Since this is a file for internal use, does this need to be in the changelog?

Not sure about the Azure CI black job failure. Is sklearn/externals exempted from black, but not sklearn/utils?

Not sure about the Azure CI black job failure. Is sklearn/externals exempted from black, but not sklearn/utils?

Yea. sklearn/externals is for vendored projects/files so we try not to change the source. sklearn/utils is maintained by us, which follows black.

@thomasjpfan Makes sense. I ran flake8 and black on the file, but some Azure CI jobs still fail.

The tests are running doctests, which are failing. For example bytescale uses from scipy.misc import bytescale which does not exist.

For future reference, you can explore these errors by clicking on details in the failed check:

and navigating to the error, which is here

Thoughts on this PR

For this PR, I would consider wrapping Pillow lightly, since we only use imread and imresize. imsave is only used for testing in sklearn/datasets/tests/test_lfw.py, so we can use pillow directly in the testing code.

There are two public functions that use functions in _pilutil.py:

1. sklearn.datasets.load_sample_images: Optionally import pillow and error if its not installed. If Pillow is installed use np.asarray(Image.open(...)) instead of imread.

2. sklearn.datasets._lfw._load_imgs method: Optionally import pillow and error if its not installed. If Pillow is installed load images as a Pillow image object using Image.open, use the Image.resize api to resize and then convert to a numpy array.

(This way, we do not need to maintain a big sklearn.utils._pilutil file.)

I had seen that page, but it doesn't make sense:

• It's too verbose.
• It looks like the problem is with the XFAIL messages.

However, now that you've pointed me to these doctest messages, I'll try fixing them:

__________________ [doctest] sklearn.utils._pilutil.bytescale __________________
[gw0] linux -- Python 3.9.7 /usr/share/miniconda/envs/testvenv/bin/python
108 
109     Returns
110     -------
111     img_array : uint8 ndarray
112         The byte-scaled array.
113 
114     Examples
115     --------
116     >>> import numpy as np
117     >>> from scipy.misc import bytescale
UNEXPECTED EXCEPTION: ImportError("cannot import name 'bytescale' from 'scipy.misc' (/usr/share/miniconda/envs/testvenv/lib/python3.9/site-packages/scipy/misc/__init__.py)")
Traceback (most recent call last):
  File "/usr/share/miniconda/envs/testvenv/lib/python3.9/doctest.py", line 1336, in __run
    exec(compile(example.source, filename, "single",
  File "<doctest sklearn.utils._pilutil.bytescale[1]>", line 1, in <module>
ImportError: cannot import name 'bytescale' from 'scipy.misc' (/usr/share/miniconda/envs/testvenv/lib/python3.9/site-packages/scipy/misc/__init__.py)
/home/vsts/work/1/s/sklearn/utils/_pilutil.py:117: UnexpectedException
___________________ [doctest] sklearn.utils._pilutil.imsave ____________________
[gw0] linux -- Python 3.9.7 /usr/share/miniconda/envs/testvenv/bin/python
243         Image format. If omitted, the format to use is determined from the
244         file name extension. If a file object was used instead of a file name,
245         this parameter should always be used.
246 
247     Examples
248     --------
249     Construct an array of gradient intensity values and save to file:
250 
251     >>> import numpy as np
252     >>> from scipy.misc import imsave
UNEXPECTED EXCEPTION: ImportError("cannot import name 'imsave' from 'scipy.misc' (/usr/share/miniconda/envs/testvenv/lib/python3.9/site-packages/scipy/misc/__init__.py)")
Traceback (most recent call last):
  File "/usr/share/miniconda/envs/testvenv/lib/python3.9/doctest.py", line 1336, in __run
    exec(compile(example.source, filename, "single",
  File "<doctest sklearn.utils._pilutil.imsave[1]>", line 1, in <module>
ImportError: cannot import name 'imsave' from 'scipy.misc' (/usr/share/miniconda/envs/testvenv/lib/python3.9/site-packages/scipy/misc/__init__.py)
/home/vsts/work/1/s/sklearn/utils/_pilutil.py:252: UnexpectedException

It looks like doctests are OK after my last commit. Please let me know if anything else is needed.

About wrapping Pillow lightly: nothing against it, but this change feels more involved than the current PR. I would rather defer such a change to another PR – provided I can find the time.

After looking at this again, I prefer removing the file all together. I opened #22743 as an alternative, which removes _pilutil.

I would also lean towards the alternative implemented in #22743.

Closing since #22743 was merged.