Online Program Home
My Program

Abstract Details

Activity Number: 216 - Modern Bayesian Computing in Ecology
Type: Invited
Date/Time: Monday, July 30, 2018 : 2:00 PM to 3:50 PM
Sponsor: ENAR
Abstract #326728 Presentation
Title: Beyond the Black Box: Flexible Algorithm Programming for Ecological Models in NIMBLE
Author(s): Christopher Paciorek* and Colin Lewis-Beck and Perry de Valpine and Daniel B. Turek and Lauren Ponisio and Nick Michaud
Companies: University of California, Berkeley and Iowa State University and UC Berkeley and Williams College and UC Riverside and UC Berkeley
Keywords: hierarchical models; algorithms; MCMC; software development; reproducible research; ecology

The NIMBLE hierarchical modeling package provides a flexible system for working with ecological models. NIMBLE is compatible with WinBUGS and JAGS for writing models but provides extensibility and flexibility. Models can include new distributions and functions provided by the user. MCMC sampler choices can be customized, and new samplers easily written. NIMBLE's algorithm programming language supports methods beyond MCMC, such as model selection and validation and provides a platform for methodologists/developers to disseminate their own methods. These features enable the use of increasingly sophisticated statistical models in ecology. We illustrate NIMBLE's use with various important ecological models. For multi-state capture-recapture models and multi-species dynamic occupancy models, NIMBLE allows direct marginalization over individual latent states, greatly reducing dimensionality and improving MCMC efficiency. For spatial capture-recapture models, new formulations of common models and customized MCMC samplers (including reversible jump) improve performance. We illustrate model selection via cross-validation and model assessment via calibrated posterior predictive p-values.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2018 program