KBinsDiscretizer's get_feature_name_out() does not work if encode != 'onehot'

[rosscleung]

Describe the bug

If the encode = 'onehot', you can get the names out as intended:

from sklearn.preprocessing import KBinsDiscretizer
import pandas as pd
kb = KBinsDiscretizer(n_bins=5, encode='onehot')
kb.fit_transform(pd.DataFrame(range(100),columns=['one']))
kb.get_feature_names_out()

output:
array(['one_0.0', 'one_1.0', 'one_2.0', 'one_3.0', 'one_4.0'],
dtype=object)

If the encode is NOT onehot:

from sklearn.preprocessing import KBinsDiscretizer
import pandas as pd
kb = KBinsDiscretizer(n_bins=5, encode='ordinal')
kb.fit_transform(pd.DataFrame(range(100),columns=['one']))
kb.get_feature_names_out()

output:
Traceback (most recent call last):
File "", line 1, in
File "/home/smds/miniconda3/envs/ross_docker_py37/lib/python3.7/site-packages/sklearn/preprocessing/_discretization.py", line 396, in get_feature_names_out
return self._encoder.get_feature_names_out(input_features)
AttributeError: 'KBinsDiscretizer' object has no attribute '_encoder'

I have hunted down the source of the bug. It's because if the encode != 'onehot', the attribute _encoder won't even be established. It's only established if encode == 'onehot':
https://github.com/scikit-learn/scikit-learn/blob/37ac6788c/sklearn/preprocessing/_discretization.py#L240-L248

You can see that when you try to call get_feature_names_out(), it's looking for the self._encoder, which wouldn't have been created in the first place if you don't use encode='onehot':
https://github.com/scikit-learn/scikit-learn/blob/37ac6788c/sklearn/preprocessing/_discretization.py#L376-L396

Steps/Code to Reproduce

from sklearn.preprocessing import KBinsDiscretizer
import pandas as pd
kb = KBinsDiscretizer(n_bins=5, encode='ordinal')
kb.fit_transform(pd.DataFrame(range(100),columns=['one']))
kb.get_feature_names_out()

Expected Results

The name of the passed in feature, in this case it should be 'one'.

Actual Results

Traceback (most recent call last):
File "", line 1, in
File "/home/smds/miniconda3/envs/ross_docker_py37/lib/python3.7/site-packages/sklearn/preprocessing/_discretization.py", line 396, in get_feature_names_out
return self._encoder.get_feature_names_out(input_features)
AttributeError: 'KBinsDiscretizer' object has no attribute '_encoder'

Versions

import sklearn; sklearn.show_versions()

System:
    python: 3.7.10 (default, Jun  4 2021, 14:48:32)  [GCC 7.5.0]
executable: /home/smds/miniconda3/envs/ross_docker_py37/bin/python
   machine: Linux-4.4.0-116-generic-x86_64-with-debian-buster-sid

Python dependencies:
          pip: 21.2.2
   setuptools: 58.0.4
      sklearn: 1.0.2
        numpy: 1.20.3
        scipy: 1.7.1
       Cython: 0.29.24
       pandas: 1.3.3
   matplotlib: 3.4.2
       joblib: 1.0.1
threadpoolctl: 2.2.0

Built with OpenMP: True

[thomasjpfan]

Thank you for opening the issue! This issue was recently fixed in #22735 and will be included in the next version of scikit-learn.