✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: cb2f28b. Link to the linter CI: here}

Thanks for the PR. Could you please add a non regression test based on the reproducer provided in #29416 and make it run on all solvers with a small value of tol such as tol=1e-12? It's possible that strict equivalence might be hard to reach for stochastic solvers such as SAG/SAGA but I am not sure.

Note: in the test_logistic_regression_sample_weight sag and saga are left out as they systematically fail

@snath-xoc could you please push a commit that triggers running the updated test in this PR for all admissible values of global_random_seed? Instructions are available in this information issue: #28959.

EDIT: before testing on the CI, you should test locally with commands like the following:

SKLEARN_TESTS_GLOBAL_RANDOM_SEED="all" pytest -k test_logistic_regression_sample_weights sklearn/linear_model/tests/test_logistic.py -vlx

at the moment, the test fails for many seeds but this can be fixed as explained in the review comments below.

I find this test too long to follow. You might want to decouple the part that tests equivalence between integer sample weights and their repeated counter part from the part of the test that checks equivalence between class weights and sample weights into two independent tests.

To actually trigger [all random seeds] tests you need to include the list of test function names after the [all random seeds] flag in the commit message as explained in #28959.

Note for later: the LogisticRegressionCV class has the following tag to skip the common test:

    def _more_tags(self):
        return {
            "_xfail_checks": {
                "check_sample_weights_invariance": (
                    "zero sample_weight is not equivalent to removing samples"
                ),
            }
        }

However check_sample_weights_invariance currently does not work well with kind="zeros" in the presence of a cv constructor parameter. This common test should be updated to drop samples without moving the CV boundaries for the remaining training points.

This should better be done in a dedicated PR though.

Merged! Thanks very much @snath-xoc!