[R--gR] deal package

[Matthias Buehlmaier]

Hi Kevin, 

You mentioned that your problem is the following: bnlearn cannot cope with both discrete and continuous variables. 

One potential solution would be to simply discretize your continuous variables using some (not necessarily equidistant) grid. I am unsure whether or not this is the best solution, but it should work. 

Cheers, 

Matthias 


Lu, Kevin wrote:
> Dear all,
> 
> I learn something from my colleagues about the beauty of Bayesian
> network. I want to use the deal package to analyze my genetic data. I
> have 3575 subjects. For each subject, I have 20 genetic marker
> information (discrete, three levels) and his or her plasma cholesterol
> levels (continuous). I want to construct the markov blanket of plasma
> cholesterol levels. I tried the deal in my computer and computer
> cluster. Due to so many genetic markers, I can not successfully pass the
> calculation of the prior probability of my network. The maximum
> variables I could put into the network is about 12. I can not add more.
> 
> The algorithms (gs, iamb, hc) in bnlearn package can not deal the
> discete and continuous variables at the same time. Could someone give me
> some suggestions how I should do it? 
> 
> Thanks in advanve,
> 
> Kevin
-- 

für nur 19,99 Euro/mtl.!* http://portal.gmx.net/de/go/dsl02