Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

BUG: .corr() values significantly higher than 1 #61120

Closed
2 of 3 tasks
krassowski opened this issue Mar 14, 2025 · 3 comments · Fixed by #61154
Closed
2 of 3 tasks

BUG: .corr() values significantly higher than 1 #61120

krassowski opened this issue Mar 14, 2025 · 3 comments · Fixed by #61154
Assignees
@j-hendricks
Labels
Bug cov/corr

Comments

@krassowski
Copy link

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

data = pd.DataFrame(dict(
    x=[0, 1],
    y=[1.35951, 1.3595100000000007]
))
data.corr().max().max()

Issue Description

The example above results in 1.1547005383792517.

This is similar to #35135 which was closed as "not an issue with pandas, but just numerical computations" but differently to that issue which showed minuscule differences (practically negligible), I am presenting an example where the Pearson correlation is over 15% above the maximum of 1.

I was able to reproduce this on multiple machines.

I think this might warrant a mention in the documentation.

Expected Behavior

These two are perfectly correlated, so we would expect 1. 1 is the result for both:

  • (data + 0.0000000000000002).corr().max().max()
  • (data - 0.0000000000000002).corr().max().max()

Interestingly, using corrwith or R leads to a different result which under-estimates the correlation (but at least is not out of range, and the relative error is smaller!):

  • data[['x']].corrwith(data['y']) returns 0.948683
  • cor in R also returns 0.9486833
cor(c(0, 1), c(1.35951, 1.3595100000000007))
[1] 0.9486833

Installed Versions

INSTALLED VERSIONS
------------------
commit                : 0691c5cf90477d3503834d983f69350f250a6ff7
python                : 3.11.4
python-bits           : 64
OS                    : Linux
OS-release            : 6.8.0-55-generic
Version               : #57-Ubuntu SMP PREEMPT_DYNAMIC Wed Feb 12 23:42:21 UTC 2025
machine               : x86_64
processor             : x86_64
byteorder             : little
LC_ALL                : None
LANG                  : en_GB.UTF-8
LOCALE                : en_GB.UTF-8

pandas                : 2.2.3
numpy                 : 1.26.3
pytz                  : 2023.3.post1
dateutil              : 2.8.2
pip                   : 23.1.2
Cython                : None
sphinx                : None
IPython               : 9.0.0.dev
adbc-driver-postgresql: None
adbc-driver-sqlite    : None
bs4                   : 4.12.3
blosc                 : None
bottleneck            : None
dataframe-api-compat  : None
fastparquet           : None
fsspec                : None
html5lib              : None
hypothesis            : None
gcsfs                 : None
jinja2                : 3.1.3
lxml.etree            : None
matplotlib            : 3.10.0
numba                 : None
numexpr               : None
odfpy                 : None
openpyxl              : None
pandas_gbq            : None
psycopg2              : None
pymysql               : None
pyarrow               : 15.0.0
pyreadstat            : None
pytest                : 7.4.4
python-calamine       : None
pyxlsb                : None
s3fs                  : None
scipy                 : None
sqlalchemy            : None
tables                : None
tabulate              : None
xarray                : None
xlrd                  : None
xlsxwriter            : None
zstandard             : None
tzdata                : 2023.4
qtpy                  : None
pyqt5                 : None
@krassowski krassowski added Bug Needs Triage Issue that has not been reviewed by a pandas team member labels Mar 14, 2025
@krassowski
Copy link
Author

krassowski commented Mar 14, 2025 •
edited
Loading

Note that numpy.corrcoef clips the values into the correct range

Due to floating point rounding the resulting array may not be Hermitian, the diagonal elements may not be 1, and the elements may not satisfy the inequality abs(a) <= 1. The real and imaginary parts are clipped to the interval [-1, 1] in an attempt to improve on that situation but is not much help in the complex case.

Image

@rhshadrach
Copy link
Member

Thanks for the report. It makes sense to me to clip the values here, in alignment with NumPy. However, I think we should not promise full compatibility with NumPy (same output on the same input).

PRs are welcome!

@rhshadrach rhshadrach added cov/corr and removed Needs Triage Issue that has not been reviewed by a pandas team member labels Mar 19, 2025
@j-hendricks
Copy link
Contributor

take

@j-hendricks j-hendricks mentioned this issue Mar 20, 2025
Merged
5 tasks
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@j-hendricks j-hendricks

Labels
Bug cov/corr
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

BUG: Clip corr edge cases between -1.0 and 1.0 j-hendricks/pandas
3 participants
@krassowski @rhshadrach @j-hendricks