Hierarchical modeling with NIMBLE
Saturday, 25 June 2022 (Afternoon)
Presenters: Perry de Valpine, Chris Paciorek, Daniel Turek
NIMBLE (r-nimble.org) is a general framework for hierarchical statistical models and methods. It includes methods such as MCMC, particle filtering, and others, all of which are implemented in its high-level algorithm programming system, embedded in R, which you can use to write customized or new methods. NIMBLE provides a new, generalized version of the BUGS model language, implemented using the algorithm programming system, and includes a compiler that uses C++ for fast execution. NIMBLE is growing steadily in use and has enabled huge gains in MCMC efficiency compared to JAGS, depending on the model and customizations. This workshop will introduce NIMBLE programming, which can be used to provide new functions and distributions to achieve case-specific model efficiency, to configure MCMC samplers and/or write new samplers tailored to your model components, or to write new algorithms of your choice. The workshop will also give an overview of some specialized NIMBLE components, including reversible-jump MCMC for variable selection, Bayesian non-parametric distributions, and conditional auto-regressive spatial models. Finally, the workshop will cover the nimbleEcology (capture-recapture, occupancy, dynamic occupancy, hidden Markov, and N-mixture models) and nimbleSCR packages, along with future directions for NIMBLE. It will be assumed that attendees have basic familiarity with writing models in BUGS, JAGS, or NIMBLE. Emphasis will be on the software tools, not on model formulation or statistical properties of methods.
FORMAT: On-line only
Cost: R650 per person (US$45)