In this paper, we propose two Markov chains for sampling random vectors that are distributed according to discretized Dirichlet distribution. Their mixing rates are bounded by O(n2 log δ) and O(n3 log δ), where n is the dimension and 1/δ is the grid size for discretization. Our second chain gives a perfect sampler that is based on monotone coupling from the past. We also report the results of simulations.
|Number of pages||33|
|Journal||Japan Journal of Industrial and Applied Mathematics|
|Publication status||Published - Jun 1 2010|
All Science Journal Classification (ASJC) codes
- Applied Mathematics