Reversible-Jump-Markov-Chain-Monte-Carlo(RJMCMC) with Simulated Annealing

Introduction

This repo is about using Reversible Jump MCMC(RJMCMC) and Simulated Annealing algorithm(SA) to train Radial Basis Function(RBF) network, so that we can obtain a model with uncertain parameter dimensions. Besides, different model choosing approaches including AIC, BIC, MDL, MAP, HQC, and their performance are compared.

Code

1. Metropolis-Hastings&Gibbs Use Metropolis Hastings algorithm and Gibbs Sampling to estimate parameters in 2D Gaussian distribution.

2. RJMCMC A simple example of Reversible Jump MCMC.

3. RJMCMC+SA Use RJMCMC and SA to train RBF network.

4. Model Choosing A comparison between different model choosing criteria, including AIC, BIC, MDL, MAP, HQC.