Re: [igraph] benchmarks for community algorithm testing

[Tamás Nepusz]

> Thanks Tamas,  I have played with the preference.game but with limited 
> success. I am struggling to see how the pref.matrix and type.matrix work. Can 
> you supply an example where the output is a GN graph?
If I remember correctly, the GN benchmark simply groups N vertices into K 
communities "in advance", then draws edges between vertices of the same group 
with probability p1 and vertices of different groups with probability p2. p2 is 
then varied from 0 to p1.

In the preference.game, types should be the number of communities; type.dist 
should be the relative sizes of these communities (in other words, the 
probability that a given vertex belongs to a given community), and pref.matrix 
should be a matrix where the diagonal elements contain p1 (this is the same 
community case) and all the other elements contain p2.

Hmm, actually I have just realized that the reason why preference.game is not 
suitable for you is because it does not tell you which group the vertices 
belong to; this should be added as a vertex attribute in order to use the 
generated graph in the GN benchmark. Sorry for the confusion.

-- 
T.