Parallel Slice Sampling

To draw a sample of a continuous variable B from a finite measure g, using a Markov chain Monte Carlo (MCMC) method, there is an easy algorithm named slice sampling. The two main problems of this algorithm are the solution of a inequality, involving the measure density g, which can be hard to find due to the irregularities g can be affected by, and the high dimensionality of the support of the density itself.Our aim is to create a library, using the slice sampling, to draw an MCMC sample from any density g, in particular to get a realization from the posterior density of a Bayesian model. We will present and discuss a solution and some statistical test applications, using a GPU parallel language, which is, nowadays, becoming more and more commonly employed in the context of Bayesian statistical models.