TST: Test failure on 32bit for TestSparseGroupBy.test_aggfuncs

[jschendel]

The test_aggfuncs test in /sparse/test_groupby.py is failing on 32bit: https://travis-ci.org/MacPython/pandas-wheels/jobs/451706570#L5281

Specifically, the pytest output:

_______________________ TestSparseGroupBy.test_aggfuncs ________________________
self = <pandas.tests.sparse.test_groupby.TestSparseGroupBy object at 0xde3f08ec>
    def test_aggfuncs(self):
        sparse_grouped = self.sparse.groupby('A')
        dense_grouped = self.dense.groupby('A')
    
        result = sparse_grouped.mean().to_sparse()
        expected = dense_grouped.mean().to_sparse()
    
>       tm.assert_frame_equal(result, expected)
/venv/lib/python3.6/site-packages/pandas/tests/sparse/test_groupby.py:50: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/venv/lib/python3.6/site-packages/pandas/util/testing.py:1185: in assert_extension_array_equal
    assert_numpy_array_equal(left_valid, right_valid)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
left = array([-0.17387645482451206, 0.3414148016424936], dtype=object)
right = array([-0.17387645482451206, 0.3414148016424937], dtype=object)
err_msg = None
    def _raise(left, right, err_msg):
        if err_msg is None:
            if left.shape != right.shape:
                raise_assert_detail(obj, '{obj} shapes are different'
                                    .format(obj=obj), left.shape, right.shape)
    
            diff = 0
            for l, r in zip(left, right):
                # count up differences
                if not array_equivalent(l, r, strict_nan=strict_nan):
                    diff += 1
    
            diff = diff * 100.0 / left.size
            msg = '{obj} values are different ({pct} %)'.format(
                obj=obj, pct=np.round(diff, 5))
>           raise_assert_detail(obj, msg, left, right)
E           AssertionError: numpy array are different
E           
E           numpy array values are different (50.0 %)
E           [left]:  [-0.17387645482451206, 0.3414148016424936]
E           [right]: [-0.17387645482451206, 0.3414148016424937]
/venv/lib/python3.6/site-packages/pandas/util/testing.py:1146: AssertionError

The test in question:

pandas/pandas/tests/sparse/test_groupby.py

Lines 43 to 48 in efd1844

	def test_aggfuncs(self):
	sparse_grouped = self.sparse.groupby('A')
	dense_grouped = self.dense.groupby('A')
	result = sparse_grouped.mean().to_sparse()
	expected = dense_grouped.mean().to_sparse()

xref #23318 : it looks like this is the pr that caused the change?

cc : @jreback @TomAugspurger @5hirish

[jreback]

can u just change this to be an all close test rather than exact match

[TomAugspurger]

Was this closed by #23660?

[jschendel]

@TomAugspurger : No, this issue is unrelated to the interval 32bit failures. The #23318 PR I xref'd is what introduced this.

I'm planning to open a new issue in a little bit related to how assert_extension_array_equal handles precision, as right now it basically does exact=True style equality, which is inconsistent with the other assert_*_equal methods. Making the assert_extension_array_equal behavior consistent would resolve this issue without needing to explicitly change the failing test to an all close test.

[jschendel]

It looks like #23808 fixed this issue, as the latest pandas-wheels build appears to include that PR and was successful: https://travis-ci.org/MacPython/pandas-wheels/builds/457775113