Test suite does not build with scikit-learn 1.2.1

[tillea]

Bug Description
Debian has moved to scikit-learn 1.2.1. When I try to build the q2-feature-classifier package against this I get test suite errors.

Steps to reproduce the behavior
Try building with scikit-learn 1.2.1. In the Debian CI you can find a build log which ends with:

Name                                                       Stmts   Miss  Cover
------------------------------------------------------------------------------
q2_feature_classifier/__init__.py                              8      0   100%
q2_feature_classifier/_blast.py                               54      0   100%
q2_feature_classifier/_consensus_assignment.py                43      0   100%
q2_feature_classifier/_cutter.py                             100      2    98%
q2_feature_classifier/_skl.py                                 58     36    38%
q2_feature_classifier/_taxonomic_classifier.py                71     38    46%
q2_feature_classifier/_version.py                              4      0   100%
q2_feature_classifier/_vsearch.py                             91     26    71%
q2_feature_classifier/classifier.py                          195     88    55%
q2_feature_classifier/custom.py                               64     40    38%
q2_feature_classifier/plugin_setup.py                          4      0   100%
q2_feature_classifier/tests/__init__.py                       19      6    68%
q2_feature_classifier/tests/test_classifier.py               147    114    22%
q2_feature_classifier/tests/test_consensus_assignment.py     178     15    92%
q2_feature_classifier/tests/test_custom.py                    40     14    65%
q2_feature_classifier/tests/test_cutter.py                    67      0   100%
q2_feature_classifier/tests/test_taxonomic_classifier.py      77     43    44%
------------------------------------------------------------------------------
TOTAL                                                       1220    422    65%
=========================== short test summary info ============================
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_autotune_reads_per_batch - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range'      parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_autotune_reads_per_batch_ceil - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range' parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_autotune_reads_per_batch_disable_if_single_job - sklearn.utils._param_validation.InvalidParameterError:  The 'ngram_range' parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_autotune_reads_per_batch_more_jobs_than_reads - sklearn.utils._param_validation.InvalidParameterError:   The 'ngram_range' parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_autotune_reads_per_batch_zero_jobs - sklearn.utils._param_validation.InvalidParameterError: The          'ngram_range' parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_class_weight - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range' parameter of     HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_classify - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range' parameter of         HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_fit_classifier - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range' parameter of   HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_fit_specific_classifiers - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range'      parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_pipeline_serialisation - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range'        parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_populate_class_weight - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range'         parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_classifier.py::ClassifierTests::test_unassigned_taxa - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range' parameter of  HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_consensus_assignment.py::HybridClassiferTests::test_classify_hybrid_vsearch_sklearn_all_exact_match - sklearn.utils._param_validation.        InvalidParameterError: The 'ngram_range' parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_consensus_assignment.py::HybridClassiferTests::test_classify_hybrid_vsearch_sklearn_mixed_query - sklearn.utils._param_validation.            InvalidParameterError: The 'ngram_range' parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_custom.py::CustomTests::test_chunked_hashing_vectorizer - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range' parameter  of HashingVectorizer must be an instance of 'tuple'. Got [8, 8] instead.
FAILED q2_feature_classifier/tests/test_custom.py::CustomTests::test_low_memory_multinomial_nb - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range' parameter   of HashingVectorizer must be an instance of 'tuple'. Got [8, 8] instead.
FAILED q2_feature_classifier/tests/test_taxonomic_classifier.py::TestFormats::test_taxonomic_classifier_dir_fmt - sklearn.utils._param_validation.InvalidParameterError: The          'ngram_range' parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_taxonomic_classifier.py::TestTransformers::test_old_dirfmt - sklearn.utils._param_validation.InvalidParameterError: The 'ngram_range'         parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_taxonomic_classifier.py::TestTransformers::test_old_sklearn_version - sklearn.utils._param_validation.InvalidParameterError: The              'ngram_range' parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_taxonomic_classifier.py::TestTransformers::test_taxo_class_dir_fmt_to_taxo_class_result - sklearn.utils._param_validation.                    InvalidParameterError: The 'ngram_range' parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
FAILED q2_feature_classifier/tests/test_taxonomic_classifier.py::TestTransformers::test_taxo_class_result_to_taxo_class_dir_fmt - sklearn.utils._param_validation.                    InvalidParameterError: The 'ngram_range' parameter of HashingVectorizer must be an instance of 'tuple'. Got [7, 7] instead.
================= 21 failed, 40 passed, 138 warnings in 51.37s =================

All these errors contain sklearn.*

Expected behavior
Tests are working also with latest version of sklearn.

Computation Environment

• OS: Debian unstable
• QIIME 2 Release: 2022.11.1

Kind regards, Andreas.

[tillea]

Thanks to Étienne Mollier there is a patch for this issue.

[ebolyen]

Thanks @tillea! We're not in a good position to update scikit-learn right this moment, but we're going to try to get scikit-learn updated by our April release (I've added this issue to that project).