graph coarsening methods does not match with the paper?

[gegemy]

The source code uses sfdp library for graph coarsening, but the paper is about the graph coarsening methods proposed by yourself. Is there any corresponding implementation?

[GTmac]

Great question. I would say that the idea of using graph coarsening is important; the exact graph coarsening method being used is not that important.

That said, I use the sfdp library mostly because it implements a number of different graph coarsening algorithms so that I can see which one works the best. The one I used in this repo is almost the same algorithm as described in the paper (perhaps plus some minor tweaking). Hope this answers your question.

[gegemy]

Thanks for your prompt reply. The sfdp library is part of the graphviz package, but I didn't find the detailed manual and parameter descriptions(like the the specific meaning of the -g2 parameter (coarsening_scheme) ) of the sfdp you used on the graphviz website(https://www.graphviz.org). Would you mind give me one?

[xyvivian]

Hello, I am trying to run the sfdp algorithm on a mesh like graph, but it does not provide any coarsening results. May I know the detailed manual of sfdp algorithm too?

[ot4f]

The meanig of the sfdp args used in this paper are: -m: number of multilevels, default infinity -p: power of repulsive force, default is usually -1, sometimes -1.8. Must be a negative real number. -q: power of stress function, default is 2. Must be a positive real number. At the moment this only applied to stress-maxent model -g: graph coarsening scheme, should be an integer between 1 and 6 or equals to 12 -d : dimension to generate layout. 2 (default) or 3 -u: wether the graph is assumed to be undirected with no loops. Default is directed. -n: whether to use node weights. Not used by default -v: verbose output. -T fmt: set output format: m(athematica) d(ot) c(oordinate), ps(postscript), png, edgelist(a binary format for gviewer_stream fast rendering of huge graphs)