small dataset neural network

fun main(args: Array<String>) {
    val seed = 13455.toLong()

    val recordReader = CSVRecordReader(0, ',')
    recordReader.initialize(FileSplit(ClassPathResource("data.csv").file))
    //reader,label index,number of possible labels
    val labelIndex = 0
    val numClasses = 6
    val batchSize = 106

    val iterator = RecordReaderDataSetIterator(recordReader, batchSize, labelIndex, numClasses)

    val allData = iterator.next()
    allData.shuffle(seed)

    val testAndTrain = allData.splitTestAndTrain(0.8)

    val trainingData = testAndTrain.train
    val testData = testAndTrain.test

    val normalizer = NormalizerStandardize()
    normalizer.fit(trainingData)
    normalizer.transform(trainingData)
    normalizer.transform(testData)

    val numInputs = 9
    val outputNum = 6
    val iterations = 1
    val epoches = 100000

    log.info("Building model...")

    val conf = NeuralNetConfiguration.Builder()
            .seed(seed)
            .iterations(iterations)
            .activation(Activation.TANH)
            .weightInit(WeightInit.XAVIER)
            .learningRate(10.0)
            .updater(Updater.ADAGRAD)
            .list()
            .layer(0, DenseLayer.Builder().nIn(numInputs).nOut(2).build())
            .layer(1, OutputLayer.Builder(LossFunctions.LossFunction.NEGATIVELOGLIKELIHOOD)
                    .activation(Activation.SOFTMAX).nIn(2).nOut(outputNum).build())
            .backprop(true)
            .pretrain(false)
            .build()

    val model = MultiLayerNetwork(conf)
    model.init()
    model.setListeners(ScoreIterationListener(100))

    for (i in 1..epoches) {
        model.fit(trainingData)

        if (i % 100 == 0) {
            val eval = Evaluation(6)
            val output = model.output(testData.featureMatrix)
            eval.eval(testData.labels, output)
            log.info(eval.accuracy().toString())
        }


    }

    val eval = Evaluation(6)
    val output = model.output(testData.featureMatrix)
    eval.eval(testData.labels, output)
    log.info(eval.accuracy().toString())
    
    }