DataFrame.mean includes datetime64[ns] dtypes in 0.24.0rc1

[TomAugspurger]

I think this isn't deliberate.

# 0.23.4
In [1]: import pandas as pd

In [2]: pd.DataFrame({"A": pd.date_range('2000', periods=12)}).mean()
Out[2]: Series([], dtype: float64)

# 0.24.0rc1
In [2]: pd.DataFrame({"A": pd.date_range('2000', periods=12)}).mean()
Out[2]:
A   2124-07-04 18:21:24.687506176
dtype: datetime64[ns]

We continue to raise on Series[datetime64[ns]].mean()

[jreback]

hmm i think we did enable this recently

[TomAugspurger]

I believe @jbrockmendel had a branch implementing mean and such, but I don't think it's been merged.

…

On Sun, Jan 13, 2019 at 11:56 AM Jeff Reback ***@***.***> wrote: hmm i think we did enable this recently — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#24752 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQHItDgd3Ctr9aSxgEq7O9zuUtSwIvRks5vC3NBgaJpZM4Z9SGm> .

[jbrockmendel]

I think this isn't deliberate.

Certainly returning a nonsense answer isn't deliberate. IIRC we made nanops.nanmean accept datetime64[ns] but didn't yet enable it in the Index/EA subclasses.

I've got a couple of branches implementing reductions; I'll rebase and see if any of them are near ready.

[jbrockmendel]

The weird result is happening in DataFrame._reduce when nanops.nanmean is called on df.values:

>>> nanops.nanmean(self.values, axis=0)
array(['2124-07-04T18:21:24.687506176'], dtype='datetime64[ns]')

If we weren't passing axis=0, we would get a more reasonable-looking:

>>> nanops.nanmean(self.values, axis=None)
Timestamp('2000-01-06 12:00:00')