DSC-2007 Common Lisp talks by Ross Ihaka and Tony Rossini (R-Project)



On 1/25/07, Mario Rodriguez <biomates at telefonica.net> wrote:

> Symbolic environments should be very powerful tools in some
> probabilistic contexts. For example, a cas with Lebesgue or Stieltjes
> integrals could be used to compute probabilities both with discrete and
> continuous models.

Agreed, that would be a good application.

> If the cas also incorporates a graph package, a possible application
> could be in 'graphical models', these are random multivariate models
> where stochastic dependencies are represented by graphs, both directed
> or undirected, as in bayesian networks, or random fields. Sometimes
> there are conditional dependencies among discrete and continuous random
> variables. The likelihood function must be constructed and numeric
> parameters (even the graph's structure) estimated from the sample.

Yep.

> I don't know if someone, somewhere, has started such a project.

I wrote some preliminary code to compute posterior distributions
in simply-connected Bayesian networks (i.e. the easy case)
and submitted a paper about it to a conference (it was rejected).
If anyone is interested to see the paper I'll be happy to send it.
I didn't really get very far with the code -- just enough to solve
some toy problems. I should probably get back on it.

All the best,
Robert