Re: [igraph] weighted pagerank?

[Eytan Bakshy]

Hi Csardi,

My bad, perhaps this is an issue with only the python version and not  
the R version.

Here are two python versions of the test example you gave in R:

        from igraph import *
        import random
        g = Graph.Barabasi(1000,m=5).simplify()
        pg1 = g.pagerank()
	pg2 = g.pagerank(weights=[random.randint(1,1000) for i in  
xrange(g.ecount())])
        max([(i-j)/i for i,j in zip(*(pg1,pg2))])

yields 0.0.  also, if i compare pg1 and pg2 directly (pg1 == pg2), the  
two vectors identical

I alternatively tried assigning the weights to edge attributes:
        for e in g.es: e['weight'] = random.randint(1,1000)
        pg2 = g.pagerank(weights='weight')
and got the same result,

Is there something I am doing incorrectly in specifying the edge  
weights in python?

Thanks,

-e


On Aug 23, 2008, at 10:02 AM, Csardi Gabor wrote:

> Eytan,
>
> the difference is more than 50 percent. I guess it would be more
> if the weights were not uniformly distributed.
>
> > library(igraph)
> > g <- simplify(ba.game(1000, m=5))
> > pg1 <- page.rank(g)
> > pg2 <- page.rank(g, weights=runif(ecount(g),1,1000))
> > max( (pg1$vector - pg2$vector) / pg1$vector)
> [1] 0.5442513
>
> Gabor
>
> On Fri, Aug 22, 2008 at 07:18:30PM -0400, Eytan Bakshy wrote:
> > Hello,
> >
> > I am using igraph 0.51 in R and igraph 0.5 in python, and it doesn't
> > seem that specifying edge weights has any effect in either
> > implementation.  I know that the igraph 0.5 release announcement
> > mentions that Page Rank and the fast greedy algorithm supports edge
> > weights, but is this just in the C library.
> >
> > In python, I tried creating several graphs where edges have the
> > attribute 'weight', which were floats.  I am trying to compute the
> > weighted pagerank by saying:
> >         prw = g.pagerank(weights='weight')
> >
> > and if i do:
> >         pr = g.pagerank()
> > the following statement yields True:
> >         prw == pr
> >
> > on several graphs using edge weights that were distributed uniformly
> > at random between (0,1), and then again between (1,1000)
> >
> >
> > I am pretty new to R, but it appears that the same thing holds using
> > page.rank(g, weights=runif(length(E(g)),1,10000))
> >
> >
> > Thanks,
> > Eytan
> >
> >
> > _______________________________________________
> > igraph-help mailing list
> > address@hidden
> > http://lists.nongnu.org/mailman/listinfo/igraph-help
>
> --
> Csardi Gabor <address@hidden>    MTA RMKI, ELTE TTK