformers ToDatetime inherits from the SingleColumnTransformer, which does not have a get_feature_names_out method anymore (partially due to the other issues I mentioned), and this means that trying to use get_feature_names_out from any of the following transformers will fail for the same reason:

• Drop
• DropUninformative
• ToCategorical
• ToDatetime
• CleanCategories
• CleanNullStrings
• _BasePeriodicEncoder
• DropIfTooManyNulls
• ToStr
• PassThrough
• ToFloat32

My suggestion is to add back get_feature_names_out to SingleColumnTransformer, and make it return the name of the column by default. Any transformer that needs to do something more complicated should implement the proper function anyway.

The alternative is doing it manually for all the methods.

I also want to add somewhere (not sure where) a test that checks this automatically for all estimators so the problem should not pop up anymore.

[GaelVaroquaux]

My suggestion is to add back get_feature_names_out to SingleColumnTransformer

I like this suggestion. Thanks!!

[auguste-probabl]

I also want to add somewhere (not sure where) a test that checks this automatically for all estimators so the problem should not pop up anymore.

I guess you mean "transformers" rather than "estimators"? If so you might use a runtime-checkable Protocol for that