[BioC] KLdist.matrix

[Wolfgang Huber]

Dear Ladan

you can see the source code of this function by typing
	
	showMethods("KLdist.matrix", inc=T)

and there you will see the code fragment

   for (i in 1:nr) {
      if (nc > 10000)
        dat <- sample(x[i, ], nc * 0.1)
    ....
    }

i.e. the data are randomly subsampled if nc, the number of columns of 
the matrix, is above 10000.

It would have been reasonable to expect that this behaviour would be 
documented in the function's manual page.

Numeric integration is a difficult task, and I would consider this 
function a first, brave move into the right direction. Feedback on its 
performance on various applications will for sure be welcome, and if you 
take it as a start point for adaptation / further development to fulfill 
the needs of your use case, that might be the wisest way of making use 
of it.

	Best wishes
	Wolfgang




Il Jul/1/11 12:27 PM, Ladan Amini ha scritto:
> Hello,
>
> I am using KLdist.matrix from bioDist package, and the KL distance between
> two time series is
> slightly different for different runs.
>
> Example:
> n.th=rnorm(10^6,mean=0,sd=sqrt(0.1))
> n.exp=rnorm(10^6,mean=0,sd=sqrt(0.5))
> x<- rbind(n.th,n.exp)
> #run no.1
>        KLdist.matrix(x, symmetrize = TRUE)
> #result:
> #           n.th
> #n.exp 1.228867
> #run no.2
>         KLdist.matrix(x, symmetrize = TRUE)
> #result:
> #           n.th
> #n.exp 1.142033
> #run no.3
>        KLdist.matrix(x, symmetrize = TRUE)
> #result:
> #           n.th
> #n.exp 1.162144
>
> I tried to fix the value of gridsizenumber, but it gives the following
> error:
>
> KLdist.matrix(x, symmetrize = TRUE,gridsizenumber=10)
> Erreur dans .local(x, ...) : argument(s) inutilisé(s) (gridsizenumber = 10)
> ERROR
>
> I would appreciate a lot if anyone can help me to solve this problem.
>
> Thank!
> Ladan
>
> 	[[alternative HTML version deleted]]
>
>
>
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at r-project.org
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor


-- 


Wolfgang Huber
EMBL
http://www.embl.de/research/units/genome_biology/huber