gistfile1.txt

public class ImageCNNDummy {
    private static class SimplePreProcessor implements DataSetPreProcessor {
        @Override
        public void preProcess(org.nd4j.linalg.dataset.api.DataSet toPreProcess) {
            toPreProcess.getFeatureMatrix().divi(255);  //[0,255] -> [0,1] for input pixel values
        }
    }

    // Images are of format given by allowedExtension -
    protected static final String[] allowedExtensions = BaseImageLoader.ALLOWED_FORMATS;

    protected static final long seed = 12345;

    public static final Random randNumGen = new Random(seed);



    public static void main(String[] args) throws Exception {

        int nChannels = 1;
        int outputNum = 3;
        int batchSize = 64;
        int nEpochs = 10;
        int iterations = 1;
        int seed = 123;
        int width = 100;
        int height = 100;
        int channels = 3;

        System.out.println("####");

        System.out.println("Hello World!");

        System.out.println("Load data....");

        File parentDir = new File(System.getProperty("user.dir"), "CNNTest/");
        FileSplit filesInDir = new FileSplit(parentDir, allowedExtensions, randNumGen);
        ParentPathLabelGenerator labelMaker = new ParentPathLabelGenerator();
        BalancedPathFilter pathFilter = new BalancedPathFilter(randNumGen, allowedExtensions, labelMaker);
        InputSplit[] filesInDirSplit = filesInDir.sample(pathFilter, 100, 0);
        InputSplit trainData = filesInDirSplit[0];
        InputSplit testData = filesInDirSplit[1];

        ImageRecordReader recordReader = new ImageRecordReader(height, width, channels, labelMaker);

        recordReader.initialize(trainData);
        DataSetIterator dataIter = new RecordReaderDataSetIterator(recordReader, batchSize, -1, outputNum);

        SimplePreProcessor SP = new SimplePreProcessor();
        //NOTE:
        //The normalization was not done correctly.
        //The net is fit with an iterator. You have to set the preprocessor you want to use to iterator as below.
        //There are in an example that explains use of the preprocessor with both datasets and the iterator.
        //Take a looksie:
        //https://github.com/deeplearning4j/dl4j-0.4-examples/blob/master/src/main/java/org/deeplearning4j/examples/dataExamples/PreprocessNormalizerExample.java
        dataIter.setPreProcessor(SP);
        //In the while loop you were create a dataset, ds and then modifying it. This does not have any impact on the iterator.
        //and this is what gets passed to you net.
        /*while (dataIter.hasNext()) {
            DataSet ds = dataIter.next();
            SP.preProcess(ds);
        }*/

        dataIter.reset();

        System.out.println("Build model....");
        MultiLayerConfiguration.Builder builder = new NeuralNetConfiguration.Builder().seed(seed).iterations(iterations)
            .regularization(true).l2(0.0005).learningRate(0.0000000000001)// .biasLearningRate(0.02)
            // .learningRateDecayPolicy(LearningRatePolicy.Inverse).lrPolicyDecayRate(0.001).lrPolicyPower(0.75)
            .weightInit(WeightInit.XAVIER).optimizationAlgo(OptimizationAlgorithm.STOCHASTIC_GRADIENT_DESCENT)
            .updater(Updater.NESTEROVS).momentum(0.9).list()
            .layer(0,
                new ConvolutionLayer.Builder(5, 5).nIn(nChannels).stride(1, 1).nOut(20).activation("identity")
                    .build())
            .layer(1,
                new SubsamplingLayer.Builder(SubsamplingLayer.PoolingType.MAX).kernelSize(2, 2).stride(2, 2)
                    .build())
            .layer(2,
                new ConvolutionLayer.Builder(5, 5).nIn(nChannels).stride(1, 1).nOut(50).activation("identity")
                    .build())
            .layer(3,
                new SubsamplingLayer.Builder(SubsamplingLayer.PoolingType.MAX).kernelSize(2, 2).stride(2, 2)
                    .build())
            .layer(4, new DenseLayer.Builder().activation("relu").nOut(500).build())
            .layer(5, new OutputLayer.Builder(LossFunctions.LossFunction.NEGATIVELOGLIKELIHOOD).nOut(outputNum)
                .activation("softmax").build())
            .backprop(true).pretrain(false);
        new ConvolutionLayerSetup(builder, width, height, channels);

        MultiLayerConfiguration conf = builder.build();
        MultiLayerNetwork model = new MultiLayerNetwork(conf);
        model.init();

        System.out.println("Train model....");
        model.setListeners(new ScoreIterationListener(1));
        for (int i = 0; i < nEpochs; i++) {
            model.fit(dataIter);
            System.out.println("*** Completed epoch {" + i + "} ***");

            System.out.println("Evaluate model....");
            //NOTE: After the fit, you need to reset the iterator if you want to
            //get data from it for eval
            dataIter.reset();

            Evaluation eval = new Evaluation(outputNum);
            while (dataIter.hasNext()) {
                DataSet ds = dataIter.next();
                INDArray output = model.output(ds.getFeatureMatrix(), false);
                eval.eval(ds.getLabels(), output);
            }
            System.out.println(eval.stats());
            //NOTE: You have to reset the iterator again since you want to run multiple epochs
            dataIter.reset();
        }
        System.out.println("****************Example finished********************");
    }
}