ROOM: RSCC, F7
NIMBLE is a relatively new software package for general hierarchical statistical modeling. NIMBLE adopts the same modeling language as JAGS and WinBUGS, but it provides much more flexibility. Users work in R and can write new functions and new distributions to be used in models. MCMC algorithms are highly configurable. Users can choose what kinds of samplers to use and even write new samplers. NIMBLE is not just for MCMC: It can be used for variable selection, particle filtering, maximum likelihood estimation, cross-validation, or whatever algorithm one writes in its algorithm language to work with general models. All models and algorithms are compiled from code-generated C++ for efficient computation. Users do not need to know anything about C++.This workshop will cover application of NIMBLE to common wildlife models such as spatial capture-recapture, occupancy and abundance models, state-space models and integrated population models, and spatial models. Attendees will learn how customizations to nimble via new functions, distributions, or MCMC sampling configurations can, in some cases, yield large boosts in MCMC performance, sometimes by order(s) of magnitude. It will be assumed that attendees have basic familiarity with JAGS and/or WinBUGS and some hierarchical models of interest. The workshop will also survey other capabilities of NIMBLE, including variable selection by reversible jump MCMC, maximum likelihood estimation by MCEM or other methods, and topics of attendee interest.
Organizers: Perry de Valpine
Supported by: Biometrics Working Group