TGMCMC

TGMCMC is a hybrid posterior inference algorithm for normalized random measure mixture models that combines MCMC sampler and tree-based discrete approximate algorithm. It builds binary trees representing the cluster structure of datasets via incremental BHC, and proposes new clustering results based on those trees and accept the proposals with Metropolis-Hastings algorithm. TGMCMC mixes much faster than conventional MCMC algorithms, yet is guaranteed to converge unlike incremental BHC.

Category Software 0 / 0
License GPL v2
Producer Admin
Package Regdate Nov 30, 2015 16:28 Package Downloaded 24
Latest version 1.0 Download
Downloaded Count 24
Recent Updates Nov 30, 2015 16:29 0 / 0

TGMCMC is a hybrid posterior inference algorithm for normalized random measure mixture models that combines MCMC sampler and tree-based discrete approximate algorithm. It builds binary trees representing the cluster structure of datasets via incremental BHC, and proposes new clustering results based on those trees and accept the proposals with Metropolis-Hastings algorithm. TGMCMC mixes much faster than conventional MCMC algorithms, yet is guaranteed to converge unlike incremental BHC.

 

The sourcecode is distributed by the following URL:

https://github.com/juho-lee/nrmm.cpp