Actually I meant - currently the graph package doesn't support convergence (within some tolerance) as in the case of PR. Should the graph package be modified to support it?
Praveen On Wed, Apr 18, 2012 at 1:22 PM, Edward J. Yoon <[email protected]>wrote: > Yes, boolean is enough. > > In PR example, iteration 30 is just a arbitrary number. I had just > copy and paste Pregel example. Actually, program should run until > convergence is achieved. If you are saying about unit test of > PageRank, then +1. > > On Wed, Apr 18, 2012 at 4:22 PM, Praveen Sripati > <[email protected]> wrote: > > 1) I was looking in the ShortestPath#bsp() (1) on when the program gets > out > > the main while loop. Not sure if I understood it properly, but it looks a > > bit complicated without much comments. Similar logic has been applied in > > the org.apache.hama.graph.GraphJobRunner#bsp() also. > > > > Basically, each of the bsp task should send a boolean to a master > depending > > on any messages have been sent to other peers or not. The master if there > > is atleast one true from any one of the peer, will send a true back to > all > > the peers to continue the processing. Or else the bsp task can send a > > boolean to all the peers instead of the master to avoid an additional > > sync(). It is as simple as that. > > > > The number of updates vertex which have been updated should not matter. > > What is the purpose of the updatesMade and the globalUpdateCounts integer > > variables? > > > > 2) Also, in the PageRank(2) the convergence is not tested, the program > just > > goes through the loop maxIteration # of times. Is a test for convergence > > required in the graph package? > > > > Praveen > > > > (1) - > > > https://svn.apache.org/repos/asf/incubator/hama/branches/0.4/examples/src/main/java/org/apache/hama/examples/ShortestPaths.java > > (2) - > > > https://svn.apache.org/repos/asf/incubator/hama/trunk/examples/src/main/java/org/apache/hama/examples/PageRank.java > > > > -- > Best Regards, Edward J. Yoon > @eddieyoon >
