CI: master failing windows py37_np141 builds

[TomAugspurger]

e.g https://dev.azure.com/pandas-dev/pandas/_build/results?buildId=15921

================================== FAILURES ===================================
______________________________ test_apply[True] _______________________________
[gw1] win32 -- Python 3.7.4 C:\Miniconda\envs\pandas-dev\python.exe

ordered = True

    @pytest.mark.parametrize("ordered", [True, False])
    def test_apply(ordered):
        # GH 10138
    
        dense = Categorical(list("abc"), ordered=ordered)
    
        # 'b' is in the categories but not in the list
        missing = Categorical(list("aaa"), categories=["a", "b"], ordered=ordered)
        values = np.arange(len(dense))
        df = DataFrame({"missing": missing, "dense": dense, "values": values})
        grouped = df.groupby(["missing", "dense"], observed=True)
    
        # missing category 'b' should still exist in the output index
        idx = MultiIndex.from_arrays([missing, dense], names=["missing", "dense"])
        expected = DataFrame([0, 1, 2.0], index=idx, columns=["values"])
    
        result = grouped.apply(lambda x: np.mean(x))
>       assert_frame_equal(result, expected)
E       AssertionError: DataFrame are different
E       
E       DataFrame shape mismatch
E       [left]:  (3, 3)
E       [right]: (3, 1)

pandas\tests\groupby\test_categorical.py:228: AssertionError
______________________________ test_apply[False] ______________________________
[gw1] win32 -- Python 3.7.4 C:\Miniconda\envs\pandas-dev\python.exe

ordered = False

    @pytest.mark.parametrize("ordered", [True, False])
    def test_apply(ordered):
        # GH 10138
    
        dense = Categorical(list("abc"), ordered=ordered)
    
        # 'b' is in the categories but not in the list
        missing = Categorical(list("aaa"), categories=["a", "b"], ordered=ordered)
        values = np.arange(len(dense))
        df = DataFrame({"missing": missing, "dense": dense, "values": values})
        grouped = df.groupby(["missing", "dense"], observed=True)
    
        # missing category 'b' should still exist in the output index
        idx = MultiIndex.from_arrays([missing, dense], names=["missing", "dense"])
        expected = DataFrame([0, 1, 2.0], index=idx, columns=["values"])
    
        result = grouped.apply(lambda x: np.mean(x))
>       assert_frame_equal(result, expected)
E       AssertionError: DataFrame are different
E       
E       DataFrame shape mismatch
E       [left]:  (3, 3)
E       [right]: (3, 1)

pandas\tests\groupby\test_categorical.py:228: AssertionError

[charlesdong1991]

ahh, i also see this error in my PR and i spent quite time debugging where PR went wrong, i thought it was my change that failed the check... I am actually about to ask how to only test code on windows environment because it passes checks on my laptop :) thanks for opening the issue! ^^

[TomAugspurger]

It seems to be a flaky test, if you push to your PR again then it may pass.

[TomAugspurger]

I set up a windows VM and was able to reproduce the failures occasionally. Trying out #27909.

[TomAugspurger]

#27909 doesn't fix it :/

[TomAugspurger]

Very strange.

We have res = f(group), where f is (AFAICT) lambda x: np.mean(x).

I have no idea what's going on.

Given that it's an only observed on older NumPy's (right?) maybe we just skip for NumPy 1.14?

[TomAugspurger]

xref #21636 too.