DataFrame.describe excludes top and freq for empty DataFrame

[TomAugspurger]

Right now, the output shape and dtype of DataFrame.describe for object columns depends on whether the DataFrame is empty.

Code Sample, a copy-pastable example if possible

In [75]: x = pd.DataFrame({"A": ['a', np.nan, np.nan]})

In [76]: x.describe()
Out[76]:
        A
count   1
unique  1
top     a
freq    1

In [77]: x.iloc[:0].describe()
Out[77]:
        A
count   0
unique  0

Problem description

This leads to instability in the output dtypes and shape.

Would people prefer that we use np.NaN or None for the top and freq in this case? I believe there's no ambiguity, since we drop missing values before computing.

While the output consistency would be nice, it's not clear to me what's actually best for users here.

[WillAyd]

Makes sense to me

[enisnazif]

Seems reasonable to me - would probably also need a note in the docs explaining what None means in this context. Mind if I attempt a fix?

[TomAugspurger]

That would be great!

…

On Sat, May 18, 2019 at 11:10 AM enisnazif ***@***.***> wrote: Seems reasonable to me - would probably also need a note in the docs explaining what None means in this context. Mind if I attempt a fix? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#26397?email_source=notifications&email_token=AAKAOIWTQX3YM2K4JJCG26LPWAS7XA5CNFSM4HM5TRIKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODVWRLSY#issuecomment-493688267>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAKAOIXHHEB2LAIQB6E764DPWAS7XANCNFSM4HM5TRIA> .

[enisnazif]

Thanks - opened a PR at #26474