[igraph] Re: Centralization in igraph?

[Uri Shwed]

Hi Gabor,
Thanks for the quick reply!
in evcent the tmaxdev value for directed graphs is simply the number of 
nodes -1, right? So a general centralization function for directed 
graphs could be:

centralize <- function(scores, normalize=TRUE) {
 cent <- sum(max(scores)-scores)
 tmax<-length(scores)-1
 if (normalize) {
   cent <- cent/tmax
 }
 cent
}

??

In addition, I've just noticed that all my graphs, created by 
"graph.edgelist(x)", have one more node than their corresponding 
networks, created by "as.network(x, matrix.type="edgelist")". Any idea 
why, or what I should do about this? I only have this problem with data 
imported as directed edgelist - my other set of undirected valued graphs 
is fine.
Thanks,
Uri



Gabor Csardi wrote:
> Uri, thanks, that's kind of you, but actually it is 
> not included in igraph. But it very simple to do it:
>
> centralize <- function(scores, tmax, normalize=TRUE) {
>   cent <- sum(max(scores)-scores)
>   if (normalize) {
>     cent <- cent/tmax
>   }
>   cent
> }
>
> It is a bit different than sna, you have to give the 'tmax' 
> argument 'by hand', as igraph centrality functions do not support 
> returning the theoretical maximum. Here is how to use it:
>
> library(igraph)
> g <- simplify(ba.game(1000, m=2))
> centralize(degree(g), tmax=997002, normalize=TRUE)
>
> g2 <- erdos.renyi.game(1000, 4/999)
> centralize(degree(g2), tmax=997002, normalize=TRUE)
>
> Unfortunately, you'll need to calculate the theoretical
> maximum for the hub scores..... You can see the evcent 
> function of SNA (tmaxdev=TRUE) to get a hint....
> Please tell me if you could find it out.
>
> Gabor
>
> ps. please consider writing to the igraph-help mailing list 
> next time. Thanks.
>
> On Thu, Jun 19, 2008 at 02:49:57PM -0400, Uri Shwed wrote:
>   
> > Dear Gabor,
> > Does igraph supports a general centralization calculation, like sna? I  
> > need a GLI for Klienberg hub and authority scores. Knowing igraph's  
> > style, I'm sure its there somewhere....
> > Thanks,
> > Uri
> >     
>
>