infer_freq borks on PeriodIndex

[jseabold]

Should PeriodIndex pass through infer_freq and return freqstr or should this be the user's problem? Either is a simple fix. The latter should just have a better error message. Right now

from pandas.tseries.api import infer_freq
from pandas import PeriodIndex

infer_freq(PeriodIndex(start="1/1/1990", periods=20, freq="M"))

[jreback]

inferring is not implemented for PeriodIndex AFAICT

In [11]: p = PeriodIndex(start="1/1/1990", periods=20, freq="M")

In [12]: p.freqstr
Out[12]: 'M'

In [13]: p.freq
Out[13]: 'M'

[jseabold]

Yes. Should it be? I.e., trivially return freqstr since you don't need to do any inferring or have a better error message than the one about converting Int64Index to Datetime. Use case is for time series models in statsmodels. Should we handle it or should infer_freq handle it? (I've already committed a fix, so I don't really care either way.)

[jreback]

I think the use would be if someone puts together multiple Periods and tries to create a PeriodIndex

But I don't think very comment (unlike doing this with Timestamps, which is pretty common)

and unless they are ALL the same freq it would bork anyhow.

So maybe just a better error message.