Filtering a column while using 'over' doesn't work as expected

[Andre-Medina]

Filtering a column while using 'over' doesn't work as expected

[ritchie46]

The query without first shouldn't be allowed. You should finish with an aggregation, or apply only element wise oprrations, otherwise the groups elements cannot be mapped to the original group.

We should raise here.

[Andre-Medina]

Sounds like a plan, so an only method for the expressions to grab the only value present in the grouped expression? e.g.

(
    data_with_groups
    .with_columns(
        pl.col('data')
        .truediv(pl.col('data').filter(pl.col('year') == 2021).only())
        .over('groups')
        .alias('ratio')
    )
)

That would be perfect for this use case I think, then also be able to flag if there's too many or not enough filtered data points per group.

[deanm0000]

My gut reaction to an .only method was to scoff at it as redundant for first but after thinking about it for a minute I think a method that assumes there will only be a single item but raises if that assumption is wrong.

It would work a little bit like expr.first(), if expr.len()!=1: raise OnlyError

[Andre-Medina]

It would work a little bit like expr.first(), if expr.len()!=1: raise OnlyError

yes exactly what i was thinking. as just using .first() on incorrect groupings would just silently fail for this use case and i feel what im trying to accomplish is pretty foundational use case.

[Andre-Medina]

@deanm0000 should I re-raise this as a feature request?

[deanm0000]

I suppose it can't hurt.