Examples of three MCMC sampling algorithms for estimating the posterior of a univariate Poisson-Gamma model, using the 1854 Broad Street Cholera Outbreak dataset. Samplers include a generic Metropolis algorithm with parallelized log-likelihood calculations, diminishing adaption and naive parallel prefetching. OpenMP is used for parallel computing.

Ensure that a version of clang with OpenMP support is available. If issues linking against OpenMP are encountered, check that clang is able to find the relevant headers (libomp-dev may need to be installed on Debian based systems).

Setup the project directory and build the sampling program.

mkdir build bin
make data/cholera_counts.txt
make bin/sampler

Run the sampler using e.g. prefetching.

./bin/sampler prefetching data/cholera_counts.txt > samples.txt