rjmc is an R package designed to perform Bayesian inference using the Reversible Jump Markov Chain Monte Carlo (RJMCMC) algorithm. This method facilitates efficient exploration of variable-dimension parameter spaces, making it especially suitable for tasks such as model selection and problems where the number of parameters can change. This implementation is inspired by the seminal work of Peter J. Green (1995).
Features
- Implements a Metropolis-Hastings-like algorithm that supports jumps between models with different dimensions.
- Flexible framework for user-defined model spaces, likelihoods, and priors.
- Effective at exploring high-dimensional posterior distributions with a variable number of parameters.
- Enables Bayesian model selection and averaging through a rigorous probabilistic framework.
Why Use Reversible Jump MCMC?
Standard MCMC techniques are limited to fixed-dimensional parameter spaces, which restricts their application to problems where the number of parameters is known and constant. RJMCMC overcomes this limitation by allowing jumps between models with different dimensions, enabling:
- Bayesian model selection (e.g., choosing between competing hypotheses).
- Joint inference over models and their parameters.
- Exploration of hierarchical or nested model structures.
By combining the strengths of MCMC with the flexibility of dynamic model spaces, RJMCMC is a powerful tool for tackling complex Bayesian inference problems.
Installation
To install the rjmc package, follow these steps:
Step 1: Install R
Make sure you have R installed on your system. You can download R from https://cran.r-project.org/.
Step 2: Install rjmc from GitHub
You can install the development version of rjmc from GitHub using the devtools package. If you don’t already have devtools installed, you can install it with:
```{r}
install.packages(“devtools”) devtools::install_github(“dchodge/rjmc”)
### Step 3: Load `rjmc` into Your R Session
After installation, you can load the `rjmc` package into your R session with:
```{r}
library(rjmc)
Getting Started
We provide examples of how to implement this package for various scenarios in vignettes:
- Example 1: How to fit a mixture distribution where the number of mixtures is unknown
Refer to the background documentation or vignettes in the package for step-by-step tutorials.
Contributing
We welcome contributions and suggestions! If you’d like to contribute to the rjmcmc package or report issues, please feel free to:
- Submit a pull request on GitHub.
- Open an issue on the repository.
For questions or feedback, contact the package maintainer:
David Hodgson
Email: david.hodgson@lshtm.ac.uk