Re: SimplePageRankVertex implementation, dangling nodes and sending messages to all nodes...

2012-05-18 Thread Sebastian Schelter
Accumulation should be possible with aggregators and the idea with applying preference vectors seems very interesting too. Maybe these should be implemented as additional features in GIRAPH-191? On 18.05.2012 11:24, Gianmarco De Francisci Morales wrote: Could you just accumulate the PR of

Re: SimplePageRankVertex implementation, dangling nodes and sending messages to all nodes...

2012-05-18 Thread Sebastian Schelter
Hi Paolo, Very good points, could you add them to GIRAPH-191? I think that the convergence check does not need an extra superstep. It is sufficient to compute the L1-norm of the current pageRank vector, which can be done by an aggregator and compare it to the previous aggregated value. I guess

Re: SimplePageRankVertex implementation, dangling nodes and sending messages to all nodes...

2012-05-18 Thread Paolo Castagna
Hi Sebastian Sebastian Schelter wrote: Very good points, could you add them to GIRAPH-191? Done. Summarized with a link to this thread. I think that the convergence check does not need an extra superstep. It is sufficient to compute the L1-norm of the current pageRank vector, which can be

Re: SimplePageRankVertex implementation, dangling nodes and sending messages to all nodes...

2012-05-17 Thread Paolo Castagna
Hi Sebastian, from SimplePageRankVertex.java, we have: if (getSuperstep() MAX_SUPERSTEPS) { long edges = getNumOutEdges(); sendMsgToAllEdges( new DoubleWritable(getVertexValue().get() / edges)); } else { voteToHalt(); } What happens if getNumOutEdges()