Optional axis parameter for pl.any , pl.Expr.any, pl.all, pl.Expr.all

[lucazanna]

Problem description

I just read a Stack Overflow answer which shows a concise way to verify equality in Polars:
https://stackoverflow.com/questions/76092263/column-and-row-wise-logical-operations-on-polars-dataframe

I like that the code is short and easy to write:

# Verify which columns contain a specific value
df.select((pl.all() == "?").any())

# Verify which rows contain a specific value
df.select(pl.any(pl.all() == "?"))

While I like the compactness of it, I feel that at first read is not very clear that pl.Expr.any does computations column-wise while pl.any does computations row-wise.
I believe the clarity of Polars expressions is one of the strengths of the library.

My proposal for this:

• is it possible to add an optional 'axis' parameter to pl.any, pl.Expr.any, pl.all, pl.Expr.all that allows to specify if the computation should be column-wise or row-wise?
• in addition to it, what about aligning these functions to run column-wise by default?
  I believe having pl.any work the same way as pl.Expr.any would be good for the predictability of the language

What are your thoughts?

[gab23r]

If you do that I am scared that you need to do that for each method: min, max, sum, mean...

[cmdlineluser]

Perhaps there should be an .arr.all() so you could write it similar to the other "row-wise" computations?

It would be slightly more verbose:

df.select(pl.concat_list(pl.all() == "?").arr.all())

[lucazanna]

Fair point @gab23r

i like @cmdlineluser ´s idea of concat_list with .arr.any or .arr.all for row-wise computations

it would be more verbose that what exists today, but it would follow Polars logic and be easy to understand

and the existing pl.any , pl.Expr.any, etc could work column-wise

This can also fit into Polars logic that fast code should be easy to write.
So column-wise code can be easier to write

if someone is doing lots of row-wise computations, maybe the dataframe should be reworked / transposed / melted

[cmdlineluser]

Just browsing old issues, I think .all_horizontal / .any_horizontal #9752 resolves this.

[lucazanna]

@cmdlineluser yes it does. closing this