[R--gR] Package for Discrete Bayesian Networks
Kjell Konis
kjell.konis at epfl.ch
Sat Mar 28 14:41:53 CET 2009
If you have access to Hugin then you may find the RHugin package useful.
http://rhugin.r-forge.r-project.org/
It provides an R API for the Hugin Decision Engine.
Cheers,
Kjell
On Mar 27, 2009, at 3:45 PM, Thomas Liebig wrote:
> Hi Matthias,
>
> I also want to utilise large discrete Bayesian Networks with R (for
> spatial applications). Actually I learn the networks outside R with
> some
> selfmade tools and am looking for a package that offers me basic
> inference and sampling functions afterwards.
> Deal doesnt provide them. Claus Dethlefsen suggested to use Rhugs
> instead that provides an interface to hugin.
>
> Therefore, i'm currently implementing these two R functions
> (inference,
> sampling) plus an additional import function. For your application you
> would also need a fast Network Search algorithm for large networks.
> Perhaps the 'Sparse Candidate Algorithm', 'Scalable Sparse Bayesian
> Network Search' or the 'Screen Based Network Search' as deal only
> provides the poor Greedy Search for Bayesian Networks.
>
> Feel free to contact me in case, you are interested in exchanging
> functions or experience. Your application sounds also very interesting
> to me, and i would be happy to learn more about your textmining
> domain.
>
> regards,
> Thomas
>
> Mat_1 at gmx.net schrieb:
>> All,
>>
>> I would like to apply Bayesian networks to text-based information.
>> That is, I would like to use Bayesian networks to extract
>> 'sentiment' (say, 'good' vs. 'bad') form unstructured text.
>>
>>> From the CRAN Task View on gRaphical Models in R it seems that I
>>> should use the combination of the two packages deal and gRain
>>> (please correct me if I overlooked something). The package deal
>>> can work both with discrete and continuous variables. For my
>>> project I only need to work with discrete variables. Due to the
>>> large amount of data that I potentially have to handle, I am
>>> concerned that maybe deal will not run fast enough because it is
>>> 'too general' for my purpose.
>>
>> In this sense, I would like to ask the following: is there an R
>> package that I overlooked and that is tailored towards Bayesian
>> networks with discrete variables?
>>
>> Many thanks,
>>
>> Matthias
>>
>
> --
> Thomas Liebig
> Fraunhofer-Institut für Intelligente Analyse- und
> Informationssysteme (IAIS)
>
> Schloss Birlinghoven, D-53754 Sankt Augustin, Germany
> Email: thomas.liebig at iais.fraunhofer.de
> Phone: +49 2241 142050
> Fax: +49 2241 142072
>
> _______________________________________________
> R-sig-gR mailing list
> R-sig-gR at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/r-sig-gr
More information about the R-sig-gR
mailing list