We’ve released the newest version of NIMBLE on CRAN and on our website. NIMBLE is a system for building and sharing analysis methods for statistical models, especially for hierarchical models and computationally-intensive methods (such as MCMC, Laplace approximation, and SMC).
This is a micro release that primarily addresses some packaging changes requested by CRAN. In addition, this release includes:
- A multinomial MCMC sampler,
sampler_RW_multinomial, for random variables following a multinomial distribution. - Some enhancements to error trapping and warning messages.
- A variety of minor bug fixes.
Hello Chris, I have been following Nimble from a distance. I am especially interested in how you are getting on with automatic differentiation. I note you have used a C++ library for this. One of the most puzzling things about STAN is how hard it to get it working in a basic sense, the fact that the C++ compilation process often falls over. Have you had user problems in this regard?
I have now given up developing the BUGS software. It is time to retire and spend time trekking in the mountains. I have started a new project, a new PPL, called ProbPALA. This works by source code to source code translation and so is more like STAN, maybe more like Nimble too. I have written the AD code by hand. Things work for most of the BUGS examples, using both a Gibbs type approach and using HMC. However I am not always able to marginalize out discrete variables to get HMC working.
I think I read you had not got HMC working for CAR models. I am also stuck on that. How to implement sum to zero constraints in HMC. Maybe constraint by krigging is part of the answer.
All the best
Andrew