gistfile1.scala

 val Q = DenseMatrix.ones[Double](numRegions, numClasses) //TODO maybe initialize to 1/num neighbours or 1/numneighbours*numClass
    //TODO This Q could be stored in a spark RDD if it gets too large for memory;; Prob not necesary because thetaUnayr is the same size 
    //TODO consider storing the probabilities in a byte
    if (debug) {
      //val header = (0 until numClasses).toSeq.toList.flatMap { curclass => (0 until numRegions).toSeq.toList.map { curregion => ",Qr" + curregion + " Qc" + curclass } }
      // println("#MF_Q_LOG#,timeID,elapsedTime" + header)
      // println("#MF_E_LOG#,timeID,logTag,iter,maxE,minE,maxE_change,minE_change")
    }

    var lastMaxE = 0.0
    var lastMinE = 0.0
    var numNoChange = 0
    for (iter <- 0 until maxIterations) {

      var numUnchangedQs = 0
      val lastQ = Q;
      val xiLab = (0 until numClasses).par
      //val xis = (0 until graph.size).par

      val allXiperLabel = xiLab.map(curLab => ((curLab,
        for (xi <- 0 until graph.size) yield {

          val neigh = graph.getC(xi).toArray
          val allClasses = (0 until numClasses).toList

          val newQest = neigh.toList.map { neighIdx =>
            allClasses.foldLeft(0.0) { (running, curClass) =>
              {
                running + Math.exp(lastQ(neighIdx, curClass) * (if (DISABLE_PAIRWISE) 0 else thetaPairwise(curClass, curLab))) * Math.exp((1 / temp) * thetaUnary(xi, curLab))
              }
            }
          }.sum

          (1 / temp) * newQest
        })))

      for (labAgain <- 0 until numClasses) {
        val allXi = allXiperLabel(labAgain)._2.toArray
        Q(::, labAgain) := DenseVector(allXi)
      }
      //allXiperLabel.foreach((label:Int,listofQs:IndexedSeq[Double])=>Q(::,label):=DenseVector(listofQs.toArray))