Parallel processing in groupby operations

[mhooreman]

Hello,

It would be nice to have the possibility to do parallel processing using joblib when applying something to a groupby. I'm dreaming about a parallel optional argument which defaults to False.

I think it's best to have that parallel argument while grouping

The only option I have now is to make list on the groupby results, apply a function on that list using joblib, and concatenate the results. It leads to unefficient and difficult to read code.

############
# Examples #
############
data.groupby(...).fillna(..., parallel=True)
data.groupby(...).apply(..., parallel=True)
data.groupby(...).whatever(..., parallel=True)
data.groupby(...).whatever(...) # not parallel

[jreback]

duplicate of #13111 this is non-trivial and should only have an api in pandas not an actual implementation, this is currently possible via dask (though not directly thru a pandas api).

Certainly would take a PR, but this is a fairly large effort / surface area.