(Easy Nearest Neighbor Estimation of Mutual Information)
This Python 3 package provides simple methods for estimating mutual information of continuous variables. With one method call, you can estimate several variable pairs and time lags at once. Both unconditional and conditional MI (including multivariate condition) are supported. The interface is aimed at non-linear correlation analysis.
The package uses the nearest neighbor methods decscribed in:
- Kraskov et al. (2004): Estimating mutual information. Physical Review E 69. doi:10.1103/PhysRevE.69.066138.
- Frenzel & Pompe (2007): Partial Mutual Information for Coupling Analysis of Multivariate Time Series. Physical Review Letters 99. doi:10.1103/PhysRevLett.99.204101.
- Ross (2014): Mutual Information between Discrete and Continuous Data Sets. PLoS ONE 9. doi:10.1371/journal.pone.0087357.
NOTE: The package is beta quality. We are not planning major changes to the API, but breakages are still possible. The latest source code on GitHub might be more unstable than the released version on PyPI. You can see the roadmap of planned additions on the Milestones page.
This package requires Python 3.6 or higher, and it is tested to work on the latest versions of Ubuntu, macOS and Windows. The only hard dependencies are reasonably recent versions of NumPy and SciPy; Pandas is strongly suggested for more enjoyable data analysis.
This package is available on PyPI:
pip install ennemi
(If your machine has multiple Python installations, you may need to run e.g. pip3
.)
For documentation, please see https://polsys.github.io/ennemi.
The tests depend on pandas, so you need that installed in addition.
Additionally, pytest
and mypy
are required for building the project.
All of these are installed by the "extras" syntax of pip
.
To install the package in development mode, clone this repository and execute
pip install -e .[dev]
in the repository root folder.
All methods, including tests, are type annotated and checked with mypy
.
The CI script runs the check automatically on each pushed commit.
To run the check yourself, execute
python -m mypy ennemi/ tests/unit tests/integration
in the repository root (configuration is stored in mypy.ini
file).
Please see also the contribution guidelines.
This work is licensed under the MIT license.
In short, you are allowed to use, modify and distribute it as you wish, as long as
the original copyright notice is preserved.
This software is provided "as is", functioning to the extent of passing
the unit and integration tests in the tests
directory.
This package is archived on Zenodo.
The DOI 10.5281/zenodo.3834018
always resolves to the latest version of ennemi
.
For reproducibility, you should cite the exact version of the package you have used.
To do so, use the DOI given on the Releases page or on Zenodo.
In the future, a journal article may become the canonical reference for this package. You should still also cite the package version you have used.
This package is maintained by Petri Laarne at
Institute for Atmospheric and Earth System Research (INAR),
University of Helsinki.
Please feel free to contact me at [email protected]
about any questions or suggestions related to this project.