Follow-up fix for check_classifiers_train on prescott OpenBLAS

[ogrisel]

Addresses the problem seen in: #21702 (comment)

[ogrisel]

We could improve create_memmap_backed_data to do that automatically but I was too lazy. The real fix will be to fix joblib anyways.

[ogrisel]

BTW I test it on an Intel machine and it works as expected:

$ OPENBLAS_CORETYPE=Prescott pytest -v sklearn/tests/test_common.py -k "check_classifiers_train and svc"                                                                                              
================================================= test session starts =================================================
platform linux -- Python 3.10.0, pytest-6.2.5, py-1.11.0, pluggy-1.0.0 -- /storage/store/work/ogrisel/mambaforge/envs/d
ev/bin/python3.10                                                                                                      
cachedir: .pytest_cache
rootdir: /home/parietal/ogrisel/code/scikit-learn, configfile: setup.cfg
collected 8785 items / 8776 deselected / 9 selected                                                                   

sklearn/tests/test_common.py::test_estimators[LinearSVC()-check_classifiers_train] PASSED                       [ 11%]
sklearn/tests/test_common.py::test_estimators[LinearSVC()-check_classifiers_train(readonly_memmap=True)] PASSED [ 22%]
sklearn/tests/test_common.py::test_estimators[LinearSVC()-check_classifiers_train(readonly_memmap=True,X_dtype=float32$] PASSED [ 33%]
sklearn/tests/test_common.py::test_estimators[NuSVC()-check_classifiers_train] PASSED                           [ 44%]
sklearn/tests/test_common.py::test_estimators[NuSVC()-check_classifiers_train(readonly_memmap=True)] PASSED     [ 55%]
sklearn/tests/test_common.py::test_estimators[NuSVC()-check_classifiers_train(readonly_memmap=True,X_dtype=float32)] P$SSED [ 66%]
sklearn/tests/test_common.py::test_estimators[SVC()-check_classifiers_train] PASSED                             [ 77%]
sklearn/tests/test_common.py::test_estimators[SVC()-check_classifiers_train(readonly_memmap=True)] PASSED       [ 88%]
sklearn/tests/test_common.py::test_estimators[SVC()-check_classifiers_train(readonly_memmap=True,X_dtype=float32)] PAS$ED [100%]

[adrinjalali]

Gonna trust you on that this is now working :D

[jjerphan]

In this case, could we scope some time in the coming days to focus on joblib then?

Edit: I saw the notification for review and wanted to approve and merge but had no machine at hand to.

[ogrisel]

In this case, could we scope some time in the coming days to focus on joblib then?

Sure but I have no idea how to fix the problem in joblib. Will have to sleep on it.