Reversible-jump MCMC

tags: Markov Chain Monte Carlo

The Reversible-jump Markov Chain Monte Carlo (RJ-MCMC) algorithm is an extension of the Rosenbluth-Metropolis-Hastings algorithm where the number of dimensions of the parameter space is allowed to change.

And example of application (from Green 1995) is

the multiple changepoint model for Poisson model where the rate is supposed to be piecewise constant but changes an unknown number of times.

Frameworks

In Julia the PPL Gen implements RJ-MCMC
In Python AeHMC plans to implement RJ-MCMC

References

Reversible Jump Markov Chain Monte Carlo Computation and Bayesian Model Determination (Green 1995)
Involutive MCMC: a Unifying Framework
Automating Involutive MCMC using Probabilistic and Differentiable Programming

Links to this note

Bayesian statistics