Skip to content
/ korr Public

collection of utility functions for correlation analysis

License

Notifications You must be signed in to change notification settings

kmedian/korr

Folders and files

NameName
Last commit message
Last commit date

Latest commit

c79b38f · Jun 19, 2022

History

44 Commits
Jun 19, 2022
Apr 8, 2021
Jun 19, 2022
Jun 19, 2022
Jun 19, 2022
Apr 23, 2021
Apr 9, 2021
Apr 27, 2021
Jun 19, 2022
Jun 17, 2022
Apr 23, 2021
Jun 19, 2022
Jun 19, 2022

Repository files navigation

PyPI version Total alerts Language grade: Python

korr

collection of utility functions for correlation analysis

Usage

Check the examples folder for notebooks.

Compute correlation matrix and its p-values

  • pearson -- Pearson/Sample correlation (interval- and ratio-scale data)
  • kendall -- Kendall's tau rank correlation (ordinal data)
  • spearman -- Spearman rho rank correlation (ordinal data)
  • mcc -- Matthews correlation coefficient between binary variables

EDA, Dig deeper into results

  • flatten -- A table (pandas) with one row for each correlation pairs with the variable indicies, corr., p-value. For example, try to find "good" cutoffs with corr_vs_pval and then look up the variable indicies with flatten afterwards.
  • slice_yx -- slice a correlation and p-value matrix of a (y,X) dataset into a (y,x_i) vector and (x_j, x_k) matrices
  • corr_vs_pval -- Histogram to find p-value cutoffs (alpha) for a) highly correlated pairs, b) unrelated pairs, c) the mixed results.
  • bracket_pval -- Histogram with more fine-grained p-value brackets.
  • corrgram -- Correlogram, heatmap of correlations with p-values in brackets

Utility functions

  • confusion -- Confusion matrix. Required for Matthews correlation (mcc) and is a bitter faster than sklearn's

Parameter Stability

  • bootcorr -- Estimate multiple correlation matrices based on bootstrapped samples. From there you can assess how stable correlation estimates are (how sensitive against in-sample variation). For example, stable estimates are good candidates for modeling, and unstable correlation pairs are good candidates for P-hacking and non-reproducibility.

Variable Selection, Search Functions

  • mincorr -- From all estimated correlation pairs, pick a given n=3,5,.. of variables with low and insignificant correlations among each other. (See binsel package for an application.)
  • find_best -- Find the N "best", i.e. high and most significant, correlations
  • find_worst -- Find the N "worst", i.e. insignificant/random and low, correlations
  • find_unrelated -- Return variable indicies of unrelated pairs (in terms of insignificant p-value)

Appendix

Installation

The korr git repo is available as PyPi package

pip install korr

Install a virtual environment

python3.7 -m venv .venv
source .venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt --no-cache-dir
pip install -r requirements-dev.txt --no-cache-dir
pip install -r requirements-demo.txt --no-cache-dir

(If your git repo is stored in a folder with whitespaces, then don't use the subfolder .venv. Use an absolute path without whitespaces.)

Commands

  • Check syntax: flake8 --ignore=F401
  • Run Unit Tests: pytest
  • Remove .pyc files: find . -type f -name "*.pyc" | xargs rm
  • Remove __pycache__ folders: find . -type d -name "__pycache__" | xargs rm -rf

Publish

pandoc README.md --from markdown --to rst -s -o README.rst
python setup.py sdist 
twine upload -r pypi dist/*

Support

Please open an issue for support.

Contributing

Please contribute using Github Flow. Create a branch, add commits, and open a pull request.