HUME

HUME (High-dimensional Undirected Graphical Models for arbitrary Mixed data) is an R package for learning sparse graphical models from high-dimensional data with mixed variable types.

Overview

Graphical models are fundamental tools for understanding relationships between variables in multivariate data. While methods for homogeneous data (all continuous or all discrete) are well-developed, principled approaches for mixed data types remain challenging.

Capabilities

• Learn conditional independence structure from mixed-type data
• High-dimensional support for many variables
• Sparsity-inducing estimation methods
• Flexible handling of heterogeneous variable types
• Robust statistical guarantees

Use Cases

• Identifying meaningful relationships in complex datasets
• Feature selection and variable importance
• Network discovery in multimodal data
• Exploratory data analysis with mixed measurements

Installation

# From GitHub
devtools::install_github("konstantingoe/hume")

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.