Design And Implementation Of Parallel Sampling Method Based On ARMS

Markov chain Monte Carlo method is an important statistical method, and the difficulty of using it is the implementation of the random number generation.Adaptive Rejection Metropolis-Hastings Samping method is a sampling method. It uses Adaptive Rejection Sampling method and Metropolis method to sample from general distribution. By using Adaptive Rejection Metropolis-Hastings Samping method and Gibbs method, the thesis provides a parallel sampling method. It can be used to sample form the complex statistical models and to do the convergence diagnosis.To actualize parallel sampling, the thesis provides a combined random number generator. And it can supply different random numbers for each processor. Message passing interface is also used to transfer the information needed during the computation. After using Gibbs method, several markov chains can be gained. By comparing the variances of different chains, convergence diagnosis can be achieved. Feasibility and high parallel computation efficiency of the software are confirmed by the experimental results.