[Enh]: Add `Expr.any_value`?

[danielgafni]

We would like to learn about your use case. For example, if this feature is needed to adopt Narwhals in an open source project, could you please enter the link to it below?

https://github.com/anam-org/metaxy

Please describe the purpose of the new feature or describe the problem to solve.

I would like to perform a group by on a DataFrame and take a random (any) row from each group.

I could have used df.group_by().agg(nw.all().first()), because rows within groups are not sorted already, but this currently fails with:

E           narwhals.exceptions.InvalidOperationError: Order-dependent expressions are not supported for use in LazyFrame.
E
E           Hint: To make the expression valid, use `.over` with `order_by` specified.
E
E           For example, if you wrote `nw.col('price').cum_sum()` and you have a column
E           `'date'` which orders your data, then replace:
E
E              nw.col('price').cum_sum()
E
E            with:
E
E              nw.col('price').cum_sum().over(order_by='date')
E                                       ^^^^^^^^^^^^^^^^^^^^^^
E
E           See https://narwhals-dev.github.io/narwhals/concepts/order_dependence/.

In my case performing a sort would introduce a performance penalty, so I would like to avoid it.

It seems like the best I could do now is with .is_first_distinct() over a dummy column:

def group_and_take_any(df: FrameT, cols: Sequence[str]) -> FrameT:
     return (
            df.with_columns(nw.lit(True).alias("_dummy"))
            .filter(
                nw.col("_dummy")
                .is_first_distinct()
                .over(*cols, order_by="_dummy")
            )
            .drop("_dummy")
        )

Suggest a solution if possible.

No response

If you have tried alternatives, please describe them below.

No response

Additional information that may help us understand your needs.

No response

[danielgafni]

By the way, it looks like having within a GroupBy in Polars will be available in the future:

https://github.com/pola-rs/polars/pull/23550

[MarcoGorelli]

thanks for the request!

I like the idea of having any_value, even though it's not available in Polars

[FBruzzesi]

Wouldn't Expr.sample(n=1) be what we are after without having to diverge from the polars API?

[MarcoGorelli]

that doesn't really behave like an aggregation though

In [9]: df = pl.DataFrame({'a': [1,1,2], 'b': [4,5,6]})

In [10]: df.group_by('a').agg(pl.col('b').sample(n=1))
Out[10]:
shape: (2, 2)
┌─────┬───────────┐
│ a   ┆ b         │
│ --- ┆ ---       │
│ i64 ┆ list[i64] │
╞═════╪═══════════╡
│ 1   ┆ [5]       │
│ 2   ┆ [6]       │
└─────┴───────────┘