BUG: `DataFrame.var()` with `axis=1` returns `object` Series instead of `float64` Series?

[itholic]

Pandas version checks

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

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

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

import pandas as pd

pdf = pd.DataFrame(
    {
        "A": [1, -2, 3, -4, 5] * 300,
        "B": [1.0, -2, 3, -4, 5] * 300,
        "C": [-6.0, -7, -8, -9, 10] * 300,
        "D": [True, False, True, False, False] * 300,
    },
    index=range(10, 15001, 10),
)
pdf.var(axis=1)

Issue Description

DataFrame.var() with axis=1 returns float64 Series in Pandas 1.5.3, but now it returns object Series instead. So I wonder if it is an intended behavior or an unexpected bug:

Pandas 1.5.3

>>> pdf = pd.DataFrame(
...     {
...         "A": [1, -2, 3, -4, 5] * 300,
...         "B": [1.0, -2, 3, -4, 5] * 300,
...         "C": [-6.0, -7, -8, -9, 10] * 300,
...         "D": [True, False, True, False, False] * 300,
...     },
...     index=range(10, 15001, 10),
... )
>>> pdf.var(axis=1)
10       12.250000
20        8.916667
30       27.583333
40       13.583333
50       16.666667
           ...
14960    12.250000
14970     8.916667
14980    27.583333
14990    13.583333
15000    16.666667
Length: 1500, dtype: float64

Pandas 2.0.0 and above

>>> pdf = pd.DataFrame(
...     {
...         "A": [1, -2, 3, -4, 5] * 300,
...         "B": [1.0, -2, 3, -4, 5] * 300,
...         "C": [-6.0, -7, -8, -9, 10] * 300,
...         "D": [True, False, True, False, False] * 300,
...     },
...     index=range(10, 15001, 10),
... )
>>> pdf.var(axis=1)
10           12.25
20        8.916667
30       27.583333
40       13.583333
50       16.666667
           ...
14960        12.25
14970     8.916667
14980    27.583333
14990    13.583333
15000    16.666667
Length: 1500, dtype: object

Expected Behavior

>>> pdf = pd.DataFrame(
...     {
...         "A": [1, -2, 3, -4, 5] * 300,
...         "B": [1.0, -2, 3, -4, 5] * 300,
...         "C": [-6.0, -7, -8, -9, 10] * 300,
...         "D": [True, False, True, False, False] * 300,
...     },
...     index=range(10, 15001, 10),
... )
>>> pdf.var(axis=1)
10       12.250000
20        8.916667
30       27.583333
40       13.583333
50       16.666667
           ...
14960    12.250000
14970     8.916667
14980    27.583333
14990    13.583333
15000    16.666667
Length: 1500, dtype: float64

Installed Versions

INSTALLED VERSIONS

commit : ba1cccd
python : 3.9.17.final.0
python-bits : 64
OS : Darwin
OS-release : 22.6.0
Version : Darwin Kernel Version 22.6.0: Wed Jul 5 22:21:53 PDT 2023; root:xnu-8796.141.3~6/RELEASE_ARM64_T6020
machine : arm64
processor : arm
byteorder : little
LC_ALL : None
LANG : None
LOCALE : None.UTF-8

pandas : 2.1.0
numpy : 1.26.0
pytz : 2023.3
dateutil : 2.8.2
setuptools : 68.0.0
pip : 23.2.1
Cython : None
pytest : 7.4.0
hypothesis : None
sphinx : 3.0.4
blosc : None
feather : None
xlsxwriter : None
lxml.etree : 4.9.3
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 2.11.3
IPython : 8.15.0
pandas_datareader : None
bs4 : 4.12.2
bottleneck : None
dataframe-api-compat: None
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : 3.8.0
numba : None
numexpr : None
odfpy : None
openpyxl : 3.1.2
pandas_gbq : None
pyarrow : 13.0.0
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : 1.11.2
sqlalchemy : 2.0.20
tables : None
tabulate : 0.9.0
xarray : None
xlrd : None
zstandard : None
tzdata : 2023.3
qtpy : None
pyqt5 : None

[rhshadrach]

Thanks for the report! Confirmed on main. To compute axis=1 in this case, we take a transpose and end up with object dtype. It would be better for both behavior and performance to change all columns to float dtype and then transpose. But I wonder: is this always safe?

Also, if we pursue _reduce_axis1 as @jbrockmendel has proposed, I don't think the different dtypes would be an issue.

Further investigations and PRs to fix are welcome!

[itholic]

It would be better for both behavior and performance to change all columns to float dtype and then transpose. But I wonder: is this always safe?

Thanks for the suggestion! I can't think of any other dangerous scenarios at this time :-)

Also, if we pursue _reduce_axis1 as @jbrockmendel has proposed, I don't think the different dtypes would be an issue.

Interesting! Can I see any link to a related discussion or ticket?

[jbrockmendel]

Interesting! Can I see any link to a related discussion or ticket?

Not really. Just an informal goal to use _reduce_axis1 to handle more cases without transposing.