ENH: Period ops NaT & timedelta ops

[jreback]

see related #5200, #4551, #4056

seem that a direct subtraction of PeriodIndexes would accomplish this,
but don't seem to support those types of ops...

This is currently an intersection (with a '-'), but want to do a timedelta like op

In [40]: pd.PeriodIndex(start='20130101',end='20130501',freq='M').sub(
pd.PeriodIndex(start='20130201',end='20130401',freq='M'))

Partially supported actuallly

In [5]:  pd.PeriodIndex(start='20130101',end='20130501',freq='M').to_series()-pd.PeriodIndex(start='20130205',end='20130405',freq='M').to_series()
Out[5]: 
2013-01   NaN
2013-02     0
2013-03     0
2013-04     0
2013-05   NaN
Freq: M, dtype: float64

In [6]:  pd.PeriodIndex(start='20130101',end='20130201',freq='B').to_series()-pd.Period('20130101',freq='B')

TypeError: unsupported operand type(s) for -: 'int' and 'Period'

[jtratner]

Dunno, think you should always have - be intersection, instead of adding ambiguity like that.

[jreback]

of course.....I am talking about adding a subtraction op to datetime like Index (we already support Index + offset),
should maybe have a .sub or something (or course can just do this in Series...which is prob better anyhow as I think we figured it all out)

[jbrockmendel]

One of the OP examples now works:

>>> pd.PeriodIndex(start='20130101',end='20130201',freq='B').to_series()-pd.Period('20130101',freq='B')
2013-01-01     0
2013-01-02     1
2013-01-03     2
2013-01-04     3
2013-01-07     4
2013-01-08     5
2013-01-09     6
2013-01-10     7
2013-01-11     8
2013-01-14     9
2013-01-15    10
2013-01-16    11
2013-01-17    12
2013-01-18    13
2013-01-21    14
2013-01-22    15
2013-01-23    16
2013-01-24    17
2013-01-25    18
2013-01-28    19
2013-01-29    20
2013-01-30    21
2013-01-31    22
2013-02-01    23
Freq: B, dtype: object

[jbrockmendel]

@jreback closeable?