BUG: roll_skew and roll_kurt compute different result when input same data with diffrent length

[cutebomb]

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
import numpy as np

a = np.random.random(20)*1000
a1 = a[:10]
a2 = a[:12]
df = pd.Series(a1)
df2 = pd.Series(a2)
r1 = df.rolling(8, min_periods=1).skew()
r2 = df2.rolling(8, min_periods=1).skew()
print(np.where(r1.values[:10] == r2.values[:10]))

r1 = df.rolling(8, min_periods=1).kurt()
r2 = df2.rolling(8, min_periods=1).kurt()
print(np.where(r1.values[:10] == r2.values[:10]))

Issue Description

rolling skew and kurt compute different result when input same data with diffrent length.
With the rolling window of exactly same data, I think it is wired to have different result right?
It is basically caused by minus mean value from the series, in source tree pandas/_libs/window/aggregations.pyx:621-624 and same code exist in roll_kurt too.

Expected Behavior

rolling skew and kurt result should be same.

Installed Versions

INSTALLED VERSIONS

commit : 0f43794
python : 3.10.7.final.0
python-bits : 64
OS : Linux
OS-release : 5.19.0-46-generic
Version : #47-Ubuntu SMP PREEMPT_DYNAMIC Fri Jun 16 13:30:11 UTC 2023
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 2.0.3
numpy : 1.23.1
pytz : 2023.3
dateutil : 2.8.2
setuptools : 59.6.0
pip : 23.2.1
Cython : 0.29.34
pytest : 7.3.1
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : 3.1.2
lxml.etree : 4.9.2
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 3.1.2
IPython : 8.13.2
pandas_datareader: None
bs4 : 4.12.2
bottleneck : None
brotli : None
fastparquet : None
fsspec : 2023.5.0
gcsfs : None
matplotlib : 3.7.1
numba : 0.57.0
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : 12.0.0
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : 1.10.1
snappy : None
sqlalchemy : 2.0.16
tables : None
tabulate : 0.9.0
xarray : None
xlrd : None
zstandard : None
tzdata : 2023.3
qtpy : None
pyqt5 : None

[phofl]

Hi, thanks for your report. Are you sure that this is not caused by a loss of precision? Your values are quite large for these algorithms

[cutebomb]

I don't think it is caused by loss of precision, and value (-1000, 1000) is not very large value,is it?
The rolling algorithm nature should keep the old (hitorical) value stable with new values.
my temp fix to this problem is this commit: (by removing value minus mean code) cutebomb@c5391d4

Hi, thanks for your report. Are you sure that this is not caused by a loss of precision? Your values are quite large for these algorithms

[MattDXB]

Hi, I am running into the same issue currently. Any ideas?

[cutebomb]

Hi, I am running into the same issue currently. Any ideas?

check my patch for temp solution :)

[MattDXB]

Yes, I saw it, thanks! The issue is that I'm in a company setting and I cannot recompile Cython code...