TrainExample.java

public class TrainExample {

	public static final int BATCH_SIZE = 100;
	public static final int N_EPOCHS = 150;


	public static void main(String... args) {
		// define some paths where you would like to store the model
		String modelPath = "data" + File.separator + "models" + File.separator;
		String normalizerPath = "data" + File.separator + "normalizers" + File.separator;

		String modelName = "train.model";
		String configName = "config.json";

		// define some paths to the files which will be used for training
		String trainFeaturePath = "data" + File.separator + "trainTestSplit" + File.separator +
				"train" + File.separator + "features" + File.separator;
		String trainLabelPath = "data" + File.separator + "trainTestSplit" + File.separator +
				"train" + File.separator + "labels" + File.separator;

		// there are 31 files/examples
		// for the sake of the example, maybe generate some random time series example files with integer values?
		Integer maxFieldId = 30;

		// setup helper for serialization
		SerializationUtils serializationHelper = new SerializationUtils(modelPath, modelName, normalizerPath, configName);

		// init network from config
		MultiLayerConfiguration config = serializationHelper.loadNetworkConfig();

		MultiLayerNetwork net = new MultiLayerNetwork(config);
		net.init();
		net.setListeners(new ScoreIterationListener(1000));


		// load data for training
		SequenceRecordReader trainFeatures = new CSVSequenceRecordReader();
		SequenceRecordReader trainLabels = new CSVSequenceRecordReader();

		try {
			File featuresDirTrain = new File(trainFeaturePath);
			File labelsDirTrain = new File(trainLabelPath);

			trainFeatures.initialize(new NumberedFileInputSplit(featuresDirTrain.getAbsolutePath() + "/%d.csv", 0, maxFieldId));
			trainLabels.initialize(new NumberedFileInputSplit(labelsDirTrain.getAbsolutePath() + "/%d.csv", 0, maxFieldId));
		} catch (Exception e) {
			// log error
		}

		boolean regression = true;
		int numClasses = -1; //not used for regression

		DataSetIterator trainingData = new SequenceRecordReaderDataSetIterator(
				trainFeatures,
				trainLabels,
				BATCH_SIZE,
				numClasses,
				regression,
				SequenceRecordReaderDataSetIterator.AlignmentMode.ALIGN_END);

		// setup normalizer
		NormalizerStandardize normalizer = new NormalizerStandardize();
		normalizer.fitLabel(true);
		normalizer.fit(trainingData);

		trainingData.reset();
		trainingData.setPreProcessor(normalizer);
		// store it
		serializationHelper.storeNormalizer(normalizer);

		// train
		for (int j = 0; j < N_EPOCHS; j++) {
			trainingData.reset();
			net.fit(trainingData);
		}

		// store network model
		serializationHelper.storeNetworkModel(net);

		// Successfully created and trained the NN model
	}
	
}