[R] R equivalent to Matlab's Bayes net toolbox

[Jose]

Søren Højsgaard <Soren.Hojsgaard <at> agrsci.dk> writes:

> 
> Jose,
> I am not entirely sure what Matlabs Bayes net toolbox does, but I guess it 
implements as propagation
> algorithm for Bayesian networks. There is no such package on CRAN - yet - but 
there will be soon: I've
> created a package called gRbayesnet which implements the "Lauritzen-
Spiegelhalter" propagation
> algorithm. I expect to upload it to CRAN within the next few days. 
> Best regards
> Søren
> 
> ________________________________
> 
> Fra: r-help-bounces <at> stat.math.ethz.ch på vegne af Jose Quesada 
> Sendt: ma 16-07-2007 17:58
> Til: r-help <at> lists.r-project.org
> Emne: [R] R equivalent to Matlab's Bayes net toolbox
> 
> Hi,
> 
> I'm attending  summer School at UCLA (IPAM) on "probabilistics models of 
> cognition". I have been an R-user since v. 1.4.1, but was trained in the 
> frequentist tradition (as most psychologists!). I found that all faculty 
> here use matlab and Murphy's bayes net toolbox. I have not had the need to 
> use matlab before, and would love to stick to R for graphics models and 
> bayesian modeling in general (even if it takes me extra time to cross-code 
> the examples in matlab into R).
> 
> I'm trying to find an R equivalent to Matlab's Bayes net toolbox.
> 
> I have found packages 'deal' and 'gR', and played around with:
> http://www.ci.tuwien.ac.at/gR/
> 
> But I cannot really figure out how all these packages are integrated. 
> Also, appendix B of 'bayesian AI' lists gR as "vaporware" (although this 
> could well be outdated by now).
> 
> Is there any R news article on bayesian networks? It's hard to find, 
> because I don't think the content of R-news is indexed in CRAN. I could 
> download every issue and search the TOC, but it'd be time-consuming.
> 
> Even though the examples in the documentation in package 'deal' are good, 
> they fall short. A good tutorial would be great.
> What I'd like to know from you is whether R is a sensible choice or 
> whether BNT is just easier and more mature.
> 
> Right now I could easily chose R or Matlab, since I have made little 
> investment in any form of bayesian networks modeling; However, since I 
> have a better background in R than in Matlab, I'd love to stay with R.
> 
> Any resources (mailing lists, books, tutorials) would be greatly 
> appreciated.
> 
> Thanks a lot in advance,
> -Jose
> 
> --
> Jose Quesada, PhD.
> http://www.andrew.cmu.edu/~jquesada
> 
> ______________________________________________
> R-help <at> stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 
> ______________________________________________
> R-help <at> stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 
> 



Hi Søren,

It looks like bnt implements several algorithms for learning both parameters 
and structure:
http://bnt.sourceforge.net

MCMC for example is implemented. I didn't know about this Lauritzen-
Spiegelhalter" propagation algorithm. 

The thing that I don't understand in the gR page is why there are so many 
different packages and why they are not very integrated:
boa
CoCo
coda
deal
dynamicGraph (core)
ggm
gRbase (core)
mathgraph
mimR
R2WinBUGS
rbugs
SIN
http://cran.r-project.org/src/contrib/Views/gR.html

I tried deal, for example, and it's not really comparable to bnt.

In Matlab, there seems to be a package (well, toolbox) only, and there seems to 
be some of a community around it:
http://tech.groups.yahoo.com/group/BayesNetToolbox/

Is there any kind of community doing graphical models stuff on R?

Thanks,
-Jose