I have recently encountered a somewhat weird behaviour when calculating eigenvector centralities (with the evcent function). I have a graph with ~600k edges and ~6k nodes, and I use the Lgl format as my input. I noticed a strange difference in eigenvector centralities depending on minor differences in the format.
If I specify my edges as follows:
'# node1' 'node2 weight2' 'node3 weight3' ...
than, not surprisingly, many of my nodes will have zero eigenvector centrality.
However, if I put no space between # and node 1 (’#node1’), than most of the previously null values will be approx. 2.7e-18. Other centralities (betweenness, closeness, degree, transitivity) are unaffected, and in case of small, “test” graphs this does not happen.
Another issue is that I have noticed that the eigenvector centralities are calculated somewhat differently by iGraph than by, for example NetworkX. In the following graph (see below, Lgl format) node 3 is disconnected from the rest. NetworkX assigns an eigenvector centrality 0 to this node, while iGraph ~0.35, and I wonder what might be the cause of the difference.
(g.evcent(directed=False, scale=True, weights=Weights, return_eigenvalue=False))
I would greatly appreciete any help or suggestion with these issues.
#---------------- #3 #0 1 2 4 #1 2 4 #2 5 #4 5