R · Open source
VBMM
Variational Bayesian Gaussian Mixture Models in R
An R package for coordinate-ascent variational inference in Dirichlet process Gaussian mixture models and hierarchical mixtures.
Overview
VBMM implements coordinate-ascent variational inference for Gaussian mixture models with stick-breaking priors. It includes a Dirichlet process mixture and a hierarchical model where known groups share Gaussian components.
The repository contains simulation code, convergence checks, posterior plots, and the derivation of the variational updates and evidence lower bound used by the implementation.
Technical highlights
- Coordinate-ascent variational inference
- Dirichlet process Gaussian mixture models
- Hierarchical mixtures with shared components
- ELBO and posterior diagnostic plots
Topics
Variational inferenceBayesian nonparametricsGaussian mixturesClusteringR