⚠️ CI failed on Linux_Nightly.pylatest_pip_scipy_dev ⚠️

[scikit-learn-bot]

CI is still failing on Linux_Nightly.pylatest_pip_scipy_dev (Aug 16, 2022)

• test_grid_search_failing_classifier
• test_searchcv_raise_warning_with_non_finite_score[GridSearchCV-specialized_params0-False]
• test_searchcv_raise_warning_with_non_finite_score[GridSearchCV-specialized_params0-True]
• test_searchcv_raise_warning_with_non_finite_score[RandomizedSearchCV-specialized_params1-False]
• test_searchcv_raise_warning_with_non_finite_score[RandomizedSearchCV-specialized_params1-True]
• test_multi_output_predict_proba

[Micky774]

This will also require a change to utisl.tests.test_extmath::test_uniform_weights

[lesteve]

Hmmm for some reason we are stuck with an old version of the Scipy development wheel uploaded one month ago: https://anaconda.org/scipy-wheels-nightly/scipy/files?version=1.10.0.dev0%2B0.df3fe4e&page=1 (you can see it in one of the log "Successfully installed numpy-1.24.0.dev0+569.gf1e50253a pandas-1.5.0.dev0+1172.ga7c5773d19 scipy-1.10.0.dev0+0.df3fe4e")

There have been plenty of more recent scipy wheels uploaded recently:
https://anaconda.org/scipy-wheels-nightly/scipy/files?version=&page=1

My guess is that the naming conventions is always scipy-1.10.0.dev0+0.<something>.whl instead of being scipy-1.10.0.dev0+<increasing-number>.<something>.whl (for OSX wheels the increasing number is there).

Compare this to numpy for example: https://anaconda.org/scipy-wheels-nightly/numpy/files?version=&page=1

I'll open a scipy issue about this. Edit: done in scipy/scipy#16702

[scikit-learn-bot]

CI is no longer failing! ✅

Successful run on Aug 17, 2022

[ogrisel]

Thank you @lesteve for the final push!

[lesteve]

Nice to see this one green!