Ciao Peter,
It's not easy to provide you a test case suitable for debugging ...
I'm running many test to complete my thesis (I'm citing neo4j) on some
natural language processing problem. My dataset is very large. Let me
think about it ....
If you can access the sources of 0.3 stable version, you can see these lines:
// If the first value is negative (possibly the whole vector), negate
// the whole vector
if ( values.get( nodeSet.iterator().next() ) < 0 )
{
for ( Node node : nodeSet )
{
values.put( node, -values.get( node ) );
}
}
That's always true? if the first one is negative, then you should
always negate the whole vector??
I'll make a try with the 0.6-snapshot version and I'll let you know.
The last thing: the power method is not guaranted to converge always
... especially if the graph is directed (neo4j is always directed).
Then a method like Arnoldi Iteration is necessary IMHO.
I've developed the simplest centrality measure, the degree centrality
and I should develop PageRank and/or HITS (not now ..). Maybe you can
be interested in these algos (for now the degree centrality). Let me
know.
Best regards,
Lorenzo
On Sun, May 30, 2010 at 8:56 PM, Peter Neubauer
<[email protected]> wrote:
> Lorenzo,
> I think the lesser used algos are in need of seeing over. Could you
> provide a failing testcase that can be fixed in order to get the
> Eigenvector into shape? Would be great in order to get the problem
> pinned down while you are at it!
>
> Cheers,
>
_______________________________________________
Neo4j mailing list
[email protected]
https://lists.neo4j.org/mailman/listinfo/user
