Re: [igraph] communities_spinglass original method with negative weights

[Tamas Nepusz]

Hi,

> http://igraph.org/c/doc/igraph-Community.html#igraph_community_spinglass
>
> "The vector giving the edge weights, it may be NULL, in which case all
> edges are weighted equally. Edge weights should be positive, altough
> this is not tested."
>
> What does this mean exactly? What happens if I pass negative weights?

Well, to be honest, I have no idea, because we simply use the original
implementation of Reichardt & Bornholdt for the spinglass clustering
method. Probably the result is useless, and indeed we should add an
explicit check in the C layer for negative or zero weights for the
default implementation. (Although I suspect that zero weights should
not cause a problem).

Regards,
T.