Java MomentumBackpropagation类代码示例

import org.neuroph.nnet.learning.MomentumBackpropagation; //导入依赖的package包/类
@Override
public void doRun() {
    try {
        System.out.println("Starting training thread....." + sampleDimension.toString() + " and " + imageLabels.toString());

        HashMap<String, BufferedImage> imagesMap = new HashMap<String, BufferedImage>();
        for (File file : srcDirectory.listFiles()) {
            imageLabels.add(FilenameUtils.removeExtension(file.getName()));
            if (sampleDimension.getWidth() > 0 && sampleDimension.getHeight() > 0) {
                Double w = sampleDimension.getWidth();
                Double h = sampleDimension.getHeight();
                imagesMap.put(file.getName(), ImageUtilities.resizeImage(ImageUtilities.loadImage(file), w.intValue(), h.intValue()));
            }
        }
        Map<String, FractionRgbData> imageRgbData = ImageUtilities.getFractionRgbDataForImages(imagesMap);
        DataSet learningData = ImageRecognitionHelper.createRGBTrainingSet(imageLabels, imageRgbData);

        nnet = NeuralNetwork.load(new FileInputStream(nnFile)); //Load NNetwork
        MomentumBackpropagation mBackpropagation = (MomentumBackpropagation) nnet.getLearningRule();
        mBackpropagation.setLearningRate(learningRate);
        mBackpropagation.setMaxError(maxError);
        mBackpropagation.setMomentum(momentum);

        System.out.println("Network Information\nLabel = " + nnet.getLabel()
                + "\n Input Neurons = " + nnet.getInputsCount()
                + "\n Number of layers = " + nnet.getLayersCount()
        );

        mBackpropagation.addListener(this);
        System.out.println("Starting training......");
        nnet.learn(learningData, mBackpropagation);
        //Training Completed
        listener.batchImageTrainingCompleted();
    } catch (FileNotFoundException ex) {
        System.out.println(ex.getMessage() + "\n" + ex.getLocalizedMessage());
    }

}
 

import org.neuroph.nnet.learning.MomentumBackpropagation; //导入依赖的package包/类
/**
 * Creates and returns a new instance of Multi Layer Perceptron
 * @param layersStr space separated number of neurons in layers
 * @param transferFunctionType transfer function type for neurons
 * @return instance of Multi Layer Perceptron
 */
public static MultiLayerPerceptron createMLPerceptron(String layersStr, TransferFunctionType transferFunctionType, Class learningRule,  boolean useBias, boolean connectIO) {
	ArrayList<Integer> layerSizes = VectorParser.parseInteger(layersStr);
               NeuronProperties neuronProperties = new NeuronProperties(transferFunctionType, useBias);
	MultiLayerPerceptron nnet = new MultiLayerPerceptron(layerSizes, neuronProperties);
               
               // set learning rule - TODO: use reflection here
               if (learningRule.getName().equals(BackPropagation.class.getName()))  {
                   nnet.setLearningRule(new BackPropagation());
               } else if (learningRule.getName().equals(MomentumBackpropagation.class.getName())) {
                   nnet.setLearningRule(new MomentumBackpropagation());
               } else if (learningRule.getName().equals(DynamicBackPropagation.class.getName())) {
                   nnet.setLearningRule(new DynamicBackPropagation());
               } else if (learningRule.getName().equals(ResilientPropagation.class.getName())) {
                   nnet.setLearningRule(new ResilientPropagation());
               } 

               // connect io
               if (connectIO) {
                   nnet.connectInputsToOutputs();
               }

	return nnet;
}
 

import org.neuroph.nnet.learning.MomentumBackpropagation; //导入依赖的package包/类
/**
 * Instantiates a new animal network.
 *
 * @param input the input
 * @param hidden the hidden
 * @param output the output
 */
public AnimalNetwork(int input,int hidden,int output) {
	super();
	System.out.println("network is created");
	initializeNeurons();
	initializeQuestions();
	animal_network = new MultiLayerPerceptron(TransferFunctionType.SIGMOID,Data.INPUTUNITS,Data.HIDDENUNITS,Data.OUTPUTUNITS);
	animal_network.setNetworkType(NeuralNetworkType.MULTI_LAYER_PERCEPTRON);
	animal_network.randomizeWeights();  //randomize weights 
	// set parameters 
	((LMS) animal_network.getLearningRule()).setMaxError(MAXERROR);//0-1 
	((LMS) animal_network.getLearningRule()).setLearningRate(LEARNINGRATE);//0-1
	((LMS) animal_network.getLearningRule()).setMaxIterations(MAXITERATIONS);//0-1
	MomentumBackpropagation backpropogation = new MomentumBackpropagation(); // define momentum
	backpropogation.setMomentum(0.7); // set momentum
	animal_network.setLearningRule(backpropogation); 
}
 

import org.neuroph.nnet.learning.MomentumBackpropagation; //导入依赖的package包/类
public void startLearning() {
    Thread t1 = new Thread(new Runnable() {
        public void run() {
            console.addLog("Loading test set");
            testSet = loader.loadDataSet(testSetPath);
            console.addLog("Test set loaded");

            console.addLog("Loading training set");
            trainingSet = loader.loadDataSet(trainingSetPath);
            console.addLog("Training set loaded. Input size: " + trainingSet.getInputSize() +
                    " Output size: " + trainingSet.getOutputSize());

            nnet = new MultiLayerPerceptron(TransferFunctionType.SIGMOID,
                    trainingSet.getInputSize(), 86, 86, trainingSet.getOutputSize());

            MomentumBackpropagation backPropagation = new MomentumBackpropagation();
            backPropagation.setLearningRate(learningRate);
            backPropagation.setMomentum(momentum);

            LearningTestSetEvaluator evaluator =
                    new LearningTestSetEvaluator(nnetName, testSet, trainingSet, console);
            backPropagation.addListener(evaluator);
            backPropagation.addListener(new LearningEventListener() {
                @Override
                public void handleLearningEvent(LearningEvent event) {
                    if (event.getEventType() == LearningEvent.Type.LEARNING_STOPPED) {
                        listeners.forEach((listener) -> listener.learningStopped(LearningNetTask.this));
                    }
                }
            });
            nnet.setLearningRule(backPropagation);
            console.addLog("Started neural net learning with momentum: "
                    + momentum + ", learning rate: " + learningRate);
            nnet.learnInNewThread(trainingSet);
        }
    });
    t1.start();
}
 

import org.neuroph.nnet.learning.MomentumBackpropagation; //导入依赖的package包/类
private static void learnNeuralNet(DataSet trainingSet, DataSet testSet) {
    TestSetEvaluator testEvaluator = new TestSetEvaluator(NNET_NAME, testSet, trainingSet);
    MultiLayerPerceptron nnet = new MultiLayerPerceptron(TransferFunctionType.SIGMOID, INPUT_LAYER, 86, 86, OUTPUT_LAYER);

    MomentumBackpropagation bp = new MomentumBackpropagation();
    bp.setLearningRate(LEARINING_RATE);
    bp.setMomentum(MOMENTUM);
    bp.addListener(testEvaluator);

    nnet.setLearningRule(bp);
    nnet.learn(trainingSet);
    nnet.save(NNET_NAME + "last");
}
 

import org.neuroph.nnet.learning.MomentumBackpropagation; //导入依赖的package包/类
private static void learnNeuralNet(DataSet trainingSet, DataSet testSet) {
    TestSetEvaluator testEvaluator = new TestSetEvaluator(NNET_NAME, testSet, trainingSet);
    MultiLayerPerceptron nnet = new MultiLayerPerceptron(TransferFunctionType.SIGMOID, INPUT_LAYER, 140, OUTPUT_LAYER);

    MomentumBackpropagation bp = new MomentumBackpropagation();
    bp.setLearningRate(LEARINING_RATE);
    bp.setMomentum(MOMENTUM);
    bp.addListener(testEvaluator);

    nnet.setLearningRule(bp);
    nnet.learn(trainingSet);
    nnet.save(NNET_NAME + "last");
}
 

import org.neuroph.nnet.learning.MomentumBackpropagation; //导入依赖的package包/类
private static void learnNeuralNet(DataSet trainingSet, DataSet testSet) {
    TestSetEvaluator testEvaluator = new TestSetEvaluator(NNET_NAME, testSet, trainingSet);
    MultiLayerPerceptron nnet = new MultiLayerPerceptron(TransferFunctionType.SIGMOID, INPUT_LAYER, 76, 76, OUTPUT_LAYER);

    MomentumBackpropagation bp = new MomentumBackpropagation();
    bp.setLearningRate(LEARINING_RATE);
    bp.setMomentum(MOMENTUM);
    bp.addListener(testEvaluator);

    nnet.setLearningRule(bp);
    nnet.learn(trainingSet);
    nnet.save(NNET_NAME + "last");
}
 

import org.neuroph.nnet.learning.MomentumBackpropagation; //导入依赖的package包/类
@Override
public void doRun() {
    HashMap<String, BufferedImage> imagesMap = new HashMap<String, BufferedImage>();
    String fileName = "";
    if (!isReal) {
        fileName = "real";
    } else {
        fileName = "faked";
    }

    System.out.println("Teaching as " + fileName);
    imagesMap.put(fileName, image);
    Map<String, FractionRgbData> imageRgbData = ImageUtilities.getFractionRgbDataForImages(imagesMap);
    DataSet learningData = ImageRecognitionHelper.createRGBTrainingSet(labels, imageRgbData);
    MomentumBackpropagation mBackpropagation = (MomentumBackpropagation) nnet.getLearningRule();
    mBackpropagation.setLearningRate(learningRate);
    mBackpropagation.setMaxError(maxError);
    mBackpropagation.setMomentum(momentum);

    System.out.println("Network Information\nLabel = " + nnet.getLabel()
            + "\n Input Neurons = " + nnet.getInputsCount()
            + "\n Number of layers = " + nnet.getLayersCount()
    );

    mBackpropagation.addListener(this);
    System.out.println("Starting training......");
    nnet.learn(learningData, mBackpropagation);

    //Mark nnet as dirty. Write on close
    isDirty = true;
}
 

import org.neuroph.nnet.learning.MomentumBackpropagation; //导入依赖的package包/类
/**
 * Runs this sample
 */
public void run() {
	
    // create training set (logical XOR function)
    DataSet trainingSet = new DataSet(2, 1);
    trainingSet.addRow(new DataSetRow(new double[]{0, 0}, new double[]{0}));
    trainingSet.addRow(new DataSetRow(new double[]{0, 1}, new double[]{1}));
    trainingSet.addRow(new DataSetRow(new double[]{1, 0}, new double[]{1}));
    trainingSet.addRow(new DataSetRow(new double[]{1, 1}, new double[]{0}));

    // create multi layer perceptron
    MultiLayerPerceptron myMlPerceptron = new MultiLayerPerceptron(TransferFunctionType.TANH, 2, 3, 1);
    // enable batch if using MomentumBackpropagation
    if( myMlPerceptron.getLearningRule() instanceof MomentumBackpropagation ){
    	((MomentumBackpropagation)myMlPerceptron.getLearningRule()).setBatchMode(true);
    	((MomentumBackpropagation)myMlPerceptron.getLearningRule()).setMaxError(0.00001);
    }

    LearningRule learningRule = myMlPerceptron.getLearningRule();
    learningRule.addListener(this);
    
    // learn the training set
    System.out.println("Training neural network...");
    myMlPerceptron.learn(trainingSet);

    // test perceptron
    System.out.println("Testing trained neural network");
    testNeuralNetwork(myMlPerceptron, trainingSet);

    // save trained neural network
    myMlPerceptron.save("myMlPerceptron.nnet");

    // load saved neural network
    NeuralNetwork loadedMlPerceptron = NeuralNetwork.load("myMlPerceptron.nnet");

    // test loaded neural network
    System.out.println("Testing loaded neural network");
    testNeuralNetwork(loadedMlPerceptron, trainingSet);
}
 

import org.neuroph.nnet.learning.MomentumBackpropagation; //导入依赖的package包/类
/**
 * Benchmrk preparation consists of training set and neural networ creatiion.
 * This method generates training set with 100 rows, where every row has 10 input and 5 output elements
 * Neural network has two hiddden layers with 8 and 7 neurons, and runs learning rule for 2000 iterations
 */
@Override
public void prepareTest() {
    int trainingSetSize = 100;
    int inputSize = 10;
    int outputSize = 5;
    
    this.trainingSet = new DataSet(inputSize, outputSize);
    
    for (int i = 0; i < trainingSetSize; i++) {
        double input[] = new double[inputSize];
        for( int j=0; j<inputSize; j++)
            input[j] = Math.random();

        double output[] = new double[outputSize];
        for( int j=0; j<outputSize; j++)
            output[j] = Math.random();            
        
        DataSetRow trainingSetRow = new DataSetRow(input, output);
        trainingSet.addRow(trainingSetRow);
    }
    
    
    network = new MultiLayerPerceptron(inputSize, 8, 7, outputSize);
    ((MomentumBackpropagation)network.getLearningRule()).setMaxIterations(2000);
    
}
 

import org.neuroph.nnet.learning.MomentumBackpropagation; //导入依赖的package包/类
/**
 * Creates MultiLayerPerceptron Network architecture - fully connected
 * feed forward with specified number of neurons in each layer
 * 
 * @param neuronsInLayers
 *            collection of neuron numbers in getLayersIterator
 * @param neuronProperties
 *            neuron properties
 */
private void createNetwork(List<Integer> neuronsInLayers, NeuronProperties neuronProperties) {

	// set network type
	this.setNetworkType(NeuralNetworkType.MULTI_LAYER_PERCEPTRON);

               // create input layer
               NeuronProperties inputNeuronProperties = new NeuronProperties(InputNeuron.class, Linear.class);
               Layer layer = LayerFactory.createLayer(neuronsInLayers.get(0), inputNeuronProperties);

               boolean useBias = true; // use bias neurons by default
               if (neuronProperties.hasProperty("useBias")) {
                   useBias = (Boolean)neuronProperties.getProperty("useBias");
               }

               if (useBias) {
                   layer.addNeuron(new BiasNeuron());
               }

               this.addLayer(layer);

	// create layers
	Layer prevLayer = layer;

	//for(Integer neuronsNum : neuronsInLayers)
               for(int layerIdx = 1; layerIdx < neuronsInLayers.size(); layerIdx++){
                       Integer neuronsNum = neuronsInLayers.get(layerIdx);
		// createLayer layer
		layer = LayerFactory.createLayer(neuronsNum, neuronProperties);

                       if ( useBias && (layerIdx< (neuronsInLayers.size()-1)) ) {
                           layer.addNeuron(new BiasNeuron());
                       }

		// add created layer to network
		this.addLayer(layer);
		// createLayer full connectivity between previous and this layer
		if (prevLayer != null)
			ConnectionFactory.fullConnect(prevLayer, layer);

		prevLayer = layer;
	}

	// set input and output cells for network
                 NeuralNetworkFactory.setDefaultIO(this);

                 // set learnng rule
	//this.setLearningRule(new BackPropagation(this));
	this.setLearningRule(new MomentumBackpropagation());
              // this.setLearningRule(new DynamicBackPropagation());
               
               this.randomizeWeights(new NguyenWidrowRandomizer(-0.7, 0.7));
			
}