Re: [igraph] Weighted graphs in Python interface

[Tamas Nepusz]

Hi,

> 1. Working with the Python interface, I would like to generate a  
> directed weighted graph (without multiple edges and self-loops) from  
> an adjacency matrix, where the elements of the matrix store the  
> weight if an edge exists and are otherwise zero. The matrix is a  
> numpy array or list of lists. How can this be achieved in a most  
> direct way?
I already submitted a patch to the igraph 0.6 dev tree that  
accomplishes this goal (the new graph constructor will be called  
Graph.Weighted_Adjacency). If you are willing to hack around with the  
igraph source and then recompile igraph, I can send you a patch.  
Otherwise, try this simple solution:

edges, weights, vcount = [], [], 0
for v1, line in enumerate(file("adjacency-matrix.dat")):
    vcount += 1
    parts = map(float, line.strip().split())
    for v2, weight in parts:
        if v2 > 0:
            edges.append((v1,v2))
            weights.append(weight)

g = Graph(vcount, edges, directed=True)
g.es["weight"] = weights

I hope this helps. I haven't tried it, but the basic idea should be  
correct.

> The weights are all positive in my application. Is this implemented  
> yet in iGraph?
Weighted shortest path lengths will be implemented in igraph 0.6.  
Weighted closeness centrality is not implemented directly in the dev  
tree yet, but I think it's not too difficult to calculate using the  
matrix returned by the weighted shortest path calculations. Weighted  
betweenness centrality seems more complicated - I'm not familiar with  
the implementation details of igraph_betweenness, maybe Gabor can  
comment on it.

> If not, do you have suggestions for other fast libraries  
> implementing these features?
I think the Boost Graph Library implements weighted betweenness  
centrality:
http://www.boost.org/doc/libs/1_35_0/libs/graph/doc/betweenness_centrality.html

--
Tamas