Re: [R--gR] deal: heuristic search lowers the score?

Dear Sam

The score should not be positive, so something is not right here. This is possibly caused by numerical
instabilities, either because the jointprior has a too low size of the imaginary database, or because the
continuous variables need to be rescaled. If, for example, you have a continuous variable taking values
around 0.001, try multiplying the variable by 1000.

I have not explored this further, so I have not got a really good explanation for what is going on.

The relscore in a networklist (e.g. after you have sorted the networks according to their score using
'nwfsort') is the relative score of the networks using the network with the highest score as reference.
(see help(nwfsort)).

Best,
Claus

________________________________
Claus Dethlefsen, Msc, PhD
Statistiker ved Kardiovaskulært Forskningscenter

 
Forskningens Hus
Aalborg Sygehus 
Sdr. Skovvej 15
9000 Aalborg

Tlf:   9932 6863
email: aas.claus.dethlefsen <at> nja.dk <mailto:aas.claus.dethlefsen <at> nja.dk> 

________________________________

Fra: r-sig-gr-bounces <at> stat.math.ethz.ch på vegne af Sam Steingold
Sendt: on 22-03-2006 20:08
Til: dealsoftware <at> yahoogroups.com
Cc: r-sig-gr <at> stat.math.ethz.ch
Emne: [R--gR] deal: heuristic search lowers the score?

Hi,
I noticed that the network created by heuristic() has lower score than
the one created by autosearch():

prior <- jointprior(net,nrow(df)*10)
search <- autosearch(net,df,prior,timetrace=TRUE,removecycles=TRUE)
heuristic <- heuristic(getnetwork(search),df,trylist=gettrylist(search))
score(getnetwork(search))
[1] 378.5992
score(getnetwork(heuristic))
[1] 357.014
why?  does this mean that the score is actually the _negative_ of log
likelihood?

also, it is not clear what the relscore is.

PS. I read the paper and the manual...

--
Sam Steingold (http://www.podval.org/~sds) on Fedora Core release 4 (Stentz)
http://www.honestreporting.com http://www.savegushkatif.org
http://ffii.org http://www.dhimmi.com http://www.camera.org
Democrats, get out of my wallet! Republicans, get out of my bedroom!

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