Deprecate ``shuffle`` keyword in favour of ``shuffle_method`` for DataFrame methods about dask HOT 5 OPEN

Are there cases where shuffle= isn't specific enough? All other things being equal, I think a shorter name is better

from dask.

shuffle for me is just very unspecific, could be a bool or something else. shuffle_method expresses the intent of the option very clearly. It's inconsistent with the config option as well

from dask.

shuffle for me is just very unspecific, could be a bool or something else. shuffle_method expresses the intent of the option very clearly. It's inconsistent with the config option as well

+1

from dask.

Hmm, there's a tradeoff between making things very explicit and having overly long parameter names. In my mind shuffle="p2p" is clear enough (especially if it's clearly documented and used consistently as a keyword).

Though to be clear, I don't care a ton about this specific case -- my hope is most people are happy with the default shuffle algorithm and won't ever need to specify this option. I mostly just want to bring up the explicit vs. concise tradeoff to make sure we're considering it.

from dask.

shuffle for me is just very unspecific, could be a bool or something else. shuffle_method expresses the intent of the option very clearly. It's inconsistent with the config option as well

Note that the shuffle keyword can be a bool for groupby-aggregates:

In light of this, renaming shuffle to shuffle_method for all the cases where only a method is accepted is even more relevant to highlight the distinction.

We may also want to reconsider dumping all kinds of things into shuffle for groupby-aggregates.

from dask.