nn2.scala

    val iterations = 1
    val seed = 456
    val listenerFreq = 1
    val batchSize = 100
    val numEx = 5000

    val server = UiServer.getInstance();
    log.info("Started on port " + server.getPort());

    log.info("Load data....")
    val iter = new GamesEpoch(batchSize, numEx, 5)

    val h1 = 100
    val h2 = 40
//    val h3 = 40
    val h4 = 40
    log.info("Build model....")
    val conf: MultiLayerConfiguration = new NeuralNetConfiguration.Builder()
      .seed(seed)
      .iterations(iterations)
//      .miniBatch(false)
      .learningRate(1e-3f)
//      .momentum(0.6f)
      .optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT)
    //      .l1(1e-1).l2(2e-4)
      .list(3)
      .layer(0, new DenseLayer.Builder()
        .nIn(50)
        .nOut(h1)
        .weightInit(WeightInit.RELU)
        .activation("relu")
        .updater(Updater.NESTEROVS)
        .build()
    )
      .layer(1, new DenseLayer.Builder()
        .nIn(h1)
        .nOut(h2)
        .weightInit(WeightInit.RELU)
        .activation("relu")
        .updater(Updater.NESTEROVS)
        .build()
    )
    .layer(2, new OutputLayer.Builder(LossFunctions.LossFunction.RMSE_XENT)
      .nIn(h2)
      .nOut(1)
      .weightInit(WeightInit.XAVIER)
      .activation("sigmoid")
      .build()
    )
      .pretrain(false)
      .backprop(true)
      .build()
    val model = new MultiLayerNetwork(conf)
    model.init()

    model.setListeners(new HistogramIterationListener(1), new ScoreIterationListener(listenerFreq))
    log.info("Train model....")
    model.fit(iter)
    /*
    log.info("Evaluate weights....")
    model.getLayers.foreach { case (layer: org.deeplearning4j.nn.api.Layer) =>
      val w: INDArray = layer.getParam(DefaultParamInitializer.WEIGHT_KEY)
      log.info("Weights: " + w)
    }*/

    log.info("Eval data....")
    val test = new GamesIterator(batchSize, numEx).next()

    println(test.getLabels.data().asFloat().toList.take(10))
    println(model.output(test.getFeatureMatrix, Layer.TrainingMode.TEST).data().asFloat().toList.take(10))