API: value_counts on a Categorical Series should have a CategoricalIndex

[TomAugspurger]

Thoughts on this?

In [8]: import pandas.util.testing as tm

In [9]: s = pd.Series(tm.makeCategoricalIndex(k=100))

In [10]: s.value_counts()
Out[10]:
vcKH    40
vXR9    31
Zn8J    29
dtype: int64

In [11]: s.value_counts().index
Out[11]: Index(['vcKH', 'vXR9', 'Zn8J'], dtype='object')

Now that we have CategoricalIndex (thanks Jeff), should that type be preserved so that Out[11] is a CategoricalIndex? My use-case (not shown in this example) is when the original categories are ordered, you get your value_counts and then want to sort the index.

[jorisvandenbossche]

I think that is the logical consequence of having a categorical type in series and having a CategorialIndex, so OK for me!
It is similar as how df.groupby('cat').count() now also preserves this categorical type. So in that regard, it would only be consistent that value_counts returns that as well.

[jorisvandenbossche]

There can of course be some incompatibilities, such as with plotting afterwards, but that is similar as eg the groupby case (we had some reports on that, eg #10140)

[jreback]

yes I would agree with that logic. go for it @TomAugspurger

[TomAugspurger]

Cool. I'll shot for next week.