Re: [igraph] python ig.Graph.multilevel_community

[Tamás Nepusz]

Hi Lucio,

Sorry for the late reply. I went back to the revision of the code that was used 
when igraph 0.6.5 was released and indeed you are right, there is a bug with 
the weighted multilevel community detection code. I did not notice it because I 
was accidentally using the development version where this was fixed already; 
see the following commit:

https://github.com/igraph/igraph/commit/228dae0f88a22b76efb6bfd6be67e24d3097d376

You can either apply the patch manually on your copy of igraph's source code 
and then recompile, get a nightly build from http://code.google.com/p/igraph 
(and compile it of course), or use a simple hack for the time being. The 
problem stems from the fact that we accidentally used an integer to store the 
maximum edge weight instead of a floating-point number, so the maximum weight 
is rounded down when it is stored in this variable. A workaround is simply to 
multiply your weights by a large number (say, 1000000) and then round them. I 
tried it with a multiplicative factor of 1000000 and the result had a 
modularity of 0.0215057.

All the best,
Tamas

On 8 Oct 2013, at 14:10, Lucio Floretta <address@hidden> wrote:

> Hi Tamas,
> 
> I have run the commands you suggested, but the outcome of len(com1) is 1619 
> and not 63. 
> 
> I am working on a MacBook with Snow Leopard, python 2.6.8 and igraph 0.6.5. I 
> installed the library and the python module from macports.
> 
> Thank you for helping,
> Lucio Floretta 
> 
>