gradientbasedmethod.java

利用Java实现的神经网络工具箱

/*
 * Created on 09/10/2004
 *
 * TODO To change the template for this generated file go to
 * Window - Preferences - Java - Code Generation - Code and Comments
 */
package neuralnetworktoolkit.methods.gradientbased;

import neuralnetworktoolkit.methods.TrainingMethod;
import neuralnetworktoolkit.neuralnetwork.INeuralNetwork;

/**
 * @author iver
 *
 * TODO To change the template for this generated type comment go to
 * Window - Preferences - Java - Code Generation - Code and Comments
 */
public abstract class GradientBasedMethod extends TrainingMethod {
	
	/* (non-Javadoc)
	 * @see neuralnetworktoolkit.methods.ITrainingMethod#train(neuralnetworktoolkit.INeuralNetwork, double, double[][], double[][], int[])
	 */
	
	/**
	 * Calculates the neural network derivatives vector.
	 * 
	 * @param neuralNetwork Neural network to calculate derivatives
	 * 						vector.
	 * 
	 * @return Derivatives vector.
	 */
	public double[] calculateDerivativesVector(INeuralNetwork neuralNetwork) {

		int numberOfSynapses = neuralNetwork.numberOfSynapses();
		double[] derivativesVector = new double[numberOfSynapses];
		int index				   = 0;
		int index2 				   = 0;
						
		// calcula os termos do vetor de derivadas para a camada de saida
		for (int i = (neuralNetwork.getNetworkSize() - 1);
			i >= 0;
			i--) {
			
			for (int j = 0;
				j < neuralNetwork.getLayer(i).getLayerSize();
				j++) {
				
				derivativesVector[index2] =
					neuralNetwork.getLayer(i).getNeuron(j).getDelta();
				index2++;
				for (int k = 0;
					k < neuralNetwork.getLayer(i).getWeightSize(j);
					k++) {
					
					
						if (i != 0) {
							derivativesVector[index2] =
								neuralNetwork
									.getLayer(i)
									.getNeuron(j)
									.getDelta()
									* neuralNetwork
										.getLayer(i - 1)
										.getNeuron(k)
										.getOutputValue();
							index2++;
						} else {
							derivativesVector[index2] =
								neuralNetwork
									.getLayer(i)
									.getNeuron(j)
									.getDelta()
									* neuralNetwork.getStaticInputValues()[k];
							index2++;
						}
					

				}

			}

		}

		return derivativesVector;

	} //calculateDerivativesVector()

	
}