期刊
SOFTWAREX
卷 14, 期 -, 页码 -出版社
ELSEVIER
DOI: 10.1016/j.softx.2021.100664
关键词
Reversible jump Markov chain Monte Carlo methods; RJMCMC; Model selection; Parallel MCMC
资金
- Science Mathematics and Research for Transformation (SMART) program - USD/R&E, United States (Under Secretary of DefenseResearch and Engineering)
- National Defense Education Program (NDEP)/BA1, Basic Research, United States
- Alfred P. Sloan Foundation's Minority Ph.D. (MPHD) Program, United States [70481]
CU-MSDSp is a parallel RJMCMC implementation that aims to increase accessibility of RJMCMC to practitioners. It independently forms Markov Chains to approximate the posterior distribution of model parameters, and uses these approximations to estimate the posterior distribution of the model space. This software eliminates the need for designing a trans-dimensional proposal distribution, while ensuring the same theoretical guarantees as the non-parallel algorithm.
Reversible jump Markov chain Monte Carlo (RJMCMC) is a powerful Bayesian trans-dimensional algorithm for performing model selection while inferring the distribution of model parameters. The present work introduces CU-MSDSp as an open source and fully automated parallel RJMCMC implementation that aims to increase the accessibility of RJMCMC to practitioners. CU-MSDSp begins by independently forming Markov Chains to approximate the posterior distribution of each model's parameters. These approximations are then used to estimate the posterior distribution of the model space. This embarrassingly parallelizable software eliminates the need of designing a trans dimensional proposal distribution and Jacobian all while ensuring the same theoretical guarantees as the non-parallel RJMCMC algorithm. Finally, CU-MSDSp enables practitioners to rely on their previous knowledge of fixed dimension MCMC convergence assessment and simulation design. (C) 2021 The Authors. Published by Elsevier B.V.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据