[igraph] Re: igraph

[Gábor Csárdi]

Agus,

three methods can handle edge weights: fastgreedy.community expects
non-negative values for the weights, it checks that this holds;
walktrap.community does not make any checks, but in the paper the
authors state that the weights should be positive and we use their
implementation; spinglass.community accepts any values for the weights
and there is nothing about this in the paper; according to my quick
checks, however, it does not handle negative weights properly.

Best,
Gabor

On Mon, Nov 3, 2008 at 6:31 PM, Agustin Lobo <address@hidden> wrote:
> Thanks,
>
> do the weights have to fulfill any particular
> requirement for each method?
>
> Agus
>
> Gábor Csárdi wrote:
>>
>> Dear Agus,
>>
>> currently the following community detection algorithms are implemented
>> in igraph:
>>
>> fastgreedy.community
>> walktrap.community
>> spinglass.community
>> edge.betweenness.community
>> leading.eigenvector.community
>>
>> Please see their manual pages for references.
>>
>> spinglass.community and leading.eigenvector.community returns the
>> communities themselves, the other three methods return dendrograms,
>> you need to cut these to get the actual communities. A fairly standard
>> way is to cut at the highest modularity value, the modularity values
>> are returned by the fastgreedy and the walktrap methods (if you give
>> the modularity=TRUE argument). For the edge betweenness based method
>> you have calculate them by calling 'modularity'. See the example at
>> http://igraph.sf.net -> Screenshots -> Creating animations.
>>
>> These methods use different heuristics to maximize modularity, if you
>> get consistent results with the different methods that is a good sign.
>>
>> Btw. these questions are more suited to the igraph-help mailing list,
>> see http://igraph.sf.net -> Community -> igraph-help
>>
>> Best Regards,
>> Gabor
>>
>> On Mon, Nov 3, 2008 at 2:44 PM, Agustin Lobo <address@hidden> wrote:
>>>
>>> Dear Gabor Csardi,
>>>
>>> I've been trying your R package igraph, but feel a bit
>>> lost in the methods, so I'm taking the liberty of
>>> seeking your advice.
>>> I have a matrix of transportation (nb of trips, T) among cities. I've
>>> defined
>>>
>>> t(i,j) = T(i,j)/sum(t(i,.)
>>>
>>> Then I've created a graph with:
>>> g3 <- graph.adjacency(fluxmetr.contir, weighted=TRUE,mode="directed")
>>>
>>> My goal is to find communities defined by those cities that most
>>> interact.
>>> I've tried
>>> g3w <- walktrap.community(g3)
>>> community.to.membership(g3, g3w$merges, steps=9)
>>>
>>> but I'm not really understanding what I'm doing. Could you point me
>>> to some references and/or advice on which method would be best suited for
>>> my
>>> goal?
>>>
>>> Thanks!
>>>
>>> Agus
>>>
>>> --
>>> Dr. Agustin Lobo
>>> Institut de Ciencies de la Terra "Jaume Almera" (CSIC)
>>> LLuis Sole Sabaris s/n
>>> 08028 Barcelona
>>> Spain
>>> Tel. 34 934095410
>>> Fax. 34 934110012
>>> email: address@hidden
>>> http://www.ija.csic.es/gt/obster
>>>
>>
>>
>>
>
> --
> Dr. Agustin Lobo
> Institut de Ciencies de la Terra "Jaume Almera" (CSIC)
> LLuis Sole Sabaris s/n
> 08028 Barcelona
> Spain
> Tel. 34 934095410
> Fax. 34 934110012
> email: address@hidden
> http://www.ija.csic.es/gt/obster
>



-- 
Gabor Csardi <address@hidden>     UNIL DGM