Bug in eigenvector centrality function?
Hi, I have a network with 500 nodes and wanted to calculate the eigenvector centrality with graph_tool. First it hung up (or took basically forever.) Then I decided to try with scipy.sparse.linalg.eigsh On another network with 500 nodes as well (both newman-watts small-world topology) The outcome for the eigenvalue (should be the largest ev of the adjacency matrix, if I understand right) is not the same with gt.centrality.eigenvector and scipy.sparse.linalg.eigsh. If I calculate all evs of the adjacency matrix with scipy.linalg (not sparse here) the largest one is the one that scipy gets as well. As for the eigenvector, the eigenvectors that are found seem to be roughly the same |v1-v2|\approx 2e-7 Attached is an archive that contains a script which demonstrates the behavior as well as the two example graphs.