Inconsistent implicit type casting when using max_horizontal and mean_horizontal

[NicolasMuellerQC]

Checks

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of Polars.

Reproducible example

import polars as pl

df = pl.DataFrame(data=dict(x=[1, 2, 3]))
print(
    df.with_columns(
        (1 - 0.5 * pl.col("x")).alias("y"),
        pl.max_horizontal(0, 1 - 0.5 * pl.col("x")).alias("z"),
        pl.max_horizontal(0.0, 1 - 0.5 * pl.col("x")).alias("w"),
        pl.max_horizontal(0, 1.0 - 0.5 * pl.col("x")).alias("r"),
    )
)

# Output:
# shape: (3, 5)
# ┌─────┬──────┬─────┬─────┬─────┐
# │ x   ┆ y    ┆ z   ┆ w   ┆ r   │
# │ --- ┆ ---  ┆ --- ┆ --- ┆ --- │
# │ i64 ┆ f64  ┆ i32 ┆ f64 ┆ f64 │
# ╞═════╪══════╪═════╪═════╪═════╡
# │ 1   ┆ 0.5  ┆ 1   ┆ 1.0 ┆ 0.5 │
# │ 2   ┆ 0.0  ┆ 0   ┆ 0.0 ┆ 0.0 │
# │ 3   ┆ -0.5 ┆ 0   ┆ 0.0 ┆ 0.0 │
# └─────┴──────┴─────┴─────┴─────┘


print(
    df.with_columns(
        (1 - 0.5 * pl.col("x")).alias("y"),
        pl.mean_horizontal(0, 1 - 0.5 * pl.col("x")).alias("z"),
        pl.mean_horizontal(0.0, 1 - 0.5 * pl.col("x")).alias("w"),
        pl.mean_horizontal(0, 1.0 - 0.5 * pl.col("x")).alias("r"),
    )
)
# Output:
# shape: (3, 5)
# ┌─────┬──────┬─────┬─────┬───────┐
# │ x   ┆ y    ┆ z   ┆ w   ┆ r     │
# │ --- ┆ ---  ┆ --- ┆ --- ┆ ---   │
# │ i64 ┆ f64  ┆ f64 ┆ f64 ┆ f64   │
# ╞═════╪══════╪═════╪═════╪═══════╡
# │ 1   ┆ 0.5  ┆ 0.5 ┆ 0.5 ┆ 0.25  │
# │ 2   ┆ 0.0  ┆ 0.0 ┆ 0.0 ┆ 0.0   │
# │ 3   ┆ -0.5 ┆ 0.0 ┆ 0.0 ┆ -0.25 │
# └─────┴──────┴─────┴─────┴───────┘

Log output

No output.

Issue description

The implicit type casting when using max_horizontal is inconsistent:

• The expression (1 - 0.5 * pl.col("x")).alias("y") on the df has type f64 and correctly calculates e.g. 1-0.5*1 = 0.5 in the first row.
• The expression pl.max_horizontal(0, 1 - 0.5 * pl.col("x")).alias("z") has type i32 even though the second expression has type f64 (by point one). I think operations involving i32 and f64 should be casted to f64 if there is any implicit casting.
• The expression pl.max_horizontal(0.0, 1 - 0.5 * pl.col("x")).alias("w") has type f64, but it looks like the second column was casted to i32 before the max aggregation, even though it has type f64 (by point one).
• The expression pl.max_horizontal(0, 1.0 - 0.5 * pl.col("x")).alias("r")has type f64 and the output is what I would expect.

The same problem applies to mean_horizontal.

Expected behavior

I would expect the columns "z", "w", "r" to be equal in both cases. The column "r" is correct.

Installed versions

--------Version info---------
Polars:               0.20.23
Index type:           UInt32
Platform:             macOS-14.4.1-arm64-arm-64bit
Python:               3.11.9 | packaged by conda-forge | (main, Apr 19 2024, 18:34:54) [Clang 16.0.6 ]

----Optional dependencies----
adbc_driver_manager:  <not installed>
cloudpickle:          2.2.1
connectorx:           0.3.2
deltalake:            <not installed>
fastexcel:            <not installed>
fsspec:               2024.3.1
gevent:               <not installed>
hvplot:               <not installed>
matplotlib:           3.8.4
nest_asyncio:         1.6.0
numpy:                1.26.4
openpyxl:             3.1.2
pandas:               2.2.2
pyarrow:              15.0.2
pydantic:             1.10.14
pyiceberg:            <not installed>
pyxlsb:               <not installed>
sqlalchemy:           2.0.29
xlsx2csv:             <not installed>
xlsxwriter:           3.1.9

[TobiasDummschat]

This seems to be fixed in 0.20.24. Now all columns added columns have type f64.

[NicolasMuellerQC]

Thanks, I can confirm that it's fixed in 0.20.24!