humlpy

mixed-gm provides tools for learning undirected graphical models when data contains variables of different types. This addresses the common challenge in real-world applications where measurements span continuous, discrete, binary, ordinal, and count variables.

Key Features

• Handles arbitrary mixed variable types
• Latent Gaussian copula framework
• Scalable high-dimensional methods
• Leverages polychoric and polyserial correlations
• Both theoretical and empirical validation

Supported Data Types

• Continuous variables
• Binary/categorical variables
• Count data
• Ordinal variables
• Mixed combinations

Methods

The package implements flexible and scalable methodology building on classical ideas of polychoric and polyserial correlations within a latent Gaussian copula framework, enabling principled joint analysis of mixed data.

Citation

Göbler, K., Drton, M., Mukherjee, S., & Miloschewski, A. (2024). High-dimensional undirected graphical models for arbitrary mixed data. Electronic Journal of Statistics, 18(1), 2339-2404.