BUG/API: PeriodIndex.unique returns ndarray

[sinhrks]

Currently, Index.unique returns ndarray but DatetimeIndex.unique returns DatetimeIndex to preserve datetime information. I think PeriodIndex.unique should also return PeriodIndex (currently it returns ndarray).

I'd like to ask which is better to fix all Index.unique to return Index for consistency, or only fix PeriodIndex.

didx = pd.DatetimeIndex([datetime.datetime(2014, 1, 1), datetime.datetime(2014, 2, 1)])
didx.unique()
# <class 'pandas.tseries.index.DatetimeIndex'>
# [2014-01-01, 2014-02-01]

pidx = pd.PeriodIndex(['2014-01', '2014-02'], freq='M')
pidx.unique()
# [528 529]
type(pidx.unique())
# <type 'numpy.ndarray'>

[jreback]

this is a historical / compat as well. In theory could fix all of them, but I think lots of things (tests) will break. I am not sure if anything real will actually break. Give a try and see (but I agree PeriodIndex.unique should return a PeriodIndex