[CI][Python] New unbiased skew and kurtosis test fails on test-debian-12-python-3-i386

[raulcd]

Describe the bug, including details regarding any error messages, version, and platform.

The test-debian-12-python-3-i386 job fails with:

______________ test_unbiased_skew_and_kurtosis[input0-expected0] _______________

input = [1.0, 2.0, 3.0, 40.0, None]
expected = {'kurtosis': 3.9631931024230695, 'skew': 1.988947740397821}

    @pytest.mark.parametrize("input, expected", (
        (
            [1.0, 2.0, 3.0, 40.0, None],
            {'skew': 1.988947740397821, 'kurtosis': 3.9631931024230695}
        ),
        ([1, 2, 40], {'skew': 1.7281098503730385, 'kurtosis': None}),
        ([1, 40], {'skew': None, 'kurtosis': None}),
    ))
    def test_unbiased_skew_and_kurtosis(input, expected):
        arrow_skew = pc.skew(input, skip_nulls=True, biased=False)
        arrow_kurtosis = pc.kurtosis(input, skip_nulls=True, biased=False)
        assert arrow_skew == pa.scalar(expected['skew'], type=pa.float64())
>       assert arrow_kurtosis == pa.scalar(expected['kurtosis'], type=pa.float64())
E       assert <pyarrow.DoubleScalar: 3.9631931024230713> == <pyarrow.DoubleScalar: 3.9631931024230695>
E        +  where <pyarrow.DoubleScalar: 3.9631931024230695> = <cyfunction scalar at 0xe097a648>(3.9631931024230695, type=DataType(double))
E        +    where <cyfunction scalar at 0xe097a648> = pa.scalar
E        +    and   DataType(double) = <cyfunction float64 at 0xe0977b70>()
E        +      where <cyfunction float64 at 0xe0977b70> = pa.float64

arrow-dev/lib/python3.11/site-packages/pyarrow/tests/test_compute.py:473: AssertionError
______________ test_unbiased_skew_and_kurtosis[input1-expected1] _______________

input = [1, 2, 40], expected = {'kurtosis': None, 'skew': 1.7281098503730385}

    @pytest.mark.parametrize("input, expected", (
        (
            [1.0, 2.0, 3.0, 40.0, None],
            {'skew': 1.988947740397821, 'kurtosis': 3.9631931024230695}
        ),
        ([1, 2, 40], {'skew': 1.7281098503730385, 'kurtosis': None}),
        ([1, 40], {'skew': None, 'kurtosis': None}),
    ))
    def test_unbiased_skew_and_kurtosis(input, expected):
        arrow_skew = pc.skew(input, skip_nulls=True, biased=False)
        arrow_kurtosis = pc.kurtosis(input, skip_nulls=True, biased=False)
>       assert arrow_skew == pa.scalar(expected['skew'], type=pa.float64())
E       assert <pyarrow.DoubleScalar: 1.7281098503730388> == <pyarrow.DoubleScalar: 1.7281098503730385>
E        +  where <pyarrow.DoubleScalar: 1.7281098503730385> = <cyfunction scalar at 0xe097a648>(1.7281098503730385, type=DataType(double))
E        +    where <cyfunction scalar at 0xe097a648> = pa.scalar
E        +    and   DataType(double) = <cyfunction float64 at 0xe0977b70>()
E        +      where <cyfunction float64 at 0xe0977b70> = pa.float64

Component(s)

Python, Continuous Integration

[raulcd]

Issue resolved by pull request 45929
#45929