Possible direction of development can allow seqences of indexes in case of grouping by map. Currently you can build grouped dataset out of a map with keys as a group names and vals as list of row indices. If we allow seq of row indices too it could solve this issue.
Contributor should verify against regresion all functions relying on group-name especially: ungroup and aggregate. Also pivot operations.
grouped dataset usually has unique group names, however it can contain names duplicated (what is unique it's group-id). So there is an option to build such grouping to make partition-by functionality happen, as described here: https://clojurians.zulipchat.com/#narrow/stream/264992-ml-study/topic/session.2013.2E1/near/228939222
Possible direction of development can allow seqences of indexes in case of grouping by map. Currently you can build grouped dataset out of a map with keys as a group names and vals as list of row indices. If we allow seq of row indices too it could solve this issue.
Contributor should verify against regresion all functions relying on
group-nameespecially:ungroupandaggregate. Also pivot operations.