REGR: incorrect results with std on rolling window since 1.2.0

[cragod]

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

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

• (optional) I have confirmed this bug exists on the master branch of pandas.

---

Code Sample

import pandas as pd
import numpy as np

df = pd.DataFrame({'A': [1e-8, -1.1e-8, 1.2e-8, -1.1e-8] * 5})
df['std'] = df['A'].rolling(10).std(ddof=0)
print(df)

Problem description

when i ran above code in 1.2.0, the output is:

               A               std
0   1.000000e-08  NaN
1  -1.100000e-08  NaN
2   1.200000e-08  NaN
3  -1.100000e-08  NaN
4   1.000000e-08  NaN
5  -1.100000e-08  NaN
6   1.200000e-08  NaN
7  -1.100000e-08  NaN
8   1.000000e-08  NaN
9  -1.100000e-08  0.0
10  1.200000e-08  0.0
11 -1.100000e-08  0.0
12  1.000000e-08  0.0
13 -1.100000e-08  0.0
14  1.200000e-08  0.0
15 -1.100000e-08  0.0
16  1.000000e-08  0.0
17 -1.100000e-08  0.0
18  1.200000e-08  0.0
19 -1.100000e-08  0.0

Expected Output

ran in 1.1.5:

               A           std
0   1.000000e-08           NaN
1  -1.100000e-08           NaN
2   1.200000e-08           NaN
3  -1.100000e-08           NaN
4   1.000000e-08           NaN
5  -1.100000e-08           NaN
6   1.200000e-08           NaN
7  -1.100000e-08           NaN
8   1.000000e-08           NaN
9  -1.100000e-08  1.092200e-08
10  1.200000e-08  1.112160e-08
11 -1.100000e-08  1.112160e-08
12  1.000000e-08  1.092200e-08
13 -1.100000e-08  1.092200e-08
14  1.200000e-08  1.112160e-08
15 -1.100000e-08  1.112160e-08
16  1.000000e-08  1.092200e-08
17 -1.100000e-08  1.092200e-08
18  1.200000e-08  1.112160e-08
19 -1.100000e-08  1.112160e-08

Output of pd.show_versions()

INSTALLED VERSIONS

commit : 3e89b4c
python : 3.7.5.final.0
python-bits : 64
OS : Darwin
OS-release : 20.2.0
Version : Darwin Kernel Version 20.2.0: Wed Dec 2 20:39:59 PST 2020; root:xnu-7195.60.75~1/RELEASE_X86_64
machine : x86_64
processor : i386
byteorder : little
LC_ALL : None
LANG : zh_CN.UTF-8
LOCALE : zh_CN.UTF-8

pandas : 1.2.0
numpy : 1.18.2
pytz : 2019.3
dateutil : 2.8.1
pip : 20.3.3
setuptools : 41.2.0
Cython : None
pytest : 5.4.1
hypothesis : None
sphinx : 1.8.5
blosc : None
feather : None
xlsxwriter : None
lxml.etree : 4.5.2
html5lib : None
pymysql : 0.9.3
psycopg2 : None
jinja2 : 2.10.3
IPython : 7.11.1
pandas_datareader: None
bs4 : None
bottleneck : None
fsspec : None
fastparquet : None
gcsfs : None
matplotlib : 3.3.1
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyxlsb : None
s3fs : None
scipy : None
sqlalchemy : 1.3.15
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
numba : None

[simonjayhawkins]

Thanks @cragod for the report.

first bad commit: [846cff9] REF: Remove rolling window fixed algorithms (#36567) cc @mroeschke

[phofl]

Isn't the output in 1.1.5 wrong too?

Edit; Forget it, was running var not std...

[phofl]

This has the same underlying cause as mentioned in #37051 and non trivial to fix

[mroeschke]

Thanks for the report @cragod.

As @phofl mentioned, the cause of this is #37051. Let's track this issue in that thread, so going to close this as a duplicate. Happy to have eyes on that issue!

[simonjayhawkins]

This has the same underlying cause as mentioned in #37051 and non trivial to fix

"non trivial" sounds to me alot like not suitable for backport.

re-opening this issue for now for discussion on how to resolve the issue on the 1.2.x branch.

[jreback]

we cannot handle this as a backport and its non-trivial to fix this. that said this is an edge case so won't fix for 1.2.x