MCMC for Normalized Random Measure Mixture Models

Unix | CodeCov | License ---- | ------- | ------- | |

What's this about

This package is a Julia port of the Java code released written by Yee Whye Teh for the article MCMC for Normalized Random Measure Mixture Models. Do not hesitate to create pull requests for enhancements or to open an issue. In any case, you can write to me (Emile Mathieu) at: emile.mathieu-at-stats-dot-ox-dot-ac-dot-uk.

Installation and requirements

Requirements:

• Julia in [0.6.x]
• 64-bit architecture

In the Julia REPL:

Pkg.clone("https://github.com/BigBayes/BNPMix.jl")
using BNPMix

Algorithms implemented

• Marginalized Samplers:
  • Neal’s Algorithm 8 generalized
  • The Reuse algorithm
• Conditional Slice Sampler

Notes

• Conjugate version not implemented
• Only Normal-Gamma-Independent emission implemented
• Multidimensional data observation not implemented

Example

You can run the Reuse conditional sampler on the galaxy dataset modeled as a mixtures of Gaussian with a normalized generalized Gamma prior:

cd examples
julia galaxy.jl

Tests

To localy run the tests, run in Julia:

Pkg.test("BNPMix")

References

• Favaro, Teh, MCMC for Normalized Random Measure Mixture Models, 2013