Add "season" as a time/date component to DatetimeIndex

[shoyer]

To help pandas to get better traction in the Earth sciences, it would be awesome to add support for climatological "season" as a Time/Date Component of pandas.DatetimeIndex. Season comes up in geoscience at least as often as "quarter" does in finance.

The standard seasons (the ones worth adding a shortcut for) would be ['DJF', 'MAM', 'JJA', 'SON'], as labeled by the first letter of each month. These could be represented by integers (in a pinch) but string labels would be much more natural -- people talk about the season "JJA" (or "summer" in the Northern hemisphere), not "season 3".

Actually, this might be a nice show case for the brand new CategoricalBlock (#7217).

For reference, here is some vectorized code to do the calculation from month numbers:

SEASONS = np.array(['DJF', 'MAM', 'JJA', 'SON'])

month = np.arange(12) + 1
season = SEASONS[(month // 3) % 4]
print season 

(We actually already use this code to add "season" to xray, but I think this would be useful more broadly as well.)

[jreback]

You just want this, yes? (maybe with labels) (converting to series just to show it)

In [8]: pd.date_range('20131201',periods=5,freq='QS-DEC').to_series()
Out[8]: 
2013-12-01   2013-12-01
2014-03-01   2014-03-01
2014-06-01   2014-06-01
2014-09-01   2014-09-01
2014-12-01   2014-12-01
Freq: QS-DEC, dtype: datetime64[ns]

[shoyer]

Yes, that's right.

[SoulofAkuma]

take

[SoulofAkuma]

I think there has been some miscommunication regarding what is actually expected here. The original enhancement request mentioned providing season as a Time/Date component, which is not equivalent to just generating a date_range from a given season. If I understand this correctly, this refers to having a property season on a timestamp, that returns the season of a given timestamp. I could extend the timestamps to support this, but I am not sure if this could be considered too niche to add to the generic timestamp implementation.

The other thing is that it is not entirely clear what generating a date_range based on a season should look like. If a season is provided, should that include the months of the season and then skip to the next year? Or just the beginning of the season and then go to the next year or season? This requires some clarification.

[SoulofAkuma]

Will drop working on this as advised by @mroeschke