crossvalidation.java

利用Java实现的神经网络工具箱

/*
 * Created on 23/10/2004
 *
 * TODO To change the template for this generated file go to
 * Window - Preferences - Java - Code Generation - Code and Comments
 */
package neuralnetworktoolkit.validation;

import neuralnetworktoolkit.*;
import neuralnetworktoolkit.methods.*;
import neuralnetworktoolkit.neuralnetwork.*;
import neuralnetworktoolkit.neuralnetwork.weightinitialization.WeightInitialization;
import neuralnetworktoolkit.normalization.*;

/**
 * @author iver
 *
 * TODO To change the template for this generated type comment go to
 * Window - Preferences - Java - Code Generation - Code and Comments
 */
public class CrossValidation extends Validation {

	/* (non-Javadoc)
	 * @see neuralnetworktoolkit.validation.Validation#validate(neuralnetworktoolkit.neuralnetwork.INeuralNetwork)
	 */
	public ValidationStatistics validate(INeuralNetwork nn, WeightInitialization initialization, INormalization inputNormalization, INormalization outputNormalization, ITrainingMethod trainingMethod, TrainingParameters param) {
		ValidationStatistics result = new ValidationStatistics();
		StatisticalResults trainStat;
		double error = 0;
		double trainError;
		double totalError = 0;
		double[][] inputs = param.getInputs();
		double[][] outputs = param.getOutputs();
		double[][] instantInputs = new double[inputs.length-1][inputs[0].length];
		double[][] instantOutputs = new double[inputs.length-1][inputs[0].length];
		double[][] results = new double[1][outputs[0].length]; 
		double[][] instantOutput = new double[1][outputs[0].length];
		int index = 0;
		
		if ( inputNormalization != null ) {
			System.out.println("Normalizei inputs!!!");
			inputs = inputNormalization.normalize(inputs);
			if ( outputNormalization != null ) {
				System.out.println("Normalizei outputs!!!");
				outputs = outputNormalization.normalize(outputs);
			}
		}
		
		for (int i = 0; i < inputs.length; i++) {
			System.out.println("Round " + (i+1)+ "---------------------------------------------------");
			
			for (int j = 0; j < inputs.length; j++) {
				if (i!=j) {
					//System.out.println("Index: " + index);
					instantInputs[index] = inputs[j];
					instantOutputs[index] = outputs[j];
					index++;
				}
			}
			System.out.println(">>> Treinando");
			initialization.initialize(nn);
			param.setInputs(inputs);
			param.setOutputs(outputs);
			trainStat = trainingMethod.train(nn, param);
			System.out.println(">>> Erro: "+trainStat.getError());
			System.out.println(">>>>>>>>>>>>>");
			nn.inputLayerSetup(inputs[i]);
			nn.propagateInput();
			//System.out.println("entrada 1: "+ inputs[i][0] + " entrada 2: "+inputs[i][1]);
			
			results[0] = nn.retrieveFinalResults();
			
			if ( outputNormalization != null ) {
				results = outputNormalization.unnormalize(results);
				instantOutput[0] = outputs[i];
				instantOutput = outputNormalization.unnormalize(instantOutput);
			}
			
			for (int k = 0; k < nn.retrieveFinalResults().length; k++) {
				System.out.println("Saída rede: " + results[0][k] + " Original: " + instantOutput[0][k] + " Erro " + Math.abs( instantOutput[0][k] - results[0][k] ) );
				error = error + ( Math.abs( instantOutput[0][k] - results[0][k] ) );
			}
			
			totalError = totalError +error;
			error = 0;
			index = 0;
						
			System.out.println("----------------------------------------------------------");
		}
		
		result.setTotalError(totalError/inputs.length);
		
		return result;
	}
}