Introduction to NIMBLE: the fanciest member in the BUGS family
Instructors: Perry de Valpine, University of California-Berkeley
Date: 23–25 April 2018
Venue: Swiss Ornithological Institute, Sempach, Switzerland
Computers: Bring your own laptop with latest R and NIMBLE
Registration: CHFr. 350 (normal rate), CHFr. 200 (student rate)
Complex hierarchical models are often fitted with software that uses the BUGS language for model description (e.g., JAGS, WinBUGS or OpenBUGS). For non-statisticians the simplicity of the BUGS language is a key to be able to fit such models in practice. However, the fitting algorithms implemented in these programs (MCMC) are mostly black boxes, and there are only very limited possibilities for the user to affect them. Therefore, in some cases convergence might be difficult to achieve at all or the algorithms may be inefficient. The new NIMBLE software (https://r-nimble.org/) offers solutions to these problems. It comes as an R package and uses the BUGS language to describe statistical models. In addition it gives very wide user control over the MCMC samplers. Moreover, NIMBLE interacts in a more direct way with R allowing the possibility to easily specify your own functions as needed.
Perry de Valpine is a creator and main developer of NIMBLE and will give a comprehensive overview about the possibilities of this exciting new package.
The workshop is geared towards scientists from ecology and related fields (e.g. fisheries and wildlife). The general topics in hierarchical statistical modeling and computational methods apply as well to many other applications. Contents include the following:
1. Introduction to NIMBLE
- Migrating from BUGS and JAGS
- The NIMBLE workflow in R
- Querying and manipulating models in R
2. MCMC in NIMBLE
- Configuring samplers
- Comparing different kinds of MCMC (within and outside of NIMBLE)
3. Programming in NIMBLE
- The nimbleFunction system for writing model-generic algorithms
- The NIMBLE compiler
4. Writing new samplers, new distributions and new functions to use in models
5. Maximum likelihood with NIMBLE
6. Sequential Monte Carlo (particle filtering) with NIMBLE
In this intermediate-level workshop about 3/4 of the time is spent on lecturing and 1/4 on solving exercises. No previous experience with the NIMBLE software is assumed. However, a good working knowledge of modern regression methods (ANOVA, ANCOVA, GLMs) and of program R is required. Moreover, knowledge in of the programs using the BUGS language (JAGS, WinBUGS) is highly advantageous.
Send your application to Michael Schaub (firstname.lastname@example.org), with CC to Marc Kéry (email@example.com); describing your background and knowledge in statistical modeling, R and WinBUGS/OpenBUGS/JAGS/NIMBLE and capture-recapture, by 28 February 2018 at the latest. Workshop invitations will be sent out immediately afterwards.