Maximum Likelihood Estimation in Multivariate Probability Distributions Using EM Algorithms

mvem is a Python package that provides maximum likelihood estimation methods for multivariate probability distributions using expectation–maximization (EM) algorithms. Additionally, it includes some functionality for fitting non-Gaussian multivariate mixture models. For fitting a wide range of univariate probability distributions, we refer to scipy.stats.

Currently included:

• normal (mvem.stats.multivariate_norm)
• skew normal (mvem.stats.multivariate_skewnorm)
• Student's t (mvem.stats.multivariate_t)
• normal-inverse Gaussian (mvem.stats.multivariate_norminvgauss)
• generalised hyperbolic skew Student's t (mvem.stats.multivariate_genskewt)
• generalised hyperbolic (mvem.stats.multivariate_genhyperbolic)
• hyperbolic (mvem.stats.multivariate_hyperbolic)
• variance-gamma (mvem.stats.multivariate_vargamma)

Multivariate mixture models currently included:

• skew normal (mvem.mixture.skewnorm)

Where to get it

The source code is currently hosted on GitHub at: https://github.com/krisskul/mvem

Binary installers for the latest released version are available at the Python Package Index (PyPI).

pip install mvem

Quickstart

Documentation and a few introductory notebooks are hosted at readthedocs.

from mvem.stats import multivariate_norminvgauss
# assume p-variate data x
params = multivariate_norminvgauss.fit(x)

Requirements

Tested on Python 3.6

numpy
scikit-learn
scipy>=1.6.0