We need to bump up the dependency of scipy to 1.5 or 1.6 in the _min_dependencies.py file first (and regenerate the CI config with build_tools/update_environments_and_lock_files.py).

Currently, the min-dependencies doc build fails with:

Traceback (most recent call last):
  File "/home/runner/mambaforge/envs/testenv/lib/python3.8/site-packages/sphinx/cmd/build.py", line 280, in build_main
    app.build(args.force_all, filenames)
  File "/home/runner/mambaforge/envs/testenv/lib/python3.8/site-packages/sphinx/application.py", line 387, in build
    self.events.emit('build-finished', None)
  File "/home/runner/mambaforge/envs/testenv/lib/python3.8/site-packages/sphinx/events.py", line 101, in emit
    results.append(listener.handler(self.app, *args))
  File "/home/runner/mambaforge/envs/testenv/lib/python3.8/site-packages/sphinx_gallery/gen_gallery.py", line 640, in summarize_failing_examples
    raise ExtensionError(
sphinx.errors.ExtensionError: Here is a summary of the problems encountered when running the examples
Unexpected failing examples:
/home/runner/work/scikit-learn/scikit-learn/examples/gaussian_process/plot_compare_gpr_krr.py failed leaving traceback:
Traceback (most recent call last):
  File "/home/runner/mambaforge/envs/testenv/lib/python3.8/site-packages/sphinx_gallery/gen_gallery.py", line 159, in call_memory
    return 0., func()
  File "/home/runner/mambaforge/envs/testenv/lib/python3.8/site-packages/sphinx_gallery/gen_rst.py", line 466, in __call__
    exec(self.code, self.fake_main.__dict__)
  File "/home/runner/work/scikit-learn/scikit-learn/examples/gaussian_process/plot_compare_gpr_krr.py", line 180, in <module>
    from scipy.stats import loguniform
ImportError: cannot import name 'loguniform' from 'scipy.stats' (/home/runner/mambaforge/envs/testenv/lib/python3.8/site-packages/scipy/stats/__init__.py)

As I tried to indicate in the related issue, we tend to do bump minimum requirements and update sklearn.utils.fixes close to the release which will roughly happen in November (we try to do a release every 6 months) ...

Basically I created the issue more as a reminder to ourselves than anything else ...

As I tried to indicate in the related issue, we tend to do bump minimum requirements and update sklearn.utils.fixes close to the release which will roughly happen in November (we try to do a release every 6 months) ...

Basically I created the issue more as a reminder to ourselves than anything else ...

Maybe we can leave this PR as it is for now and revisit when close to the next release :)

Maybe we can leave this PR as it is for now and revisit when close to the next release :)

Sounds good to me!

Note for ourselves in a few months, there are still instances of sklearn.utils.fixes.loguniform left in this PR branch.

It should be removed from the doc completely. Either we remove loguniform from sklearn/utils/fixes.py or we deprecate it and remove it in two releases. Not sure the status of sklearn.utils.fixes.py with regards to the two releases deprecation cycle policy.

❯ git grep -P 'fixes.+loguniform' | grep -v whats_new
doc/modules/grid_search.rst::class:`~sklearn.utils.fixes.loguniform`. This is a continuous version of
doc/modules/grid_search.rst:  from sklearn.utils.fixes import loguniform
examples/model_selection/plot_randomized_search.py:from sklearn.utils.fixes import loguniform
sklearn/utils/fixes.py:    >>> from sklearn.utils.fixes import loguniform

Not sure the status of sklearn.utils.fixes.py with regards to the two releases deprecation cycle policy.

Historically, we usually still deprecate over two versions for anything in utils.

Linking issue: I proposed a faster deprecation cycle for utils #15801

Looking at another PR: #22894 In that case we removed sparse_lsqr without deprecation.

@MaxwellLZH do you want to reboot this PR now that other deprecation related PRs are getting merged for the upcoming release?

I'll make a new PR for this

Moving it to the 1.3 milestone since we did not bump scipy min version in the end.

done in #24665