Bug of Aync parameter merger

["Edward J. Yoon" <ed...@samsung.com>]

Today I've tried to test Aync parameter merger on my cluster but it throws
below exeptions. The reason is maybe that the objects are not writable.

// exchange parameter update with master
    SmallLayeredNeuralNetworkMessage inMessage =
proxy.merge(avgTrainingError,
        weightUpdates, this.inMemoryModel.getWeightMatrices());

16/04/18 13:44:37 ERROR bsp.BSPTask: Error running bsp setup and bsp
function.
java.lang.reflect.UndeclaredThrowableException
	at com.sun.proxy.$Proxy5.merge(Unknown Source)
	at
org.apache.horn.bsp.SmallLayeredNeuralNetworkTrainer.calculateUpdates(SmallL
ayeredNeuralNetworkTrainer.java:203)
	at
org.apache.horn.bsp.SmallLayeredNeuralNetworkTrainer.bsp(SmallLayeredNeuralN
etworkTrainer.java:157)
	at org.apache.hama.bsp.BSPTask.runBSP(BSPTask.java:171)
	at org.apache.hama.bsp.BSPTask.run(BSPTask.java:144)
	at
org.apache.hama.bsp.GroomServer$BSPPeerChild.main(GroomServer.java:1255)
Caused by: java.io.IOException: Call to tserver/127.0.1.1:40052 failed on
local exception: java.io.IOException: Can't write: [-9.15038855924827E-4,
-6.59978486972921E-5, -3.2776380633814714E-4, -4.908153521598804E-4,
-1.6535398987379825E-4, -2.481217176171659E-5, -3.097885126985827E-4,
-4.5677556953153244E-5, -2.4586775269751806E-5]
[0.007494058256877595, 5.350221684534322E-4, 0.002670209123842312,
0.004006823375780923, 0.0013388943782793772, 1.9042423341881118E-4,
0.002527834024659683, 3.6902914343781094E-4, 2.1355016559231787E-4]
[0.003978872861757267, 2.894127056140916E-4, 0.0014220632229977248,
0.0021264410749675996, 7.121792101231689E-4, 9.615850486945849E-5,
0.001346404836613541, 1.983544516681021E-4, 1.1456125145256654E-4]
[-0.00695218464799201, -5.06352811206525E-4, -0.0024558201932702453,
-0.0037067439613160842, -0.001223792213555592, -1.3795325608127702E-4,
-0.002327019982822582, -3.333356053453033E-4, -2.2891282141905694E-4]
[0.0055693438721696985, 3.880990234702207E-4, 0.0019653686596401037,
0.002976654095812432, 9.959956034508214E-4, 1.3512799657517912E-4,
0.0018675147363488187, 2.6864207467334106E-4, 1.466807720784203E-4]
[-0.004406974432832138, -3.253702689795474E-4, -0.0015839399468065797,
-0.002365407518184666, -7.935665906444789E-4, -1.2452109168041786E-4,
-0.0014916074949176499, -2.241789663094838E-4, -1.2922820566750723E-4]
[-0.007248671217487145, -5.286530805287898E-4, -0.0025899737055120175,
-0.003882832442858956, -0.001298906448842776, -1.8706905596266497E-4,
-0.002449470417876311, -3.6278584202388065E-4, -2.0959879825765203E-4]
[0.004184985972170336, 3.028041744683768E-4, 0.0015042967415655265,
0.0022373908691310524, 7.448922331540512E-4, 1.0410505256316024E-4,
0.0014161086335274698, 2.189835200884028E-4, 1.1446038058615025E-4]
 as interface org.apache.hama.commons.math.DoubleMatrix
	at org.apache.hama.ipc.Client.wrapException(Client.java:945)
	at org.apache.hama.ipc.Client.call(Client.java:913)
	at org.apache.hama.ipc.RPC$Invoker.invoke(RPC.java:239)

And also, with below code, the master task will be just finished as soon as
it starts.

@Override
  public void bsp(
      BSPPeer<LongWritable, VectorWritable, NullWritable, NullWritable,
SmallLayeredNeuralNetworkMessage> peer)
      throws IOException, SyncException, InterruptedException {
    if (!isMaster(peer)) {
      while (!this.isConverge.get()) {
        // each slave-worker calculate the matrices updates according to
local data
        // and merge them with master
        calculateUpdates(peer);
      }
    }
  }

I'll fix and commit them to the master directly today.

--
Best Regards, Edward J. Yoon