Add base test for Series[ExtensionArray].sort_values with sort=False

[akulagrawal]

• closes #Add base test for Series[ExtensionArray].value_counts with sort=False #23074
• one base test added / passed
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

[pep8speaks]

Hello @akulagrawal! Thanks for submitting the PR.

• There are no PEP8 issues in the file pandas/tests/extension/base/methods.py !

[TomAugspurger]

Some failing tests that'll need to be fixed up. I haven't looked closely.

This raises an ExtensionArray API question: should we require that .value_counts() preserve order? Though IIRC, I think that @jorisvandenbossche wanted to remove .values_counts from the API?

At the very least, we should ensure that our EAs implement value_counts correctly, so that the following match

In [21]: ser = pd.Series(pd.core.arrays.integer_array([1, 2, 2, 2, 1, 0]))

In [22]: ser.value_counts(sort=False)
Out[22]:
2    3
1    2
0    1
dtype: int64

In [23]: ser.astype(int).value_counts(sort=False)
Out[23]:
0    1
1    2
2    3
dtype: int64

[akulagrawal]

Oh my bad! I am a beginner and this is my first contribution to pandas. As for the match, according to my system the output is:

In [21]: ser = pd.Series([1, 2, 2, 2, 1, 0])
In [22]: ser.value_counts(sort=False)
Out[22]:
0 1
1 2
2 3
dtype: int64
In [23]: ser.astype(int).value_counts(sort=False)
Out[23]:
0 1
1 2
2 3
dtype: int64

I understand they are incorrect but they don't match with yours atleast

[TomAugspurger]

Yeah, I'm not sure why those don't match. Presumably there's whatever internally does this hashing doesn't preserve order.

I'd recommend holding off on this PR and maybe finding another issue until we can decide exactly what we want the requirements on order to be here.

[akulagrawal]

Oh... first of all there was some error in my last comment. I meant they matched on my system (output updated in the comment), though both seem to be incorrect.
Alright, moving onto other issue. Thanks @TomAugspurger !

[jorisvandenbossche]

Closing the PR (to shorten our backlog of open PRs). We still have the issue to track it, so if we decide to keep value_counts on that issue, we can simply reopen this PR. @akulagrawal That is OK?

[akulagrawal]

yes sure :)