We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

stat.ME

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Methodology

Title: Parallel Markov Chain Monte Carlo for Bayesian Hierarchical Models with Big Data, in Two Stages

Abstract: Due to the escalating growth of big data sets in recent years, new Bayesian Markov chain Monte Carlo (MCMC) parallel computing methods have been developed. These methods partition large data sets by observations into subsets. However, for Bayesian nested hierarchical models, typically only a few parameters are common for the full data set, with most parameters being group-specific. Thus, parallel Bayesian MCMC methods that take into account the structure of the model and split the full data set by groups rather than by observations are a more natural approach for analysis. Here, we adapt and extend a recently introduced two-stage Bayesian hierarchical modeling approach, and we partition complete data sets by groups. In stage 1, the group-specific parameters are estimated independently in parallel. The stage 1 posteriors are used as proposal distributions in stage 2, where the target distribution is the full model. Using three-level and four-level models, we show in both simulation and real data studies that results of our method agree closely with the full data analysis, with greatly increased MCMC efficiency and greatly reduced computation times. The advantages of our method versus existing parallel MCMC computing methods are also described.
Comments: 30 pages, 2 figures. New simulation example for logistic regression. MCMC efficiency measure added. Details of convergence diagnostics added. One additional table
Subjects: Methodology (stat.ME); Distributed, Parallel, and Cluster Computing (cs.DC); Computation (stat.CO); Machine Learning (stat.ML)
Cite as: arXiv:1712.05907 [stat.ME]
  (or arXiv:1712.05907v2 [stat.ME] for this version)

Submission history

From: Erin Conlon [view email]
[v1] Sat, 16 Dec 2017 06:14:18 GMT (349kb)
[v2] Wed, 16 Jan 2019 22:07:54 GMT (539kb)

Link back to: arXiv, form interface, contact.