randomcov

Generating random covariance and correlation matrices with beautiful geodesic interpolation visualizations! 🎯

Install

pip install randomcov 

or for latest

pip install git+https://github.com/microprediction/randomcov.git

Example

from randomcov import random_covariance_matrix
cov = random_covariance_matrix(n=50, corr_method='residuals', var_method='lognormal')

Motivation

To collect standard but also novel correlation and covariance generation methods, in order to better understand when some estimation methods work better than others in different contexts: such as the construction of machine learning model ensembles, combinations of forecasts, or financial portfolios.

The geodesic interpolation capabilities enable smooth transformations between covariance structures while preserving mathematical properties, making it ideal for portfolio optimization and risk management applications.

Generator catalog

Every method: random_correlation_matrix(n, corr_method="<name>", corr_kwargs={...}). All generators accept rng= (int seed or numpy Generator) for reproducibility.

There is no canonical "uniform" measure over covariance matrices, and the elliptope-uniform (LKJ eta=1) concentrates on weak correlations as n grows; robust conclusions require batteries across NAMED ensembles, which is what this catalog is for.

Correlation Inflation

An example of the kind of thing I wish to test against generative models.

Correlation Inflation: A Working Paper

The transformation preserves geometric properties while smoothly interpolating towards perfect correlation structure using differential geometry on the manifold of positive definite matrices.