fix(python): typing for Series methods that can return None

[MatveyF]

Resolves #6623

Fix the type hints for some Series methods that return None.

Also Resolves this

In [4]: pl.Series([None]).cast(pl.Datetime).dt.median()
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In [4], line 1
----> 1 pl.Series([None]).cast(pl.Datetime).dt.median()

File ~/tmp/.venv/lib/python3.8/site-packages/polars/internals/series/datetime.py:94, in DateTimeNameSpace.median(self)
     74 """
     75 Return median as python DateTime.
     76
   (...)
     91
     92 """
     93 s = pli.wrap_s(self._s)
---> 94 out = int(s.median())
     95 return _to_python_datetime(out, s.dtype, s.time_unit)

TypeError: int() argument must be a string, a bytes-like object or a number, not 'NoneType'

This will now return None instead of raising an error.

Notes:

• methods affected are min(), max(), mean(), median(), and quantile()
• added unit tests for median() & mean() when used with datetime

[alexander-beedie]

Nice; FYI, as the tests have to run on python 3.7, you can't use the walrus operator := (which requires >= 3.8). If you can fix that, everything else looks good ;)

[MarcoGorelli]

Hi @MatveyF

Sorry to jump in, just wanted to point out that this would break the following:

result = pl.date_range(datetime(1969, 12, 31), datetime(1970, 1, 2), "1d").dt.median()
assert result == datetime(1970, 1, 1)

You probably need to explicitly check if out is not None rather than just if out

[MarcoGorelli]

Nice! I think you also fixed a bug here - the following fails on master

In [4]: pl.Series([None]).cast(pl.Datetime).dt.median()
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In [4], line 1
----> 1 pl.Series([None]).cast(pl.Datetime).dt.median()

File ~/tmp/.venv/lib/python3.8/site-packages/polars/internals/series/datetime.py:94, in DateTimeNameSpace.median(self)
     74 """
     75 Return median as python DateTime.
     76
   (...)
     91
     92 """
     93 s = pli.wrap_s(self._s)
---> 94 out = int(s.median())
     95 return _to_python_datetime(out, s.dtype, s.time_unit)

TypeError: int() argument must be a string, a bytes-like object or a number, not 'NoneType'

but works in your branch. Might be worth adding it as a test case to make sure nobody accidentally brings it back in the future?

[MarcoGorelli]

No objections, over to maintainers (test could be parametrised, instead of having two separate tests, something like

@pytest.mark.parametrize(
    ('value', 'expected'),
    [(dt.datetime(1970, 1, 1), (dt.datetime(1970, 1, 1)], [None, None])
)
def test_median(value, expected) -> None:
    result = pl.Series([value]).cast(pl.Datetime).dt.median()
    assert result == expected

, but I might be nitpicking here, feel free to ignore me 😄 )

[ritchie46]

Thanks!