NVIDIA / NVIDIA/cudf

[FEA] Support for dynamic groupby in cudf-polars

Open
#17,144 4 comments 0 reactions 1 assignee Claimed by @wence- View on GitHub
cudf-polars feature request libcudf Python
Dominant language
C++
Stars
9.8k
Forks
1.1k
Avg merge
3d 6m
Merged PRs (30d)
278

Description

**Is your feature request related to a problem? Please describe.**

Polars supports `group_by_dynamic` that does (approximately) grouped rolling windows.

We do not currently support this in cudf-polars, what is more, the IR doesn't distinguish between normal groupbys and dynamic groupbys when we get it from polars. Consequently, we only fall back to CPU based on some heuristics, and sometimes just do the wrong thing.

**Describe the solution you'd like**

- [x] Expose dynamic groupbys in the IR (pola-rs/polars#19385, in polars 1.11)
- [x] Raise that they're not implemented (#17154, in cudf nightlies)
- [ ] Implement them.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.