Bill Northcott’s binary distribution of JAGS 3.3.0 for Mac OS X is now available. This distribution supports 10.8 (Mountain Lion) and 10.7 (Lion). As explained by Bill in the installation manual, changes in Apple’s developer tools make it difficult to support earlier versions.
In response to a comment by Emmanuel Charpentier, I should write a few words about what has changed in JAGS 3.3.0.
Back in December I wrote about the online machine learning course from Stanford, and how I was looking forward to the course on probabilistic graphical models (PGMs). Unfortunately the second course did not work out so well for me, despite my obvious interest in the topic, and I never completed it. The workload was simply too intense and the weekly deadlines were incompatible with my busy travel schedule during the springtime.
The Stanford courses are now part of an umbrella organization called Coursera, which is aggregating online courses from Universities all over the world. This 20-minute TED talk by Daphne Koller, co-founder of Coursera and lecturer on the PGM course, explains Coursera’s social mission. I now realise I am not really part of the target audience. In fact Coursera has its eye on the global marketplace for education. By removing financial and logistic barriers to entry they aim to make higher education accessible to people who would never otherwise have a chance to follow such courses.
An interesting point made by Koller is that online courses generate large amounts of data, which can be analyzed to improve the course the next time it is repeated. The same idea is reiterated by Peter Norvig in a 6-minute TED talk. Norvig’s talk also explains why the weekly deadlines – which I complained about above – are necessary to make these courses work.
Not discouraged by my failure on the PGM course, I have signed up for two autumn courses: Heterogeneous Parallel Programming and Functional Programming Principles in Scala. Both are relatively short (6 and 7 weeks respectively) so should be more manageable although I am sure I will have to give up one of them.
Andrew Gelman has announced the release of Stan version 1.0.0 and its R interface RStan. Stan – named after Stanislaw Ulam, the inventor of the Monte Carlo method – is a new MCMC program that represents a major technological leap forward. It works flawlessly on my Linux desktop and is very, very fast. (Note that I have nothing to do with the Stan project: I am just posting this here because it is clearly of interest to JAGS users). Continue reading
Kodi Arfer sent a script that reveals a bug in the glm module. If you set the random number generator seed in the initial values, then the output from jags should be identical every time you run the model. Unfortunately, Kodi’s script shows that this is not the case. In the interests of transparency I am putting a note here on the blog but I currently have no idea how to fix this.
The script, and some sample output, is below