[R--gR] learning networks with a large number of variables and pre-set parents.

[DED (David George Edwards)]

Hi zhilua li

I dont know of any software that can handle 10000 (discrete) variables.
CoCo can handle a couple of hundred I believe.
I dont think that gR is looking at this type of problem as yet, this
must be phase 2.

Best regards
David 


-----Original Message-----
From: r-sig-gr-bounces at stat.math.ethz.ch
[mailto:r-sig-gr-bounces at stat.math.ethz.ch] On Behalf Of zhihua li
Sent: 25. marts 2005 06:13
To: r-sig-gr at stat.math.ethz.ch
Cc: r-help at stat.math.ethz.ch
Subject: [R--gR] learning networks with a large number of variables
andpre-set parents.


hi netters: 

I have a series of  discrete variables which form a network and  I want
to 
learn the network structure from some training data. I could have used 
packages like deal but there are two problems. 

First of all, I have 10000 variables. So the possible network structure
is 
awfully huge, I don't know how long it will take my PC to find the 
highest-scoring network..........maybe a month? 
Secondly, I have some prior knowledge that only 500 out of the 10000 
variales are possible parents. In another word, only those arrows
startting 
from the 500 variables and pointing to the remaining 99500 variables are

allowed in the network.  In deal an assignment to "banlist" should help
me 
rule out the impossible arrows. But in my case the number of "impossible

arrows" is  500*499+99500*99549, and so the "banlist" would get 
unacceptable long. Are there any methods (in deal or other packages) to 
specify the parents set in advance? 

Thanks a lot!

_______________________________________________
R-sig-gR mailing list
R-sig-gR at stat.math.ethz.ch
https://stat.ethz.ch/mailman/listinfo/r-sig-gr