CI: Remove float16 fixture value

[SaturnFromTitan]

• xref CI: value_counts tests failing sometimes #32220
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff

Hopefully, this fixes the flaky test

[WillAyd]

@jreback any reason you know of we should keep this fixture value?

[jreback]

we very little support for float16; it does work in places but if it’s causing issues then maybe should just xfail those

[SaturnFromTitan]

If we don't support it, then removing it from the fixture values seems more maintainbale than adding a lot of xfail(strict=False) @jreback

[simonjayhawkins]

If we don't support it, then removing it from the fixture values seems more maintainbale than adding a lot of xfail(strict=False) @jreback

It appears that the test design is flawed. so we should change the test. Looking at the differences -0.68505859375 appears once in the result and twice in the expected. The result is correct. The test assumes that the randomly generated numbers in the Series will be unique. This is not always the case.

Series length are different
E       [left]:  28, Float64Index([  -0.355712890625,    -0.71435546875,     0.39794921875,
E                      -0.1519775390625,       -1.14453125,       1.181640625,
E                         -1.3037109375,        2.13671875,      -0.751953125,
E                           2.005859375,   -0.399169921875, -0.06842041015625,
E                        -0.68505859375,        1.07421875,      0.6845703125,
E                       -0.133544921875,      1.6826171875,     0.41357421875,
E                        -0.03857421875,          1.015625,      0.2802734375,
E                       -0.477294921875,  -0.1910400390625,     0.33544921875,
E                         0.59423828125,     -0.9111328125,   -0.098388671875,
E                                   nan],
E                    dtype='float64')
E       [right]: 29, Float64Index([  -0.355712890625,    -0.71435546875,     0.39794921875,
E                      -0.1519775390625,       -1.14453125,       1.181640625,
E                         -1.3037109375,        2.13671875,      -0.751953125,
E                           2.005859375,   -0.399169921875, -0.06842041015625,
E                            1.07421875,      0.6845703125,   -0.133544921875,
E                          1.6826171875,    -0.68505859375,     0.41357421875,
E                        -0.03857421875,          1.015625,      0.2802734375,
E                       -0.477294921875,  -0.1910400390625,     0.33544921875,
E                         0.59423828125,    -0.68505859375,     -0.9111328125,
E                       -0.098388671875,               nan],
E                    dtype='float64')

[simonjayhawkins]

If we don't support it, then removing it from the fixture values seems more maintainbale than adding a lot of xfail(strict=False) @jreback

It appears that the test design is flawed. so we should change the test.

but to get ci to green quickly for now could just...

diff --git a/pandas/tests/base/test_ops.py b/pandas/tests/base/test_ops.py
index 625d55900..54e459ed2 100644
--- a/pandas/tests/base/test_ops.py
+++ b/pandas/tests/base/test_ops.py
@@ -277,6 +277,8 @@ class TestIndexOps(Ops):
             pytest.skip(f"values of {klass} cannot be changed")
         elif isinstance(orig, pd.MultiIndex):
             pytest.skip("MultiIndex doesn't support isna")
+        elif not orig.is_unique:
+            pytest.xfail("FIXME: test assumes fixture has unique values gh-32220")
 
         # special assign to the numpy array
         if is_datetime64tz_dtype(obj):

and not mark the issue as closed.

[SaturnFromTitan]

@simonjayhawkins Ah ok, I got confused because the issue mentioned test_value_counts_unique_nunique failing and that one already has an xfail for duplicated values.

However, you're right and test_value_counts_unique_nunique_null is actually failing. I think adding the xfail now and refactoring the tests later is the best approach 👍

[SaturnFromTitan]

Changed the PR to use the same xfail message as test_value_counts_unique_nunique. Both share the same flawed implementation

[simonjayhawkins]

Changed the PR to use the same xfail message as test_value_counts_unique_nunique. Both share the same flawed implementation

makes sense and we should leave the issue open.

[simonjayhawkins]

no need to backport as #32046 was not backported.

[simonjayhawkins]

ci is green ex. conda issue.

[simonjayhawkins]

since #32241 is merged, merging this to get ci to green.

[simonjayhawkins]

Thanks @SaturnFromTitan