BUG: std using numpy.float32 dtype gives incorrect result on contant array.

[jamespreed]

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

# this should be 0 or very close to 0
print(pd.Series([271.46] * 150000, dtype='float32').std())
# returns: 0.229433074593544

# same result using numpy dtype directly (as you would expect)
print(pd.Series([271.46] * 150000, dtype=np.float32).std())
# returns: 0.229433074593544


# the pandas float32 dtype does not have this issue.
print(pd.Series([271.46] * 150000, dtype=pd.Float32Dtype()).std())
# returns: 0.0

# neither does using the numpy standard deviation function on the numpy values
print(np.std(pd.Series([271.46] * 150000, dtype=np.float32).values))
# returns: 0.0

Issue Description

When using the numpy float32 dtype by passing either the string 'float32' or the numpy dtype np.float32, the value of the standard deviation is incorrect for a constant array (array of identical values). It should return zero, but instead returns a value.

Switching to using the Pandas float32 dtype alleviates the error, as does using np.std on the numpy values of the series.

Expected Behavior

The expected behavior is the following evaluating to True

pd.Series([271.46] * 150000, dtype='float32').std() == 0.0

Installed Versions

INSTALLED VERSIONS
------------------
commit                : f538741432edf55c6b9fb5d0d496d2dd1d7c2457
python                : 3.11.7.final.0
python-bits           : 64
OS                    : Windows
OS-release            : 10
Version               : 10.0.22621
machine               : AMD64
processor             : Intel64 Family 6 Model 151 Stepping 5, GenuineIntel
byteorder             : little
LC_ALL                : None
LANG                  : None
LOCALE                : English_United States.1252

pandas                : 2.2.0
numpy                 : 1.26.3
pytz                  : 2023.3.post1
dateutil              : 2.8.2
setuptools            : 68.2.2
pip                   : 23.3.1
Cython                : None
pytest                : None
hypothesis            : None
sphinx                : None
blosc                 : None
feather               : None
xlsxwriter            : None
lxml.etree            : 4.9.3
html5lib              : None
pymysql               : None
psycopg2              : None
jinja2                : None
IPython               : 8.15.0
pandas_datareader     : None
adbc-driver-postgresql: None
adbc-driver-sqlite    : None
bs4                   : None
bottleneck            : 1.3.7
dataframe-api-compat  : None
fastparquet           : None
fsspec                : None
gcsfs                 : None
matplotlib            : 3.8.0
numba                 : None
numexpr               : 2.8.7
odfpy                 : None
openpyxl              : None
pandas_gbq            : None
pyarrow               : None
pyreadstat            : None
python-calamine       : None
pyxlsb                : None
s3fs                  : None
scipy                 : 1.11.4
sqlalchemy            : None
tables                : None
tabulate              : None
xarray                : None
xlrd                  : None
zstandard             : None
tzdata                : 2023.3
qtpy                  : None
pyqt5                 : None

[Groni3000]

Have no problems:

OS: Windows 10
Python: 3.11.4
Working in venv with such dependencies:

contourpy==1.2.0
cycler==0.12.1
fonttools==4.49.0
kiwisolver==1.4.5
matplotlib==3.8.3
numpy==1.26.4
packaging==23.2
pandas==2.2.0
pillow==10.2.0
pyarrow==15.0.0
pyparsing==3.1.1
python-dateutil==2.8.2
pytz==2024.1
seaborn==0.13.2
six==1.16.0
tzdata==2024.1

Sorry, don't know how to make such "Installed Versions".

Tried the same code with 1.26.3 numpy version, still no problems.

[jamespreed]

Updated numpy to 1.26.4 with the same result. Is this a "just on my machine" bug?

[jamespreed]

Sorry, don't know how to make such "Installed Versions".

You can use:

pd.show_versions()

[Groni3000]

@jamespreed Thank you very much! Here is my installed versions:

INSTALLED VERSIONS
------------------
commit                : f538741432edf55c6b9fb5d0d496d2dd1d7c2457
python                : 3.11.4.final.0
python-bits           : 64
OS                    : Windows
OS-release            : 10
Version               : 10.0.19045
machine               : AMD64
processor             : AMD64 Family 23 Model 113 Stepping 0, AuthenticAMD
byteorder             : little
LC_ALL                : None
LANG                  : en_US.UTF-8
LOCALE                : Russian_Ukraine.1251

pandas                : 2.2.0
numpy                 : 1.26.4
pytz                  : 2024.1
dateutil              : 2.8.2
setuptools            : 65.5.0
pip                   : 24.0
Cython                : None
pytest                : 8.0.1
hypothesis            : None
sphinx                : None
blosc                 : None
feather               : None
xlsxwriter            : None
lxml.etree            : None
html5lib              : None
pymysql               : None
psycopg2              : None
jinja2                : None
IPython               : None
pandas_datareader     : None
adbc-driver-postgresql: None
adbc-driver-sqlite    : None
bs4                   : None
bottleneck            : None
dataframe-api-compat  : None
fastparquet           : None
fsspec                : None
gcsfs                 : None
matplotlib            : 3.8.3
numba                 : None
numexpr               : None
odfpy                 : None
openpyxl              : None
pandas_gbq            : None
pyarrow               : 15.0.0
pyreadstat            : None
python-calamine       : None
pyxlsb                : None
s3fs                  : None
scipy                 : None
sqlalchemy            : None
tables                : None
tabulate              : None
xarray                : None
xlrd                  : None
zstandard             : None
tzdata                : 2024.1
qtpy                  : None
pyqt5                 : None

[Groni3000]

@jamespreed Just out of curious, can you try to create venv and reproduce this bug?
Make sure you are using same python version as described in bug report.

python -m venv PROJECT_FOLDER & cd PROJECT_FOLDER & .\Scripts\activate & pip install numpy pandas

Create your file and try again.

[jamespreed]

Ok, that is interesting. The pip installed version gives the same outputs that as you are seeing:

The outputs I am originally seeing are from the conda version of the packages.

[Groni3000]

Ye, really interesting: package manager affects math results ʘ‿ʘ At least you know how to use std with reliable results...

[phofl]

Are you able to isolate whether it's the pandas or NumPy package that causes this?

[asishm]

The outputs I am originally seeing are from the conda version of the packages.

Can you provide your conda env export file? (or steps to create the env using conda)

[jamespreed]

I am extremely confused now. The issue does NOT appear when clean installing via the following methods:

venv

mkdir test
python -m venv test
cd test
./Scripts/activate
pip install pandas

conda env with numpy, pip install pandas

conda create -n test2 python=3.11 numpy
conda activate test2
pip install conda-forge::pandas==2.2.0 -y

clean conda env

conda create -n test3 python=3.11 conda-forge::pandas==2.2.0 -y

[jamespreed]

The issue is bottleneck. Whenever I install it into an environment, the float32 issue appears.

In all of the above cases, installing bottleneck using either pip or conda exhibits the problem.

conda activate test2
pip install bottleneck
# issue appears again.

[phofl]

is this only on the new pandas version or does this also happen if you use an older version of pandas?

[jamespreed]

It happens on Pandas 1.4.4, 1.5.3, 2.0.0, 2.1.0, 2.1.4, and 2.2.0

[phofl]

that means that it is most likely a bottleneck issue, could you report there?

[rhshadrach]

Looking at the linked issue, it appears very likely this is a bottleneck issue. Closing for now.

[rhshadrach]

Also, note you can disable the use of bottleneck with

pd.set_option("compute.use_bottleneck", False)

You can also temporarily disable it for certain lines:

with pd.option_context("compute.use_bottleneck", False):
    print(pd.Series([271.46] * 150000, dtype='float32').std())