Since I haven't seen a reply to this, I will offer a couple of
comments. I've never used "deal", but it sounded interesting, so
I
thought I'd look at it.
Have you looked at Susanne G. B?ttcher and Claus Dethlefsen. deal: A
Package for Learning Bayesian Networks. Journal of Statistical Software,
8(20), 2003, and the deal reference manual downloadable under
"documentation" from "www.math.aau.dk/~dethlef/novo/deal"?
If yes and
you still would like more help from this listserve, please submit
another post including a simple, self-contained example explaining
something you've tried and why it doesn't seem to answer your question?
(This is suggested in the posting guide!
'www.R-project.org/posting-guide.html'.)
This documentation might answer your questions. Even though I've not
read them, I will guess potential answers to your two questions, hoping
some other reader may disabuse us both of our ignorance:
From what I saw in the examples, I would guess that "deal" supports
two types of distributions: normal and finite (discrete). If so, it
does NOT support a Poisson. If it were my problem and I still held that
view after reviewing this documentation, I might write to the maintainer
[listed with help(package="deal")] and ask him for suggestions. Then
if
it were sufficiently important, I might think about how I would modify
the code to allow for a Poisson.
Regarding simulations, have you looked at "rnetwork", which
provides
"simulation of data sets with a given dependency structure"?
Hope this helps,
Spencer Graves
Carsten Steinhoff wrote:> Hello,
>
> I want to use R to model a bayesian belief network of frequencies for
system
> failiures in various departments of a company.
>
> For the nodes I want to use a poisson-distribution parameterized with
expert
> knowledge (e.g. with a gamma prior).
> Then I want to fill in learning-data to improve the initial estimates and
> get some information about possible connections.
> Later I want to simulate dependend random variables from the network
>
> I tryed to use the package "deal" for that task, which is as far
as I know
> the best (and only?) R-package for that task.
> But a few questions rose that I could not solve with the documentation:
>
> (1) Is it possible to parameterize the prior distribution (for example
> (dpois(x,lambda=60) directly and non gaussian ?
>
> (2) If I've chosen a structure, can I simulate dependend states that
are non
> gausian distributed?
>
> Thank you for any idea!
>
> Regards, Carsten
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> 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