使用encog 3.0库构建elman递归神经网络时遇到问题。几次迭代后我得到了NaN错误。
这是我在MATLAB中的代码：
net=newlrn (minmax(usdjpytrain),[4,1], {'logsig', 'purelin'}, 'trainscg', 'learngdm', 'mse');
net.trainParam.epochs = 1000;
net.trainParam.goal = 1e-5;
我想在java中创建如上所述的循环神经网络。默认情况下，MATLAB中的学习速率值为0.01，动量为0.9。但是如果我使用这样的学习率和动量，我就会出现 NaN 错误。这是我在java中的代码：
/*
* To change this template, choose Tools | Templates
* and open the template in the editor.
*/
package modifikasiJSR;
import org.encog.Encog;
import org.encog.engine.network.activation.ActivationSigmoid;
import org.encog.ml.data.MLDataSet;
import org.encog.ml.data.basic.BasicMLDataSet;
import org.encog.ml.train.MLTrain;
import org.encog.neural.networks.BasicNetwork;
import org.encog.neural.networks.training.CalculateScore;
import org.encog.neural.networks.training.TrainingSetScore;
import org.encog.neural.networks.training.propagation.back.Backpropagation;
import org.encog.neural.pattern.ElmanPattern;
import dataSet.DataDefault;
import database.Parameter;
import org.encog.ml.data.MLData;
import org.encog.ml.data.MLDataPair;
import org.encog.ml.train.strategy.HybridStrategy;
import org.encog.neural.networks.structure.AnalyzeNetwork;
import org.encog.neural.networks.training.propagation.scg.ScaledConjugateGradient;
/**
*
* @author Akmal
*/
public class MseTrain{
private String mataUang;
private int jumlahDataSet;
Parameter param = new Parameter();
private double [] bias_hid_out = new double[2];
private double [] bias_hid_in = new double[param.getHidden_node()];
private double [] bobot_hid_out = new double[param.getHidden_node()];
private double [] bobot_hid_in = new double[7*param.getHidden_node()];
public MseTrain(String mataUang){
this.mataUang = mataUang;
}
static BasicNetwork createElmanNetwork() {
// construct an Elman type network
Parameter param = new Parameter();
ElmanPattern pattern = new ElmanPattern();
pattern.setActivationFunction(new ActivationSigmoid());
pattern.setInputNeurons(7);
pattern.addHiddenLayer(param.getHidden_node());
pattern.setOutputNeurons(1);
return (BasicNetwork)pattern.generate();
}
public void mse(double[][] data, double [][] target) {
System.out.println("menjalankan method MSE");
DataDefault dataSet = new DataDefault(mataUang);
jumlahDataSet = dataSet.getJumTraining();
final MLDataSet trainingSet = new BasicMLDataSet(data, target);
final BasicNetwork elmanNetwork = MseTrain.createElmanNetwork();
final double elmanError = trainNetwork("Elman", elmanNetwork,trainingSet);
System.out.println("Best error rate with Elman Network: " + elmanError);
Encog.getInstance().shutdown();
}
private double trainNetwork(final String what,final BasicNetwork network, final MLDataSet trainingSet) {
network.reset();
// train the neural network
System.out.println(network.getActivation(0));
System.out.println(network.getActivation(1));
System.out.println(network.getActivation(2));
CalculateScore score = new TrainingSetScore(trainingSet);
final MLTrain trainMain = new Backpropagation(network, trainingSet, 0.01, 0.9);
final MLTrain trainAlt = new ScaledConjugateGradient(network, trainingSet);
trainMain.addStrategy(new HybridStrategy(trainAlt));
//set Bobot hid In
network.setWeight(0, 0, 0, 3.0439);
network.setWeight(0, 0, 1, 2.8421);
network.setWeight(0, 0, 2, -2.8529);
network.setWeight(0, 0, 3, 2.7013);
network.setWeight(0, 1, 0, 1.2741);
network.setWeight(0, 1, 1, -2.8055);
network.setWeight(0, 1, 2, -1.6862);
network.setWeight(0, 1, 3, 0.3049);
network.setWeight(0, 2, 0, 4.5054);
network.setWeight(0, 2, 1, 3.3152);
network.setWeight(0, 2, 2, -2.6663);
network.setWeight(0, 2, 3, 3.1311);
network.setWeight(0, 3, 0, 4.4113);
network.setWeight(0, 3, 1, -0.1022);
network.setWeight(0, 3, 2, 2.2913);
network.setWeight(0, 3, 3, -2.3347);
network.setWeight(0, 4, 0, -1.2521);
network.setWeight(0, 4, 1, 4.8178);
network.setWeight(0, 4, 2, 3.6979);
network.setWeight(0, 4, 3, 4.9682);
network.setWeight(0, 5, 0, 1.4512);
network.setWeight(0, 5, 1, -3.1329);
network.setWeight(0, 5, 2, 2.5726);
network.setWeight(0, 5, 3, 2.7323);
network.setWeight(0, 6, 0, 2.3088);
network.setWeight(0, 6, 1, 2.4109);
network.setWeight(0, 6, 2, 2.4835);
network.setWeight(0, 6, 3, -0.9406);
//set Bias hid In
network.setWeight(0, 7, 0, -9.3723);
network.setWeight(0, 7, 1, -2.1356);
network.setWeight(0, 7, 2, -5.0966);
network.setWeight(0, 7, 3, -1.5829);
//set Bobot Hid out
network.setWeight(1, 0, 0, -0.6263);
network.setWeight(1, 1, 0, -0.0205);
network.setWeight(1, 2, 0, -0.1088);
network.setWeight(1, 3, 0, 0.2926);
//set Bias Hid out
network.setWeight(1, 4, 0, 0.4187);
int epoch = 0;
do{
trainMain.iteration();
System.out.println("Training " + what + ", Epoch #" + epoch + " Error:" + trainMain.getError());
System.out.println("Error:" + trainMain.getError());
epoch++;
}while(trainMain.getError()>0.00001&&epoch<=1);
System.out.println("Neural Network Results:");
for(MLDataPair pair: trainingSet ) {
final MLData output = network.compute(pair.getInput());
System.out.println(
pair.getInput().getData(0)
+ ","
+ pair.getInput().getData(1)
+ ","
+ pair.getInput().getData(2)
+ ","
+ pair.getInput().getData(3)
+ ","
+ pair.getInput().getData(4)
+ ","
+ pair.getInput().getData(5)
+ ","
+ pair.getInput().getData(6)
+ ", actual="
+ output.getData(0)
+ ",ideal="
+ pair.getIdeal().getData(0));
}
System.out.println("bobot : "+network.dumpWeights());
AnalyzeNetwork bobotBias = new AnalyzeNetwork(network);
int jumBobotIn = param.getHidden_node()*7;
System.out.println("jumBobotIn : "+jumBobotIn);
int indexBobotIn = 0;
int indexBobotOut = 0;
for(int i=0; i<bobotBias.getWeightValues().length; i++){
System.out.println("bobot : "+bobotBias.getWeightValues()[i]);
if(i<jumBobotIn){
bobot_hid_in[indexBobotIn] = bobotBias.getWeightValues()[i];
indexBobotIn++;
}
else{
bobot_hid_out[indexBobotOut] = bobotBias.getWeightValues()[i];
indexBobotOut++;
}
}
int indexBiasIn = 0;
int indexBiasOut = 0;
int jumBiasIn = param.getHidden_node();
for(int i=0; i<bobotBias.getBiasValues().length; i++){
System.out.println("bias : "+bobotBias.getBiasValues()[i]);
if(i<jumBiasIn){
bias_hid_in[indexBiasIn] = bobotBias.getBiasValues()[i];
indexBiasIn++;
}
else{
bias_hid_out[indexBiasOut] = bobotBias.getBiasValues()[i];
indexBiasOut++;
}
}
return trainMain.getError();
}
/**
* @return the bias_hid_out
*/
public double[] getBias_hid_out() {
return bias_hid_out;
}
/**
* @return the bias_hid_in
*/
public double[] getBias_hid_in() {
return bias_hid_in;
}
/**
* @return the bobot_hid_out
*/
public double[] getBobot_hid_out() {
return bobot_hid_out;
}
/**
* @return the bobot_hid_in
*/
public double[] getBobot_hid_in() {
return bobot_hid_in;
}
}