scikit-learn: Failing CI for check_sample_weight_equivalence_on_dense_data with LinearRegerssion on debian_32bit

Here is the last scheduled run (from 1 day ago) that passed:

https://dev.azure.com/scikit-learn/scikit-learn/_build/results?buildId=75127&view=logs&j=86340c1f-3d76-5202-0821-7817a0f52092&t=a73eff7b-829e-5a65-7648-23ff8e83ea2d

and here is a more recent run that failed (all CI is failing today):

https://dev.azure.com/scikit-learn/scikit-learn/_build/results?buildId=75179&view=logs&j=86340c1f-3d76-5202-0821-7817a0f52092&t=a73eff7b-829e-5a65-7648-23ff8e83ea2d

FAILED tests/test_common.py::test_estimators[LinearRegression(positive=True)-check_sample_weight_equivalence_on_dense_data] - AssertionError: 
FAILED utils/tests/test_estimator_checks.py::test_check_estimator_clones - AssertionError: 
= 2 failed, 34214 passed, 4182 skipped, 174 xfailed, 66 xpassed, 4252 warnings in 1489.21s (0:24:49) =

Full failure log:

2025-03-28T06:36:32.3433619Z =================================== FAILURES ===================================
2025-03-28T06:36:32.3434358Z _ test_estimators[LinearRegression(positive=True)-check_sample_weight_equivalence_on_dense_data] _
2025-03-28T06:36:32.3434613Z 
2025-03-28T06:36:32.3434838Z estimator = LinearRegression(positive=True)
2025-03-28T06:36:32.3435117Z check = functools.partial(<function check_sample_weight_equivalence_on_dense_data at 0xd8591e88>, 'LinearRegression')
2025-03-28T06:36:32.3435705Z request = <FixtureRequest for <Function test_estimators[LinearRegression(positive=True)-check_sample_weight_equivalence_on_dense_data]>>
2025-03-28T06:36:32.3435878Z 
2025-03-28T06:36:32.3436047Z     @parametrize_with_checks(
2025-03-28T06:36:32.3436274Z         list(_tested_estimators()), expected_failed_checks=_get_expected_failed_checks
2025-03-28T06:36:32.3436498Z     )
2025-03-28T06:36:32.3436684Z     def test_estimators(estimator, check, request):
2025-03-28T06:36:32.3436909Z         # Common tests for estimator instances
2025-03-28T06:36:32.3437101Z         with ignore_warnings(
2025-03-28T06:36:32.3437316Z             category=(FutureWarning, ConvergenceWarning, UserWarning, LinAlgWarning)
2025-03-28T06:36:32.3437521Z         ):
2025-03-28T06:36:32.3437708Z >           check(estimator)
2025-03-28T06:36:32.3437793Z 
2025-03-28T06:36:32.3438019Z check      = functools.partial(<function check_sample_weight_equivalence_on_dense_data at 0xd8591e88>, 'LinearRegression')
2025-03-28T06:36:32.3438293Z estimator  = LinearRegression(positive=True)
2025-03-28T06:36:32.3438559Z request    = <FixtureRequest for <Function test_estimators[LinearRegression(positive=True)-check_sample_weight_equivalence_on_dense_data]>>
2025-03-28T06:36:32.3438707Z 
2025-03-28T06:36:32.3439155Z /io/sklearn/tests/test_common.py:122: 
2025-03-28T06:36:32.3439405Z _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
2025-03-28T06:36:32.3439768Z /io/sklearn/utils/estimator_checks.py:1570: in check_sample_weight_equivalence_on_dense_data
2025-03-28T06:36:32.3440046Z     _check_sample_weight_equivalence(name, estimator_orig, sparse_container=None)
2025-03-28T06:36:32.3440296Z         estimator_orig = LinearRegression(positive=True)
2025-03-28T06:36:32.3440498Z         name       = 'LinearRegression'
2025-03-28T06:36:32.3440774Z /io/sklearn/utils/_testing.py:145: in wrapper
2025-03-28T06:36:32.3440988Z     return fn(*args, **kwargs)
2025-03-28T06:36:32.3441218Z         args       = ('LinearRegression', LinearRegression(positive=True))
2025-03-28T06:36:32.3441452Z         fn         = <function _check_sample_weight_equivalence at 0xd8591de8>
2025-03-28T06:36:32.3441744Z         kwargs     = {'sparse_container': None}
2025-03-28T06:36:32.3441952Z         self       = _IgnoreWarnings(record=True)
2025-03-28T06:36:32.3442307Z /io/sklearn/utils/estimator_checks.py:1566: in _check_sample_weight_equivalence
2025-03-28T06:36:32.3442654Z     assert_allclose_dense_sparse(X_pred1, X_pred2, err_msg=err_msg)
2025-03-28T06:36:32.3442905Z         X          = array([[0.37454012, 0.95071431, 0.73199394, 0.59865848, 0.15601864,
2025-03-28T06:36:32.3443158Z         0.15599452, 0.05808361, 0.86617615, 0.6011..., 0.98663958, 0.3742708 , 0.37064215, 0.81279957,
2025-03-28T06:36:32.3443423Z         0.94724858, 0.98600106, 0.75337819, 0.37625959, 0.08350072]])
2025-03-28T06:36:32.3443662Z         X_pred1    = array([ 8.88178420e-16,  1.00000000e+00,  2.00000000e+00,  1.18549798e+00,
2025-03-28T06:36:32.3443929Z         4.06241761e+00,  1.00000000e+00,  2...5767e+00,  2.00000000e+00, -2.79936287e-02, -8.90642835e-01,
2025-03-28T06:36:32.3444177Z        -8.00809991e-01,  1.00000000e+00,  1.00000000e+00])
2025-03-28T06:36:32.3444412Z         X_pred2    = array([0.        , 1.        , 2.        , 0.94186541, 1.72670876,
2025-03-28T06:36:32.3444772Z        1.        , 2.        , 2.        , 1.8723887 , 2.        ,
2025-03-28T06:36:32.3444987Z        1.50877777, 0.76107365, 1.70933257, 1.        , 1.        ])
2025-03-28T06:36:32.3445218Z         X_repeated = array([[0.37454012, 0.95071431, 0.73199394, 0.59865848, 0.15601864,
2025-03-28T06:36:32.3445494Z         0.15599452, 0.05808361, 0.86617615, 0.6011..., 0.98663958, 0.3742708 , 0.37064215, 0.81279957,
2025-03-28T06:36:32.3445740Z         0.94724858, 0.98600106, 0.75337819, 0.37625959, 0.08350072]])
2025-03-28T06:36:32.3445976Z         X_weighted = array([[0.60754485, 0.17052412, 0.06505159, 0.94888554, 0.96563203,
2025-03-28T06:36:32.3446499Z         0.80839735, 0.30461377, 0.09767211, 0.6842..., 0.69673717, 0.62894285, 0.87747201, 0.73507104,
2025-03-28T06:36:32.3446765Z         0.80348093, 0.28203457, 0.17743954, 0.75061475, 0.80683474]])
2025-03-28T06:36:32.3447241Z         err_msg    = 'Comparing the output of LinearRegression.predict revealed that fitting with `sample_weight` is not equivalent to fitting with removed or repeated data points.'
2025-03-28T06:36:32.3447553Z         estimator_orig = LinearRegression(positive=True)
2025-03-28T06:36:32.3448930Z         estimator_repeated = LinearRegression(positive=True)
2025-03-28T06:36:32.3449323Z         estimator_weighted = LinearRegression(positive=True)
2025-03-28T06:36:32.3449572Z         method     = 'predict'
2025-03-28T06:36:32.3449847Z         n_samples  = 15
2025-03-28T06:36:32.3450076Z         name       = 'LinearRegression'
2025-03-28T06:36:32.3450329Z         rng        = RandomState(MT19937) at 0xCB6EECE8
2025-03-28T06:36:32.3450580Z         sparse_container = None
2025-03-28T06:36:32.3450850Z         sw         = array([3, 4, 0, 3, 1, 0, 4, 4, 0, 3, 0, 0, 3, 2, 0])
2025-03-28T06:36:32.3451124Z         y          = array([0, 1, 2, 2, 1, 1, 2, 2, 1, 2, 0, 0, 1, 1, 1])
2025-03-28T06:36:32.3451409Z         y_repeated = array([0, 0, 0, 1, 1, 1, 2, 2, 2, 2, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2025-03-28T06:36:32.3451823Z        1, 1, 1, 1, 1])
2025-03-28T06:36:32.3452390Z         y_weighted = array([1, 2, 1, 2, 1, 1, 2, 1, 0, 2, 0, 2, 0, 1, 1])
2025-03-28T06:36:32.3467337Z /io/sklearn/utils/_testing.py:283: in assert_allclose_dense_sparse
2025-03-28T06:36:32.3468147Z     assert_allclose(x, y, rtol=rtol, atol=atol, err_msg=err_msg)
2025-03-28T06:36:32.3468478Z         atol       = 1e-09
2025-03-28T06:36:32.3468965Z         err_msg    = 'Comparing the output of LinearRegression.predict revealed that fitting with `sample_weight` is not equivalent to fitting with removed or repeated data points.'
2025-03-28T06:36:32.3469366Z         rtol       = 1e-07
2025-03-28T06:36:32.3469659Z         x          = array([ 8.88178420e-16,  1.00000000e+00,  2.00000000e+00,  1.18549798e+00,
2025-03-28T06:36:32.3470005Z         4.06241761e+00,  1.00000000e+00,  2...5767e+00,  2.00000000e+00, -2.79936287e-02, -8.90642835e-01,
2025-03-28T06:36:32.3470354Z        -8.00809991e-01,  1.00000000e+00,  1.00000000e+00])
2025-03-28T06:36:32.3470648Z         y          = array([0.        , 1.        , 2.        , 0.94186541, 1.72670876,
2025-03-28T06:36:32.3470947Z        1.        , 2.        , 2.        , 1.8723887 , 2.        ,
2025-03-28T06:36:32.3471226Z        1.50877777, 0.76107365, 1.70933257, 1.        , 1.        ])
2025-03-28T06:36:32.3471533Z _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
2025-03-28T06:36:32.3471702Z 
2025-03-28T06:36:32.3472165Z actual = array([ 8.88178420e-16,  1.00000000e+00,  2.00000000e+00,  1.18549798e+00,
2025-03-28T06:36:32.3472656Z         4.06241761e+00,  1.00000000e+00,  2...5767e+00,  2.00000000e+00, -2.79936287e-02, -8.90642835e-01,
2025-03-28T06:36:32.3473239Z        -8.00809991e-01,  1.00000000e+00,  1.00000000e+00])
2025-03-28T06:36:32.3473561Z desired = array([0.        , 1.        , 2.        , 0.94186541, 1.72670876,
2025-03-28T06:36:32.3473863Z        1.        , 2.        , 2.        , 1.8723887 , 2.        ,
2025-03-28T06:36:32.3474355Z        1.50877777, 0.76107365, 1.70933257, 1.        , 1.        ])
2025-03-28T06:36:32.3474896Z rtol = 1e-07, atol = 1e-09, equal_nan = True
2025-03-28T06:36:32.3475293Z err_msg = 'Comparing the output of LinearRegression.predict revealed that fitting with `sample_weight` is not equivalent to fitting with removed or repeated data points.'
2025-03-28T06:36:32.3475668Z verbose = True
2025-03-28T06:36:32.3475847Z 
2025-03-28T06:36:32.3476109Z     def assert_allclose(
2025-03-28T06:36:32.3476425Z         actual, desired, rtol=None, atol=0.0, equal_nan=True, err_msg="", verbose=True
2025-03-28T06:36:32.3476851Z     ):
2025-03-28T06:36:32.3477457Z         """dtype-aware variant of numpy.testing.assert_allclose
2025-03-28T06:36:32.3477743Z     
2025-03-28T06:36:32.3478184Z         This variant introspects the least precise floating point dtype
2025-03-28T06:36:32.3478502Z         in the input argument and automatically sets the relative tolerance
2025-03-28T06:36:32.3478839Z         parameter to 1e-4 float32 and use 1e-7 otherwise (typically float64
2025-03-28T06:36:32.3479134Z         in scikit-learn).
2025-03-28T06:36:32.3479381Z     
2025-03-28T06:36:32.3479837Z         `atol` is always left to 0. by default. It should be adjusted manually
2025-03-28T06:36:32.3480176Z         to an assertion-specific value in case there are null values expected
2025-03-28T06:36:32.3480467Z         in `desired`.
2025-03-28T06:36:32.3480718Z     
2025-03-28T06:36:32.3480995Z         The aggregate tolerance is `atol + rtol * abs(desired)`.
2025-03-28T06:36:32.3481285Z     
2025-03-28T06:36:32.3481704Z         Parameters
2025-03-28T06:36:32.3481965Z         ----------
2025-03-28T06:36:32.3482420Z         actual : array_like
2025-03-28T06:36:32.3483027Z             Array obtained.
2025-03-28T06:36:32.3483323Z         desired : array_like
2025-03-28T06:36:32.3483598Z             Array desired.
2025-03-28T06:36:32.3483881Z         rtol : float, optional, default=None
2025-03-28T06:36:32.3484165Z             Relative tolerance.
2025-03-28T06:36:32.3484462Z             If None, it is set based on the provided arrays' dtypes.
2025-03-28T06:36:32.3484791Z         atol : float, optional, default=0.
2025-03-28T06:36:32.3485072Z             Absolute tolerance.
2025-03-28T06:36:32.3485449Z         equal_nan : bool, optional, default=True
2025-03-28T06:36:32.3485916Z             If True, NaNs will compare equal.
2025-03-28T06:36:32.3486379Z         err_msg : str, optional, default=''
2025-03-28T06:36:32.3486669Z             The error message to be printed in case of failure.
2025-03-28T06:36:32.3486959Z         verbose : bool, optional, default=True
2025-03-28T06:36:32.3487445Z             If True, the conflicting values are appended to the error message.
2025-03-28T06:36:32.3487721Z     
2025-03-28T06:36:32.3487982Z         Raises
2025-03-28T06:36:32.3488229Z         ------
2025-03-28T06:36:32.3488486Z         AssertionError
2025-03-28T06:36:32.3490170Z             If actual and desired are not equal up to specified precision.
2025-03-28T06:36:32.3490710Z     
2025-03-28T06:36:32.3491058Z         See Also
2025-03-28T06:36:32.3491386Z         --------
2025-03-28T06:36:32.3491708Z         numpy.testing.assert_allclose
2025-03-28T06:36:32.3492016Z     
2025-03-28T06:36:32.3492308Z         Examples
2025-03-28T06:36:32.3493161Z         --------
2025-03-28T06:36:32.3493425Z         >>> import numpy as np
2025-03-28T06:36:32.3493712Z         >>> from sklearn.utils._testing import assert_allclose
2025-03-28T06:36:32.3493988Z         >>> x = [1e-5, 1e-3, 1e-1]
2025-03-28T06:36:32.3494270Z         >>> y = np.arccos(np.cos(x))
2025-03-28T06:36:32.3494725Z         >>> assert_allclose(x, y, rtol=1e-5, atol=0)
2025-03-28T06:36:32.3495044Z         >>> a = np.full(shape=10, fill_value=1e-5, dtype=np.float32)
2025-03-28T06:36:32.3495339Z         >>> assert_allclose(a, 1e-5)
2025-03-28T06:36:32.3495595Z         """
2025-03-28T06:36:32.3496358Z         dtypes = []
2025-03-28T06:36:32.3496662Z     
2025-03-28T06:36:32.3496955Z         actual, desired = np.asanyarray(actual), np.asanyarray(desired)
2025-03-28T06:36:32.3497629Z         dtypes = [actual.dtype, desired.dtype]
2025-03-28T06:36:32.3497899Z     
2025-03-28T06:36:32.3498178Z         if rtol is None:
2025-03-28T06:36:32.3498471Z             rtols = [1e-4 if dtype == np.float32 else 1e-7 for dtype in dtypes]
2025-03-28T06:36:32.3498764Z             rtol = max(rtols)
2025-03-28T06:36:32.3499011Z     
2025-03-28T06:36:32.3604475Z >       np_assert_allclose(
2025-03-28T06:36:32.3607071Z             actual,
2025-03-28T06:36:32.3608138Z             desired,
2025-03-28T06:36:32.3608886Z             rtol=rtol,
2025-03-28T06:36:32.3609217Z             atol=atol,
2025-03-28T06:36:32.3625173Z             equal_nan=equal_nan,
2025-03-28T06:36:32.3639658Z             err_msg=err_msg,
2025-03-28T06:36:32.3640151Z             verbose=verbose,
2025-03-28T06:36:32.3640425Z         )
2025-03-28T06:36:32.3640827Z E       AssertionError: 
2025-03-28T06:36:32.3641237Z E       Not equal to tolerance rtol=1e-07, atol=1e-09
2025-03-28T06:36:32.3641778Z E       Comparing the output of LinearRegression.predict revealed that fitting with `sample_weight` is not equivalent to fitting with removed or repeated data points.
2025-03-28T06:36:32.3642255Z E       Mismatched elements: 6 / 15 (40%)
2025-03-28T06:36:32.3642821Z E       Max absolute difference among violations: 2.51014256
2025-03-28T06:36:32.3643306Z E       Max relative difference among violations: 2.17024526
2025-03-28T06:36:32.3643773Z E        ACTUAL: array([ 8.881784e-16,  1.000000e+00,  2.000000e+00,  1.185498e+00,
2025-03-28T06:36:32.3644416Z E               4.062418e+00,  1.000000e+00,  2.000000e+00,  2.000000e+00,
2025-03-28T06:36:32.3644878Z E               4.105658e+00,  2.000000e+00, -2.799363e-02, -8.906428e-01,
2025-03-28T06:36:32.3645289Z E              -8.008100e-01,  1.000000e+00,  1.000000e+00])
2025-03-28T06:36:32.3645729Z E        DESIRED: array([0.      , 1.      , 2.      , 0.941865, 1.726709, 1.      ,
2025-03-28T06:36:32.3646166Z E              2.      , 2.      , 1.872389, 2.      , 1.508778, 0.761074,
2025-03-28T06:36:32.3646542Z E              1.709333, 1.      , 1.      ])
2025-03-28T06:36:32.3646727Z 
2025-03-28T06:36:32.3647013Z actual     = array([ 8.88178420e-16,  1.00000000e+00,  2.00000000e+00,  1.18549798e+00,
2025-03-28T06:36:32.3647353Z         4.06241761e+00,  1.00000000e+00,  2...5767e+00,  2.00000000e+00, -2.79936287e-02, -8.90642835e-01,
2025-03-28T06:36:32.3647675Z        -8.00809991e-01,  1.00000000e+00,  1.00000000e+00])
2025-03-28T06:36:32.3647970Z atol       = 1e-09
2025-03-28T06:36:32.3648247Z desired    = array([0.        , 1.        , 2.        , 0.94186541, 1.72670876,
2025-03-28T06:36:32.3648536Z        1.        , 2.        , 2.        , 1.8723887 , 2.        ,
2025-03-28T06:36:32.3648814Z        1.50877777, 0.76107365, 1.70933257, 1.        , 1.        ])
2025-03-28T06:36:32.3649119Z dtypes     = [dtype('float64'), dtype('float64')]
2025-03-28T06:36:32.3650674Z equal_nan  = True
2025-03-28T06:36:32.3664015Z err_msg    = 'Comparing the output of LinearRegression.predict revealed that fitting with `sample_weight` is not equivalent to fitting with removed or repeated data points.'
2025-03-28T06:36:32.3664453Z rtol       = 1e-07
2025-03-28T06:36:32.3664740Z verbose    = True
2025-03-28T06:36:32.3664894Z 
2025-03-28T06:36:32.3665331Z /io/sklearn/utils/_testing.py:237: AssertionError
2025-03-28T06:36:32.3665827Z _________________________ test_check_estimator_clones __________________________
2025-03-28T06:36:32.3666186Z 
2025-03-28T06:36:32.3666476Z     def test_check_estimator_clones():
2025-03-28T06:36:32.3666804Z         # check that check_estimator doesn't modify the estimator it receives
2025-03-28T06:36:32.3667086Z     
2025-03-28T06:36:32.3667500Z         iris = load_iris()
2025-03-28T06:36:32.3667744Z     
2025-03-28T06:36:32.3667984Z         for Estimator in [
2025-03-28T06:36:32.3668245Z             GaussianMixture,
2025-03-28T06:36:32.3668775Z             LinearRegression,
2025-03-28T06:36:32.3669031Z             SGDClassifier,
2025-03-28T06:36:32.3669267Z             PCA,
2025-03-28T06:36:32.3669511Z             MiniBatchKMeans,
2025-03-28T06:36:32.3669766Z         ]:
2025-03-28T06:36:32.3670005Z             # without fitting
2025-03-28T06:36:32.3670276Z             with ignore_warnings(category=ConvergenceWarning):
2025-03-28T06:36:32.3670550Z                 est = Estimator()
2025-03-28T06:36:32.3670803Z                 set_random_state(est)
2025-03-28T06:36:32.3671081Z                 old_hash = joblib.hash(est)
2025-03-28T06:36:32.3671475Z >               check_estimator(
2025-03-28T06:36:32.3671766Z                     est, expected_failed_checks=_get_expected_failed_checks(est)
2025-03-28T06:36:32.3672037Z                 )
2025-03-28T06:36:32.3672196Z 
2025-03-28T06:36:32.3672459Z Estimator  = <class 'sklearn.linear_model._base.LinearRegression'>
2025-03-28T06:36:32.3672931Z est        = LinearRegression()
2025-03-28T06:36:32.3673224Z iris       = {'data': array([[5.1, 3.5, 1.4, 0.2],
2025-03-28T06:36:32.3673509Z        [4.9, 3. , 1.4, 0.2],
2025-03-28T06:36:32.3673763Z        [4.7, 3.2, 1.3, 0.2],
2025-03-28T06:36:32.3674269Z        [4.6, 3.1, 1.5,... width (cm)', 'petal length (cm)', 'petal width (cm)'], 'filename': 'iris.csv', 'data_module': 'sklearn.datasets.data'}
2025-03-28T06:36:32.3674627Z old_hash   = 'fdcbee8ed611695d1e19a9bdabd615ac'
2025-03-28T06:36:32.3674814Z 
2025-03-28T06:36:32.3675204Z /io/sklearn/utils/tests/test_estimator_checks.py:919: 
2025-03-28T06:36:32.3675540Z _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
2025-03-28T06:36:32.3675945Z /io/sklearn/utils/_param_validation.py:218: in wrapper
2025-03-28T06:36:32.3676248Z     return func(*args, **kwargs)
2025-03-28T06:36:32.3676516Z         args       = (LinearRegression(),)
2025-03-28T06:36:32.3676790Z         func       = <function check_estimator at 0xd8591668>
2025-03-28T06:36:32.3677211Z         func_sig   = <Signature (estimator=None, generate_only=False, *, legacy: 'bool' = True, expected_failed_checks: 'dict[str, str] | N...al['warn'] | None" = 'warn', on_fail: "Literal['raise', 'warn'] | None" = 'raise', callback: 'Callable | None' = None)>
2025-03-28T06:36:32.3677607Z         global_skip_validation = False
2025-03-28T06:36:32.3677881Z         kwargs     = {'expected_failed_checks': {}}
2025-03-28T06:36:32.3678234Z         parameter_constraints = {'callback': [<built-in function callable>, None], 'expected_failed_checks': [<class 'dict'>, None], 'generate_only': ['boolean'], 'legacy': ['boolean'], ...}
2025-03-28T06:36:32.3678659Z         params     = {'callback': None, 'estimator': LinearRegression(), 'expected_failed_checks': {}, 'generate_only': False, ...}
2025-03-28T06:36:32.3678985Z         prefer_skip_nested_validation = False
2025-03-28T06:36:32.3679252Z         to_ignore  = ['self', 'cls']
2025-03-28T06:36:32.3679647Z /io/sklearn/utils/estimator_checks.py:856: in check_estimator
2025-03-28T06:36:32.3679964Z     check(estimator)
2025-03-28T06:36:32.3680221Z         callback   = None
2025-03-28T06:36:32.3680536Z         check      = functools.partial(<function check_sample_weight_equivalence_on_dense_data at 0xd8591e88>, 'LinearRegression')
2025-03-28T06:36:32.3680953Z         check_result = {'check_name': 'check_sample_weight_equivalence_on_dense_data', 'estimator': LinearRegression(), 'exception': None, 'expected_to_fail': False, ...}
2025-03-28T06:36:32.3681402Z         estimator  = LinearRegression(positive=True)
2025-03-28T06:36:32.3681697Z         expected_failed_checks = {}
2025-03-28T06:36:32.3681899Z         generate_only = False
2025-03-28T06:36:32.3682090Z         legacy     = True
2025-03-28T06:36:32.3682266Z         name       = 'LinearRegression'
2025-03-28T06:36:32.3682442Z         on_fail    = 'raise'
2025-03-28T06:36:32.3682732Z         on_skip    = 'warn'
2025-03-28T06:36:32.3682922Z         reason     = 'Check is not expected to fail'
2025-03-28T06:36:32.3683274Z         test_can_fail = False
2025-03-28T06:36:32.3683596Z         test_results = [{'check_name': 'check_estimator_cloneable', 'estimator': LinearRegression(), 'exception': None, 'expected_to_fail': F...k_no_attributes_set_in_init', 'estimator': LinearRegression(), 'exception': None, 'expected_to_fail': False, ...}, ...]
2025-03-28T06:36:32.3684106Z /io/sklearn/utils/estimator_checks.py:1570: in check_sample_weight_equivalence_on_dense_data
2025-03-28T06:36:32.3684400Z     _check_sample_weight_equivalence(name, estimator_orig, sparse_container=None)
2025-03-28T06:36:32.3684731Z         estimator_orig = LinearRegression(positive=True)
2025-03-28T06:36:32.3684930Z         name       = 'LinearRegression'
2025-03-28T06:36:32.3685217Z /io/sklearn/utils/_testing.py:145: in wrapper
2025-03-28T06:36:32.3685439Z     return fn(*args, **kwargs)
2025-03-28T06:36:32.3685650Z         args       = ('LinearRegression', LinearRegression(positive=True))
2025-03-28T06:36:32.3685891Z         fn         = <function _check_sample_weight_equivalence at 0xd8591de8>
2025-03-28T06:36:32.3686105Z         kwargs     = {'sparse_container': None}
2025-03-28T06:36:32.3686316Z         self       = _IgnoreWarnings(record=True)
2025-03-28T06:36:32.3686643Z /io/sklearn/utils/estimator_checks.py:1566: in _check_sample_weight_equivalence
2025-03-28T06:36:32.3686902Z     assert_allclose_dense_sparse(X_pred1, X_pred2, err_msg=err_msg)
2025-03-28T06:36:32.3687138Z         X          = array([[0.37454012, 0.95071431, 0.73199394, 0.59865848, 0.15601864,
2025-03-28T06:36:32.3687418Z         0.15599452, 0.05808361, 0.86617615, 0.6011..., 0.98663958, 0.3742708 , 0.37064215, 0.81279957,
2025-03-28T06:36:32.3687667Z         0.94724858, 0.98600106, 0.75337819, 0.37625959, 0.08350072]])
2025-03-28T06:36:32.3687906Z         X_pred1    = array([ 8.88178420e-16,  1.00000000e+00,  2.00000000e+00,  1.18549798e+00,
2025-03-28T06:36:32.3688173Z         4.06241761e+00,  1.00000000e+00,  2...5767e+00,  2.00000000e+00, -2.79936287e-02, -8.90642835e-01,
2025-03-28T06:36:32.3688439Z        -8.00809991e-01,  1.00000000e+00,  1.00000000e+00])
2025-03-28T06:36:32.3688661Z         X_pred2    = array([0.        , 1.        , 2.        , 0.94186541, 1.72670876,
2025-03-28T06:36:32.3688871Z        1.        , 2.        , 2.        , 1.8723887 , 2.        ,
2025-03-28T06:36:32.3689077Z        1.50877777, 0.76107365, 1.70933257, 1.        , 1.        ])
2025-03-28T06:36:32.3689323Z         X_repeated = array([[0.37454012, 0.95071431, 0.73199394, 0.59865848, 0.15601864,
2025-03-28T06:36:32.3689588Z         0.15599452, 0.05808361, 0.86617615, 0.6011..., 0.98663958, 0.3742708 , 0.37064215, 0.81279957,
2025-03-28T06:36:32.3689834Z         0.94724858, 0.98600106, 0.75337819, 0.37625959, 0.08350072]])
2025-03-28T06:36:32.3690073Z         X_weighted = array([[0.60754485, 0.17052412, 0.06505159, 0.94888554, 0.96563203,
2025-03-28T06:36:32.3690355Z         0.80839735, 0.30461377, 0.09767211, 0.6842..., 0.69673717, 0.62894285, 0.87747201, 0.73507104,
2025-03-28T06:36:32.3690606Z         0.80348093, 0.28203457, 0.17743954, 0.75061475, 0.80683474]])
2025-03-28T06:36:32.3690894Z         err_msg    = 'Comparing the output of LinearRegression.predict revealed that fitting with `sample_weight` is not equivalent to fitting with removed or repeated data points.'
2025-03-28T06:36:32.3691216Z         estimator_orig = LinearRegression(positive=True)
2025-03-28T06:36:32.3691435Z         estimator_repeated = LinearRegression(positive=True)
2025-03-28T06:36:32.3691653Z         estimator_weighted = LinearRegression(positive=True)
2025-03-28T06:36:32.3691853Z         method     = 'predict'
2025-03-28T06:36:32.3692029Z         n_samples  = 15
2025-03-28T06:36:32.3692223Z         name       = 'LinearRegression'
2025-03-28T06:36:32.3692420Z         rng        = RandomState(MT19937) at 0xCD5304A8
2025-03-28T06:36:32.3692720Z         sparse_container = None
2025-03-28T06:36:32.3692923Z         sw         = array([3, 4, 0, 3, 1, 0, 4, 4, 0, 3, 0, 0, 3, 2, 0])
2025-03-28T06:36:32.3693289Z         y          = array([0, 1, 2, 2, 1, 1, 2, 2, 1, 2, 0, 0, 1, 1, 1])
2025-03-28T06:36:32.3693514Z         y_repeated = array([0, 0, 0, 1, 1, 1, 2, 2, 2, 2, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
2025-03-28T06:36:32.4389210Z        1, 1, 1, 1, 1])
2025-03-28T06:36:32.4392227Z         y_weighted = array([1, 2, 1, 2, 1, 1, 2, 1, 0, 2, 0, 2, 0, 1, 1])
2025-03-28T06:36:32.4394205Z /io/sklearn/utils/_testing.py:283: in assert_allclose_dense_sparse
2025-03-28T06:36:32.4395241Z     assert_allclose(x, y, rtol=rtol, atol=atol, err_msg=err_msg)
2025-03-28T06:36:32.4395958Z         atol       = 1e-09
2025-03-28T06:36:32.4396339Z         err_msg    = 'Comparing the output of LinearRegression.predict revealed that fitting with `sample_weight` is not equivalent to fitting with removed or repeated data points.'
2025-03-28T06:36:32.4396710Z         rtol       = 1e-07
2025-03-28T06:36:32.4397020Z         x          = array([ 8.88178420e-16,  1.00000000e+00,  2.00000000e+00,  1.18549798e+00,
2025-03-28T06:36:32.4397420Z         4.06241761e+00,  1.00000000e+00,  2...5767e+00,  2.00000000e+00, -2.79936287e-02, -8.90642835e-01,
2025-03-28T06:36:32.4397769Z        -8.00809991e-01,  1.00000000e+00,  1.00000000e+00])
2025-03-28T06:36:32.4398083Z         y          = array([0.        , 1.        , 2.        , 0.94186541, 1.72670876,
2025-03-28T06:36:32.4398404Z        1.        , 2.        , 2.        , 1.8723887 , 2.        ,
2025-03-28T06:36:32.4398711Z        1.50877777, 0.76107365, 1.70933257, 1.        , 1.        ])
2025-03-28T06:36:32.4399020Z _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
2025-03-28T06:36:32.4399360Z 
2025-03-28T06:36:32.4399684Z actual = array([ 8.88178420e-16,  1.00000000e+00,  2.00000000e+00,  1.18549798e+00,
2025-03-28T06:36:32.4400047Z         4.06241761e+00,  1.00000000e+00,  2...5767e+00,  2.00000000e+00, -2.79936287e-02, -8.90642835e-01,
2025-03-28T06:36:32.4400383Z        -8.00809991e-01,  1.00000000e+00,  1.00000000e+00])
2025-03-28T06:36:32.4400717Z desired = array([0.        , 1.        , 2.        , 0.94186541, 1.72670876,
2025-03-28T06:36:32.4401016Z        1.        , 2.        , 2.        , 1.8723887 , 2.        ,
2025-03-28T06:36:32.4401305Z        1.50877777, 0.76107365, 1.70933257, 1.        , 1.        ])
2025-03-28T06:36:32.4401599Z rtol = 1e-07, atol = 1e-09, equal_nan = True
2025-03-28T06:36:32.4401994Z err_msg = 'Comparing the output of LinearRegression.predict revealed that fitting with `sample_weight` is not equivalent to fitting with removed or repeated data points.'
2025-03-28T06:36:32.4402357Z verbose = True
2025-03-28T06:36:32.4402649Z 
2025-03-28T06:36:32.4403197Z     def assert_allclose(
2025-03-28T06:36:32.4403533Z         actual, desired, rtol=None, atol=0.0, equal_nan=True, err_msg="", verbose=True
2025-03-28T06:36:32.4403835Z     ):
2025-03-28T06:36:32.4404125Z         """dtype-aware variant of numpy.testing.assert_allclose
2025-03-28T06:36:32.4404412Z     
2025-03-28T06:36:32.4404713Z         This variant introspects the least precise floating point dtype
2025-03-28T06:36:32.4405098Z         in the input argument and automatically sets the relative tolerance
2025-03-28T06:36:32.4405430Z         parameter to 1e-4 float32 and use 1e-7 otherwise (typically float64
2025-03-28T06:36:32.4405731Z         in scikit-learn).
2025-03-28T06:36:32.4406156Z     
2025-03-28T06:36:32.4406440Z         `atol` is always left to 0. by default. It should be adjusted manually
2025-03-28T06:36:32.4406762Z         to an assertion-specific value in case there are null values expected
2025-03-28T06:36:32.4407464Z         in `desired`.
2025-03-28T06:36:32.4407766Z     
2025-03-28T06:36:32.4408043Z         The aggregate tolerance is `atol + rtol * abs(desired)`.
2025-03-28T06:36:32.4408313Z     
2025-03-28T06:36:32.4408563Z         Parameters
2025-03-28T06:36:32.4408818Z         ----------
2025-03-28T06:36:32.4409095Z         actual : array_like
2025-03-28T06:36:32.4409358Z             Array obtained.
2025-03-28T06:36:32.4409820Z         desired : array_like
2025-03-28T06:36:32.4410093Z             Array desired.
2025-03-28T06:36:32.4410364Z         rtol : float, optional, default=None
2025-03-28T06:36:32.4410653Z             Relative tolerance.
2025-03-28T06:36:32.4410943Z             If None, it is set based on the provided arrays' dtypes.
2025-03-28T06:36:32.4411236Z         atol : float, optional, default=0.
2025-03-28T06:36:32.4411512Z             Absolute tolerance.
2025-03-28T06:36:32.4411809Z         equal_nan : bool, optional, default=True
2025-03-28T06:36:32.4412101Z             If True, NaNs will compare equal.
2025-03-28T06:36:32.4412731Z         err_msg : str, optional, default=''
2025-03-28T06:36:32.4413109Z             The error message to be printed in case of failure.
2025-03-28T06:36:32.4413412Z         verbose : bool, optional, default=True
2025-03-28T06:36:32.4413734Z             If True, the conflicting values are appended to the error message.
2025-03-28T06:36:32.4414013Z     
2025-03-28T06:36:32.4414256Z         Raises
2025-03-28T06:36:32.4414514Z         ------
2025-03-28T06:36:32.4414771Z         AssertionError
2025-03-28T06:36:32.4415083Z             If actual and desired are not equal up to specified precision.
2025-03-28T06:36:32.4415491Z     
2025-03-28T06:36:32.4415753Z         See Also
2025-03-28T06:36:32.4416185Z         --------
2025-03-28T06:36:32.4416466Z         numpy.testing.assert_allclose
2025-03-28T06:36:32.4416900Z     
2025-03-28T06:36:32.4417154Z         Examples
2025-03-28T06:36:32.4417419Z         --------
2025-03-28T06:36:32.4417685Z         >>> import numpy as np
2025-03-28T06:36:32.4418011Z         >>> from sklearn.utils._testing import assert_allclose
2025-03-28T06:36:32.4418314Z         >>> x = [1e-5, 1e-3, 1e-1]
2025-03-28T06:36:32.4418599Z         >>> y = np.arccos(np.cos(x))
2025-03-28T06:36:32.4418899Z         >>> assert_allclose(x, y, rtol=1e-5, atol=0)
2025-03-28T06:36:32.4419379Z         >>> a = np.full(shape=10, fill_value=1e-5, dtype=np.float32)
2025-03-28T06:36:32.4420007Z         >>> assert_allclose(a, 1e-5)
2025-03-28T06:36:32.4420262Z         """
2025-03-28T06:36:32.4420502Z         dtypes = []
2025-03-28T06:36:32.4420743Z     
2025-03-28T06:36:32.4421010Z         actual, desired = np.asanyarray(actual), np.asanyarray(desired)
2025-03-28T06:36:32.4421324Z         dtypes = [actual.dtype, desired.dtype]
2025-03-28T06:36:32.4421576Z     
2025-03-28T06:36:32.4421816Z         if rtol is None:
2025-03-28T06:36:32.4422102Z             rtols = [1e-4 if dtype == np.float32 else 1e-7 for dtype in dtypes]
2025-03-28T06:36:32.4422404Z             rtol = max(rtols)
2025-03-28T06:36:32.4422867Z     
2025-03-28T06:36:32.4423151Z >       np_assert_allclose(
2025-03-28T06:36:32.4423412Z             actual,
2025-03-28T06:36:32.4423660Z             desired,
2025-03-28T06:36:32.4423929Z             rtol=rtol,
2025-03-28T06:36:32.4424278Z             atol=atol,
2025-03-28T06:36:32.4424545Z             equal_nan=equal_nan,
2025-03-28T06:36:32.4424804Z             err_msg=err_msg,
2025-03-28T06:36:32.4425069Z             verbose=verbose,
2025-03-28T06:36:32.4425335Z         )
2025-03-28T06:36:32.4425695Z E       AssertionError: 
2025-03-28T06:36:32.4426095Z E       Not equal to tolerance rtol=1e-07, atol=1e-09
2025-03-28T06:36:32.4426645Z E       Comparing the output of LinearRegression.predict revealed that fitting with `sample_weight` is not equivalent to fitting with removed or repeated data points.
2025-03-28T06:36:32.4427542Z E       Mismatched elements: 6 / 15 (40%)
2025-03-28T06:36:32.4427985Z E       Max absolute difference among violations: 2.51014256
2025-03-28T06:36:32.4428409Z E       Max relative difference among violations: 2.17024526
2025-03-28T06:36:32.4428858Z E        ACTUAL: array([ 8.881784e-16,  1.000000e+00,  2.000000e+00,  1.185498e+00,
2025-03-28T06:36:32.4429327Z E               4.062418e+00,  1.000000e+00,  2.000000e+00,  2.000000e+00,
2025-03-28T06:36:32.4429765Z E               4.105658e+00,  2.000000e+00, -2.799363e-02, -8.906428e-01,
2025-03-28T06:36:32.4430365Z E              -8.008100e-01,  1.000000e+00,  1.000000e+00])
2025-03-28T06:36:32.4430805Z E        DESIRED: array([0.      , 1.      , 2.      , 0.941865, 1.726709, 1.      ,
2025-03-28T06:36:32.4431250Z E              2.      , 2.      , 1.872389, 2.      , 1.508778, 0.761074,
2025-03-28T06:36:32.4431622Z E              1.709333, 1.      , 1.      ])
2025-03-28T06:36:32.4431810Z 
2025-03-28T06:36:32.4432416Z actual     = array([ 8.88178420e-16,  1.00000000e+00,  2.00000000e+00,  1.18549798e+00,
2025-03-28T06:36:32.4433326Z         4.06241761e+00,  1.00000000e+00,  2...5767e+00,  2.00000000e+00, -2.79936287e-02, -8.90642835e-01,
2025-03-28T06:36:32.4433865Z        -8.00809991e-01,  1.00000000e+00,  1.00000000e+00])
2025-03-28T06:36:32.4434473Z atol       = 1e-09
2025-03-28T06:36:32.4434795Z desired    = array([0.        , 1.        , 2.        , 0.94186541, 1.72670876,
2025-03-28T06:36:32.4435122Z        1.        , 2.        , 2.        , 1.8723887 , 2.        ,
2025-03-28T06:36:32.4435605Z        1.50877777, 0.76107365, 1.70933257, 1.        , 1.        ])
2025-03-28T06:36:32.4435932Z dtypes     = [dtype('float64'), dtype('float64')]
2025-03-28T06:36:32.4436253Z equal_nan  = True
2025-03-28T06:36:32.4436639Z err_msg    = 'Comparing the output of LinearRegression.predict revealed that fitting with `sample_weight` is not equivalent to fitting with removed or repeated data points.'
2025-03-28T06:36:32.4437185Z rtol       = 1e-07
2025-03-28T06:36:32.4439658Z verbose    = True
2025-03-28T06:36:32.4439971Z 
2025-03-28T06:36:32.4440407Z /io/sklearn/utils/_testing.py:237: AssertionError

Looking at the software runtime info of each I only see two differences:

  • the pip version;
  • the CPU model.

All other dependencies seem to match, including the openblas version inspected by threadpoolctl.

EDIT: this is wrong, the scipy version is not the same and I missed it.

About this issue

  • Original URL
  • State: closed
  • Created 2 years ago
  • Comments: 21 (17 by maintainers)

Most upvoted comments

@lesteve @antoinebaker actually the pyodide build is running on an older scipy release (scipy 1.14.1) than the debian_32bit build (scipy 1.15.2):

https://github.com/scikit-learn/scikit-learn/actions/runs/14163691264/job/39673230652#step:3:1863

So it’s very likely that this will crash once the pyodide distribution gets updated.

So performing a root cause analysis / minimal reproducer to report upstream is likely useful.

Hi again, just merged a fix for the reproducer and hoping that it also fixes the actual issue in the test suite. Could you please retry with the new main and let me know?

It also looks like NNLS has been rewritten in C very recently scipy/scipy#22524 so I guess you may want to build scipy from the main branch as well to see whether the problem is still there …

Unfortunately the problem remains on scipy main.

It also looks like NNLS has been rewritten in C very recently https://github.com/scipy/scipy/pull/22524 so I guess you may want to build scipy from the main branch as well to see whether the problem is still there …

Thanks very much for spotting the root cause @antoinebaker. Could you please try to craft a minimal reproducer for scipy that does not rely on scikit-learn?

Meanwhile I will change our test suite to mark this case as XFAIL.

There was a reimplementation of nnls in cython between the two versions https://docs.scipy.org/doc/scipy-1.15.1/release/1.15.0-notes.html#scipy-optimize-improvements