More Groupers / user stories / strategies

[dcherian]

We need more discussion of strategies to label groups or perhaps just more convenient Grouper objects

• QuantileBinGrouper?

  • `groupby_bins` based on quantiles pydata/xarray#7110

• SeasonGrouper?

  • xarray_ Group by season pydata/xarray#6664 : JF,MAM,JJAS,OND
  • https://stackoverflow.com/questions/70658388/how-to-select-djfm-season-instead-of-xarray-groupby
  • https://stackoverflow.com/questions/68455725/how-to-enable-season-selection-as-jjas-instead-of-jja-in-xarray
  • rolling groupby : Computing 'seasonal means' spanning 4 months (with resample or groupy) pydata/xarray#6180 - this in particular could be a special-case of SeasonGrouper. We can detect overlaps. and set up the problem appropriately.

• SeasonResampler:

  • https://stackoverflow.com/questions/71260515/xarray-resample-data-to-variable-length-seasonal-frequencies/71270288#71270288
  • Calculating ONDJFM mean for each season pydata/xarray#6865
  • https://stackoverflow.com/questions/71755223/group-by-custom-period-annually-in-xarray/71757404#71757404

• Grouping by the hour angle adjusted day of year pydata/xarray#7502

• ResampleGrouper to complement TimeResampleGrouper: Can Resample dim be spatial and not just datetime? pydata/xarray#4008 : ?

• DayofYearGrouper:

  • Handling dayofyear with leap years is also a common issue: https://stackoverflow.com/questions/68677362/daily-climatology-in-xarray-with-leap-years-accounted-for

[dcherian]

Potential SeasonGrouper syntax

SeasonGrouper(["JF", "MAM", "JJAS", "OND"])
SeasonGrouper(["DJFM", "MAMJ", "JJAS", "SOND"])

The list would be used as expected_groups so the output is in the right order.

Here's an interesting thread on SeasonGrouper design: xCDAT/xcdat#416

ds.temporal.group_average(
    'pr',
    freq='season',
    season_config={
        'dec_mode': 'DJF',  
        'drop_incomplete_djf': True, # Or drop_incomplete_season
        'custom_seasons': ['Nov', 'Dec', 'Jan', 'Feb', 'Mar']
    }    
)

All of this is a factorizing problem, I like the idea of custom Grouper objects with custom factorization

[dcherian]

Some prior art: from polars

1. https://pola-rs.github.io/polars/py-polars/html/reference/dataframe/api/polars.DataFrame.group_by_dynamic.html

   Group based on a time value (or index value of type Int32, Int64).

   Time windows are calculated and rows are assigned to windows. Different from a normal group by is that a row can be member of multiple groups. By default, the windows look like:

    [start, start + period)
   
    [start + every, start + every + period)
   
    [start + 2*every, start + 2*every + period)
   
    …
   

   where start is determined by start_by, offset, and every (see parameter descriptions below).

2. https://pola-rs.github.io/polars/py-polars/html/reference/dataframe/api/polars.DataFrame.rolling.html#polars.DataFrame.rolling (I don't think this is normal rolling?)