Thanks again Tamás.
I think I have some working code based on your recommendation now. One
more thing. It seems like, in certain cases, the number of merges is not
always Nodes-1 but smaller. Eg. I have a network with 101 nodes and 95
merges. I obtain these numbers by doing
nodes = igraph_vcount(&graph);
fprintf(stdout,"Vertices: %li\n", (long int) nodes);
number_of_merges = igraph_matrix_nrow(&merges);
fprintf(stdout,"number of merges: %li\n", (long int) number_of_merges);
Is this expected behaviour? I.e. is N-1 just an upper boundary for the
number of potential merges? Thanks!
G
On 21/05/12 13:51, Tamás Nepusz wrote:
All I've used so far is only "ready-made" implementations of the Girvan
Neumann algorithm, and I don't remember having had to set the number of
steps. All I remember getting was one community assignment presented to
me as THE GN clustering for that undirected network.
The original publication of the Girvan-Newman algorithm (see
http://www.pnas.org/content/99/12/7821, preprint:
http://www.santafe.edu/media/workingpapers/01-12-077.pdf) does not describe how
to select the number of communities (i.e. how to decide where to cut the
dendrogram). As far as I know, most of the existing implementations use the
modularity score or some other internal quality criterion to decide the number
of communities. For instance, the SNAP library uses the maximum modularity
criterion.
Cheers,
Tamas
_______________________________________________
igraph-help mailing list
[email protected]
https://lists.nongnu.org/mailman/listinfo/igraph-help
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
_______________________________________________
igraph-help mailing list
[email protected]
https://lists.nongnu.org/mailman/listinfo/igraph-help
