ENH: Add tz_localize and tz_convert to Series.dt methods

[TomAugspurger]

In [6]: s = pd.Series(pd.date_range(start='2014-01-01', periods=30, freq='d'))

In [7]: s.dt.tz_localize('US/Central')

Would return a Series with the same index whose values are Timestamps with the timezone set.

I haven't looked at this code too closely, so I'm not sure how difficult it would be and if it's doable for 0.15.

@hayd brought it up on SO: http://stackoverflow.com/questions/25656826/speeding-up-timestamp-operations/25657249?noredirect=1#comment40118386_25657249

[jorisvandenbossche]

The only thing is that the result is no longer a datetime64 column (but maybe we could just warn for that in the docstring?). But of course this is the same with the round-trip through DatetimeIndex.

Furthermore, I think to_pydatetime and to_period would also be useful to add to .dt

[hayd]

so datetime columns with tz become object even if they share tz info (I'd never noticed that!)... is this another case of wanting Series meta data or is there some way we can have dti behaviour here?

In [11]: s = pd.Series(pd.date_range(start='2014-01-01', periods=30, freq='d'));

In [12]: pd.Series(pd.DatetimeIndex(s).tz_localize('US/Central')).head()
Out[12]:
0    2014-01-01 00:00:00-06:00
1    2014-01-02 00:00:00-06:00
2    2014-01-03 00:00:00-06:00
3    2014-01-04 00:00:00-06:00
4    2014-01-05 00:00:00-06:00
dtype: object

+1 on method parity between dti & dt.

[jreback]

it IS possible to fix this and make datetime tz aware columns

need a subclass of the DatetimeBlock internally that keeps the tz (would need to be a uniform tz for the block)

not sure how tricky his would be

[ischwabacher]

The vast majority of the friction I've experienced using pandas has come from the fact that Timestamps are Good and datetime64s are Evil, and which representation my data have at a given time is a pandas sort-of-internal that leaks all over the place. Anything that reduces that friction gets a +1 from me.

[jreback]

yep sort of trivial, going to add as part of #8184 (as well as to_period/to_pydatetime)

[jreback]

In [5]: s = Series(date_range('20130101',periods=3))

In [6]: s
Out[6]: 
0   2013-01-01
1   2013-01-02
2   2013-01-03
dtype: datetime64[ns]

In [7]: s.dt.
s.dt.date              s.dt.dayofyear         s.dt.is_month_start    s.dt.is_year_end       s.dt.minute            s.dt.quarter           s.dt.to_period         s.dt.tz_localize       s.dt.weekofyear        
s.dt.day               s.dt.hour              s.dt.is_quarter_end    s.dt.is_year_start     s.dt.month             s.dt.second            s.dt.to_pydatetime     s.dt.week              s.dt.year              
s.dt.dayofweek         s.dt.is_month_end      s.dt.is_quarter_start  s.dt.microsecond       s.dt.nanosecond        s.dt.time              s.dt.tz_convert        s.dt.weekday           

In [7]: s.dt.tz_localize('US/Eastern')
Out[7]: 
0    2013-01-01 00:00:00-05:00
1    2013-01-02 00:00:00-05:00
2    2013-01-03 00:00:00-05:00
dtype: object

# this returns a scalar (as opposed to a Series)
In [4]: s.dt.tz_localize('US/Eastern').dt.tz
Out[4]: <DstTzInfo 'US/Eastern' EST-1 day, 19:00:00 STD>