Enropy in groupby failed when no discrete probabilities provided

[s-b90]

Polars version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of Polars.

Issue description

When filtering a colum in agg expression and after filtering we have 0 values. Entropy calculation fails.
Note if we do that without groupby, just for series, everything will be ok.

Reproducible example

import polars as pl

pl.DataFrame({'col_to_groupby': [2], 'time': [1672740910.967138],  'col3': [1]}).groupby('col_to_groupby').agg(
    (
        (pl.col('time').filter(pl.col('col3') == 0).unique_counts() /
         pl.col('time').filter(pl.col('col3') == 0).len()).entropy()

    ).alias('calc')
)
exceptions.ComputeError: returned aggregation is a different length: 0 than the group lengths: 1

Expected behavior

This computation should returns

col_to_groupby	calc
i64	f32
2	null

Installed versions

---Version info---
Polars: 0.15.11
Index type: UInt32
Platform: Linux-5.15.0-56-generic-x86_64-with-glibc2.35
Python: 3.8.16 (default, Dec  7 2022, 01:12:06) 
[GCC 11.3.0]
---Optional dependencies---
pyarrow: 10.0.1
pandas: <not installed>
numpy: 1.22.3
fsspec: 2022.8.2
connectorx: <not installed>
xlsx2csv: <not installed>
matplotlib: <not installed>
None

[ritchie46]

I can run this snippet in master after 0.15.15. It returns this:

shape: (1, 2)
┌────────────────┬──────┐
│ col_to_groupby ┆ calc │
│ ---            ┆ ---  │
│ i64            ┆ f64  │
╞════════════════╪══════╡
│ 2              ┆ -0.0 │
└────────────────┴──────┘

I only don't understand what you'd expect from entropy(len).

[s-b90]

Seems that it was fixed by #6036, thank you. I will close this.

I only don't understand what you'd expect from entropy(len).

It was entropy(unique_counts / len) where unique_counts / len are pk - discrete probabilities.
For some reasons, entropy calculates correctly only if values in series are prepared discrete probabilities. So we need to calculate pk before entropy.