Skip to main content
Michael Ellis
← Back to open source projects

R · Open source

dirichlet-process-mixtures

Bayesian Nonparametric Clustering With Gibbs Sampling

An R implementation of truncated Dirichlet process Gaussian mixtures with Gibbs sampling and automatic cluster discovery.

Overview

This is a compact implementation of a truncated stick-breaking Dirichlet process Gaussian mixture model. Conjugate Gibbs updates estimate the component parameters and mixture weights, while the posterior can leave unnecessary components empty.

The repository includes a reproducible simulation and MCMC trace plots so each part of the sampler can be inspected.

Technical highlights

  • Stick-breaking mixture weights
  • Conjugate Gibbs updates
  • Automatic cluster discovery
  • MCMC trace diagnostics and reproducible simulation

Topics

Dirichlet processesGibbs samplingBayesian clusteringMixture modelsR