[Python] Efficient way to iterate over groups

[MarcoGorelli]

Describe the enhancement requested

In pandas / Polars, I can do:

dict(df.group_by(['a', 'b', 'c']).__iter__())

There doesn't seem to be a built-in way to do this in PyArrow, hence I'm opening this as a feature request

Concretely, if I have

import pyarrow as pa
tbl = pa.table({'a': [1,1,3], 'b': [4, 4, 4], 'c': [1, 3, 2]})

then I'd like a way to end up with

{(3,
  4): pyarrow.Table
 a: int64
 b: int64
 c: int64
 ----
 a: [[3]]
 b: [[4]]
 c: [[2]],
 (1,
  4): pyarrow.Table
 a: int64
 b: int64
 c: int64
 ----
 a: [[1,1]]
 b: [[4,4]]
 c: [[1,3]]}

---

For context, this would be for use in Narwhals, where we have tried to come up with a workaround, but it does exhibit a noticeable slow-down as the number of grouping keys grows - 3 keys is enough for it to be slower than pandas

Component(s)

Python