Examples of org.encog.neural.networks.training.propagation.resilient.ResilientPropagation

Package org.encog.neural.networks.training.propagation.resilient

Examples of org.encog.neural.networks.training.propagation.resilient.ResilientPropagation

org.encog.neural.networks.training.propagation.resilient.ResilientPropagation
One problem with the backpropagation algorithm is that the magnitude of the partial derivative is usually too large or too small. Further, the learning rate is a single value for the entire neural network. The resilient propagation learning algorithm uses a special update value(similar to the learning rate) for every neuron connection. Further these update values are automatically determined, unlike the learning rate of the backpropagation algorithm. For most training situations, we suggest that the resilient propagation algorithm (this class) be used for training. There are a total of three parameters that must be provided to the resilient training algorithm. Defaults are provided for each, and in nearly all cases, these defaults are acceptable. This makes the resilient propagation algorithm one of the easiest and most efficient training algorithms available. The optional parameters are: zeroTolerance - How close to zero can a number be to be considered zero. The default is 0.00000000000000001. initialUpdate - What are the initial update values for each matrix value. The default is 0.1. maxStep - What is the largest amount that the update values can step. The default is 50. Usually you will not need to use these, and you should use the constructor that does not require them. @author jheaton

    if (method instanceof SVM) {
      train = new SVMTrain((SVM)method, dataSet);
    } if(method instanceof FreeformNetwork ) {
      train = new FreeformResilientPropagation((FreeformNetwork) method, dataSet);
    } else {
      train = new ResilientPropagation((ContainsFlat)method, dataSet);
    }
    EncogUtility.trainToError(train, error);
  }

View Full Code Here

    // train the neural network


    double error = Double.POSITIVE_INFINITY;
    for (int z = 0; z < this.weightTries; z++) {
      network.reset();
      final Propagation train = new ResilientPropagation(network,
          useTraining);
      final StopTrainingStrategy strat = new StopTrainingStrategy(0.001,
          5);


      train.addStrategy(strat);
      train.setThreadCount(1); // force single thread mode


      for (int i = 0; (i < this.iterations) && !getShouldStop()
          && !strat.shouldStop(); i++) {
        train.iteration();
      }


      error = Math.min(error, train.getError());
    }


    if (buffer != null) {
      buffer.close();
    }

View Full Code Here

   * @param trainingSet
   *            The training set to use.
   */
  public static void trainDialog(final BasicNetwork network,
      final MLDataSet trainingSet) {
    final Propagation train = new ResilientPropagation(network, trainingSet);
    train.setThreadCount(0);
    TrainingDialog.trainDialog(train, network, trainingSet);
  }

View Full Code Here

  public void testRPROPCont() {
    MLDataSet trainingSet = XOR.createXORDataSet();
    BasicNetwork net1 = XOR.createUnTrainedXOR();
    BasicNetwork net2 = XOR.createUnTrainedXOR();
    
    ResilientPropagation rprop1 = new ResilientPropagation(net1,trainingSet);
    ResilientPropagation rprop2 = new ResilientPropagation(net2,trainingSet);
    
    rprop1.iteration();
    rprop1.iteration();
    
    rprop2.iteration();
    rprop2.iteration();
    
    TrainingContinuation cont = rprop2.pause();
    
    ResilientPropagation rprop3 = new ResilientPropagation(net2,trainingSet);
    rprop3.resume(cont);
    
    rprop1.iteration();
    rprop3.iteration();
    
    
    for(int i=0;i<net1.getFlat().getWeights().length;i++) {
      Assert.assertEquals(net1.getFlat().getWeights()[i], net2.getFlat().getWeights()[i],0.0001);
    }

View Full Code Here

  public void testRPROPContPersistEG() {
    MLDataSet trainingSet = XOR.createXORDataSet();
    BasicNetwork net1 = XOR.createUnTrainedXOR();
    BasicNetwork net2 = XOR.createUnTrainedXOR();
    
    ResilientPropagation rprop1 = new ResilientPropagation(net1,trainingSet);
    ResilientPropagation rprop2 = new ResilientPropagation(net2,trainingSet);
    
    rprop1.iteration();
    rprop1.iteration();
    
    rprop2.iteration();
    rprop2.iteration();
    
    TrainingContinuation cont = rprop2.pause();
    
    EncogDirectoryPersistence.saveObject(EG_FILENAME, cont);
    TrainingContinuation cont2 = (TrainingContinuation)EncogDirectoryPersistence.loadObject(EG_FILENAME);
    
    ResilientPropagation rprop3 = new ResilientPropagation(net2,trainingSet);
    rprop3.resume(cont2);
    
    rprop1.iteration();
    rprop3.iteration();
    
    
    for(int i=0;i<net1.getFlat().getWeights().length;i++) {
      Assert.assertEquals(net1.getFlat().getWeights()[i], net2.getFlat().getWeights()[i],0.0001);
    }

View Full Code Here

  public void testRPROP() throws Throwable
  {
    MLDataSet trainingData = new BasicMLDataSet(XOR.XOR_INPUT,XOR.XOR_IDEAL);
    
    BasicNetwork network = NetworkUtil.createXORNetworkUntrained();
    MLTrain rprop = new ResilientPropagation(network, trainingData);
    NetworkUtil.testTraining(trainingData,rprop,0.03);
  }

View Full Code Here

    MLDataSet trainingData = XOR.createNoisyXORDataSet(10);
    
    BasicNetwork network = NetworkUtil.createXORNetworkUntrained();
    
    final FoldedDataSet folded = new FoldedDataSet(trainingData); 
    final MLTrain train = new ResilientPropagation(network, folded);
    final CrossValidationKFold trainFolded = new CrossValidationKFold(train,4);
    
    EncogUtility.trainToError(trainFolded, 0.2);
    
    XOR.verifyXOR((MLRegression)trainFolded.getMethod(), 0.2);

View Full Code Here

  public void testLimited()
  {
    MLDataSet trainingData = new BasicMLDataSet(XOR.XOR_INPUT,XOR.XOR_IDEAL);
    BasicNetwork network = NetworkUtil.createXORNetworkUntrained();    
    
    ResilientPropagation rprop = new ResilientPropagation(network,trainingData);
    rprop.iteration();
    rprop.iteration();
    network.enableConnection(1, 0, 0, false);
    network.enableConnection(1, 1, 0, false);
    
    Assert.assertTrue(network.getStructure().isConnectionLimited());
    
    Assert.assertEquals(0.0, network.getStructure().getFlat().getWeights()[0], 0.01);
    Assert.assertEquals(0.0, network.getStructure().getFlat().getWeights()[1], 0.01);
    rprop.iteration();
    rprop.iteration();
    rprop.iteration();
    rprop.iteration();
    // these connections were removed, and should not have been "trained"
    Assert.assertEquals(0.0, network.getStructure().getFlat().getWeights()[0], 0.01);
    Assert.assertEquals(0.0, network.getStructure().getFlat().getWeights()[1], 0.01);    
    rprop.finishTraining();
  }

View Full Code Here

  
  public void testRPROPConsistency() {
    MLDataSet training = EncoderTrainingFactory.generateTraining(4, false);
    BasicNetwork network = EncogUtility.simpleFeedForward(4, 2, 0, 4, true);
    (new ConsistentRandomizer(-1,1,50)).randomize(network);
    ResilientPropagation rprop = new ResilientPropagation(network,training);
    for(int i=0;i<5;i++) {
      rprop.iteration();
    }
    Assert.assertArrayEquals(EXPECTED_WEIGHTS1, network.getFlat().getWeights(),0.0001);
    for(int i=0;i<5;i++) {
      rprop.iteration();
    }
    Assert.assertArrayEquals(EXPECTED_WEIGHTS2, network.getFlat().getWeights(),0.0001);  
    
    double e = network.calculateError(training);
    Assert.assertEquals(0.0767386807494191, e, 0.00001);

View Full Code Here

    network2.setBiasActivation(-1);
    network3.setBiasActivation(0.5);
    
    MLDataSet trainingData = new BasicMLDataSet(XOR.XOR_INPUT,XOR.XOR_IDEAL);
    
    MLTrain rprop1 = new ResilientPropagation(network1, trainingData);
    MLTrain rprop2 = new ResilientPropagation(network2, trainingData);
    MLTrain rprop3 = new ResilientPropagation(network3, trainingData);


    NetworkUtil.testTraining(trainingData,rprop1,0.03);
    NetworkUtil.testTraining(trainingData,rprop2,0.01);
    NetworkUtil.testTraining(trainingData,rprop3,0.01);

View Full Code Here

0 1 2 3 4 5

TOP

Related Classes of org.encog.neural.networks.training.propagation.resilient.ResilientPropagation

com.wiieditor.neuralnetwork.NeuralNetworkWrapper

mechanics.NeuralNetwork

org.encog.ensemble.training.ResilientPropagationFactory

org.encog.examples.neural.activation.CustomActivation

org.encog.examples.neural.benchmark.MultiBench

org.encog.examples.neural.benchmark.TestConverge

org.encog.examples.neural.benchmark.ThreadCount

org.encog.examples.neural.benchmark.WeightInitialization

org.encog.examples.neural.cross.CrossValidateSunspot

org.encog.examples.neural.gui.predict.PredictSIN

All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.