CLP(Prob)

We should look at HANSEI, Church, Prism, ProbLog, CLP(BN) and all of the other interesting approaches to probabilistic logic programming.

Markov Chain Monte Carlo & Gibbs Sampling seem to be the hot topic these days. It's interesting that some promising work has been done executing this stuff on GPUs.

Existing Projects in the space

Church

• A good video on Church.
• A good tutorial

Infer.net

Infer.net is Microsoft's probabilistic programming language in F#

• The webpage links to many good resources on the area.

Stan

New project based around Hamiltonian Monte Carlo, which has advantages in situations with correlated variables.

• The Stan project
• The Reference Manual
• The code repo

Bugs and JAGS

A probabilistic programming language, based on Gibbs Sampling, that has been in use for a long time is BUGS. A more recent implementation is JAGS, which largely supercedes BUGS. The beauty of these are that you can actually get them up and running in R without much effort.

• The rjags R package lets you quickly do some probabilistic programming. In particular, you can actually run the examples in the blog post below.
• A good blog post on the sort of model BUGS/JAGS can represent.
• The jags repo
• The jags user manual is easy to miss!