Ability to sort an array of structs by struct field

[david-waterworth]

Description

There doesn't appear to be a direct way to sort an array of structs by field, i.e.

Given:

df = pl.DataFrame([{"id": 1, "data": [{"key": "A", "value": 2}, {"key": "B", "value": 1}]}])

Sort struct by value

df = pl.DataFrame([{"id": 1, "data": [{"key": "B", "value": 1}, {"key": "A", "value": 2}]}])

There are work-arounds which revolve around either the side-effect that sorting a struct appears to sort by fields in order so by (re)constructing the struct as {"value": 1, "key": "B"} sorts by value, or by exploding the struct, sorting and then re-grouping.

Ideally pl.Expr.list.sort would accept a sort key expression similar to how the python sorted function behaves.

If this is accepted I'd be inclined to implement for all sort methods?

[cmdlineluser]

Polars has a separate Expr.sort_by()

So I'm guessing you'd be asking for a list equivalent of that?

df.with_columns(
   pl.col("data").list.sort_by(
       pl.col("data").list.eval(pl.element().struct["value"])
   )
)
# AttributeError: 'ExprListNameSpace' object has no attribute 'sort_by'

[nameexhaustion]

For now, you can use .list.eval(pl.element().sort_by(pl.element().struct.field("value")))

>>> df.with_columns(
...     sorted=pl.col("data").list.eval(
...         pl.element().sort_by(pl.element().struct.field("value"))
...     )
... )
shape: (1, 3)
┌─────┬────────────────────┬────────────────────┐
│ id  ┆ data               ┆ sorted             │
│ --- ┆ ---                ┆ ---                │
│ i64 ┆ list[struct[2]]    ┆ list[struct[2]]    │
╞═════╪════════════════════╪════════════════════╡
│ 1   ┆ [{"A",2}, {"B",1}] ┆ [{"B",1}, {"A",2}] │
└─────┴────────────────────┴────────────────────┘

[david-waterworth]

Yeah either works, @cmdlineluser I guess list.sort_by is easier to find (i.e it's explicit), but the workaround posted by @nameexhaustion example works for me as well.