`pivot()` doesn't distinguish between `count` and `len` about polars HOT 3 OPEN

Yeah, it seems the warning only triggers for the "passing function by string name" case.

This doesn't warn - but I'm unsure if this use of count is deprecated also?

aggregate_function = pl.element().count()

Would the non-null len() need to be used instead?

aggregate_function = pl.element().drop_nulls().len()

from polars.

Before we rename pl.count to pl.len: pl.count() means the length of df, including the null value. But pl.count("a") equals to pl.col("a").count(), this doesn't include null value. This is sort of odd, so we rename the no-argument pl.count to pl.len.

If I remember correctly, aggregate_function='count' count the null value even when it is not deprecated(Because it actually uses pl.count() instead of pl.element().count()). You can manually use aggregate_function = pl.element().count() to achieve the behavior you expected. Of course, you could propose that aggregate_function=count shouldn't be deprecated and it's behave should be fixed as the same as pl.element().count().

from polars.

Yes, I propose this indeed, even if the workaround pl.element().count() works perfectly fine. I've seen #13673, but IMO the deprecation is not justified in this case. I've never come across a use case for including the nulls in a count in the context of a pivot, so if anything, the pl.element().count() behaviour should have priority over what len does.

If there is no support for un-deprecating count and implementing it as pl.element().count(), then I think the deprecation message should at least be changed so that it suggests using pl.element().count() if one is looking for a way to exclude the nulls.

from polars.