Henning Meyerhenke
Henning Meyerhenke
> > How long does CG take? > > On regular graph with 320k nodes and 417k edges it will take more than 100 minutes even with rtol=0.3, while LAMG...
> You can also try to reduce the graph by eliminating the degree-1 vertices, compute the resistances, then reinsert and readjust. Sounds like this should make things a lot faster...
I think this may even work by a reduction to the 2-core...
It should. What would you say would be suitable conditions to test for failure?
Sure, a zero-dim. vector should be easy to test for (and catch). I am wondering why this happens and if there is something beforehand we can do to avoid this....
Probably the coarsening step leads to one node only or so. Thanks!
This should be something for @Relux-the-Relux with guidance from @michitux and @angriman. Also the related issue in the EdgeListReader.
A possible reason could be the direction of the walks (incoming vs outgoing).
If you go to the Documentation on the homepage, then search for "commute". Commute time and resistance distance are the same except for the volume(G) factor.
There is. Use CG to set up and solve the corresponding linear system (see numerics folder). For a few pairs this should work within a few minutes, I would assume....