batchadaptivelearningratemomentumbackpropagation.java

利用Java实现的神经网络工具箱

/*
 * $RCSfile: BatchAdaptiveLearningRateMomentumBackPropagation.java,v $
 * $Revision: 1.1 $
 * $Date: 2005/05/08 02:16:28 $
 *
 * NeuralNetworkToolkit
 * Copyright (C) 2004 Universidade de Bras韑ia
 *
 * This file is part of NeuralNetworkToolkit.
 *
 * NeuralNetworkToolkit is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * NeuralNetworkToolkit is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with NeuralNetworkToolkit; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA - 02111-1307 - USA.
 */
package neuralnetworktoolkit.methods.gradientbased.backpropagation;

import neuralnetworktoolkit.StatisticalResults;
import neuralnetworktoolkit.math.NeuralMath;
import neuralnetworktoolkit.methods.TrainingParameters;
import neuralnetworktoolkit.neuralnetwork.INeuralNetwork;

/**
 * 
 * 
 * @version $Revision: 1.1 $ - $Date: 2005/05/08 02:16:28 $
 * 
 * @author <a href="mailto:[e-mail]">Hugo Iver V. Gon鏰lves</a>
 * @author <a href="mailto:[e-mail]">Rodrigo C. M. Coimbra</a>
 */
public class BatchAdaptiveLearningRateMomentumBackPropagation
		extends
			ModifiedBackPropagation {
	
	public BatchAdaptiveLearningRateMomentumBackPropagation() {
		super();
		instantaneousError = 1000000000;
	}
	
	/**
	 * Trains a neural network with backpropagation method.
	 * 
	 * @param neuralNetwork
	 *            Neural network to be trained.
	 * @param learningRate
	 *            Learning rate.
	 * @param error
	 *            Quadratic error wished for network.
	 * @param instanceSet
	 *            Training instances set.
	 * @param outputs
	 *            Network expected outputs.
	 * @param parameters
	 *            Other parameters array.
	 * 
	 * @return Statistical information about network learning.
	 */
	public StatisticalResults train(INeuralNetwork neuralNetwork,
			TrainingParameters parameters) {
		// TODO Erase instrumentation code.
		BackPropagationParameters param;
		
		param = (BackPropagationParameters) parameters;
		numberOfSynapses = neuralNetwork.numberOfSynapses();
		errorGoal = param.getError();
		maximumNumberOfEpochs = param.getMaxEpochs();
		learningRate = param.getLearningRate();
		alpha = param.getAlpha();
		lrMultiplier = param.getLrMultiplier();
		
		inicio = System.currentTimeMillis();
		
		do {
			// TODO improve this.
			deltaW = new double[neuralNetwork.numberOfSynapses()][1];
			for (int i = 0; i < param.getInputs().length; i++) {
				neuralNetwork.inputLayerSetup(param.getInputs()[i]);
				neuralNetwork.propagateInput();
				calculateNeuronDeltas(neuralNetwork, param.getOutputs()[i]);
				derivativesVector = calculateDerivativesVector(neuralNetwork);
				
				instantDeltaW = NeuralMath.constantTimesMatrix(-learningRate, derivativesVector);
				
				deltaW = NeuralMath.matrixSum(deltaW, instantDeltaW);
								
				
			}
			
			if ( (!firstIteration)&&(successfullIteration) ) {
				deltaW = NeuralMath.matrixSum(deltaW, NeuralMath.constantTimesMatrix(alpha, previousDeltaW));
			} else {
				firstIteration = false;
				//deltaW = neuralMath.constantTimesMatrix(-param.getLearningRate(), derivativesVector);
			}
			
			
			neuralNetwork.updateWeights(deltaW);
			instantaneousError = calculateTotalError(neuralNetwork, param.getInputs(), param.getOutputs());
			if ( instantaneousError < totalErrorEnergy ) {
				//System.out.println("menor");
				previousDeltaW = (double[][] )deltaW.clone();
				learningRate = learningRate*lrMultiplier;
				totalErrorEnergy = instantaneousError;
				successfullIteration = true;
			} else {
				//System.out.println("maior");
				deltaW = NeuralMath.constantTimesMatrix(-1, deltaW);
				neuralNetwork.updateWeights(deltaW);
				learningRate = learningRate/lrMultiplier;
				//successfullIteration = false;
			}
			
			numberOfEpochs++;
			
			if ( (numberOfEpochs) % 1/*(maximumNumberOfEpochs/100)*/ == 0) {
				System.out.println("N鷐ero de 蓀ocas: " + numberOfEpochs);
				System.out.println("Erro atual: " + totalErrorEnergy);
			}
			
		} while (((totalErrorEnergy / param.getInputs().length) > errorGoal)
				&& (numberOfEpochs < maximumNumberOfEpochs) && (goAhead = true));
		fim = System.currentTimeMillis();
		
		neuralNetwork.setError(totalErrorEnergy / param.getInputs().length);
		
		results.setError(totalErrorEnergy / param.getInputs().length);
		results.setNumberOfEpochs(numberOfEpochs);
		numberOfEpochs = 0;
		results.setTrainingTime((fim - inicio) / 1000);
		
		return results;
		
	} //train()
	
} //BatchAdaptiveLearningRateMomentumBackPropagation