JOONE results are mot accurate?? Help with unnormalize!!

LiuKejun — Fri, 05 Jun 2015 09:42:43 -0000

My english is not very good. If i use some wrong expressions, forgive me.

JOONE results are mot accurate?? Help with unnormalize!!

LiuKejun — Fri, 05 Jun 2015 09:25:53 -0000

I have been studying JOONE for a month, and now, i am trying to use JOONE to solve my FFNN project. I trained the neural net, validate it and test it. But the results are not as accurate as predicted... I think there my be some mistakes about normalize and unnormalize. But i don't what is it exactly.
Now, I am confused and don't know what to do.
The open discussion seems deserted. I hope someone can help me with this problem ~ I will be very grateful for that !!

I have 7 inputs and 1 output. I do normalize for each column separately.

My whole code is here:

import java.io.File;
import java.io.FileInputStream;
import java.io.FileOutputStream;
import java.io.ObjectInputStream;
import java.io.ObjectOutputStream;

import org.joone.engine.;
import org.joone.engine.learning.;
import org.joone.net.;
import org.joone.helpers.factory.JooneTools;
import org.joone.io.;
import org.joone.util.*;

public class Peat implements NeuralNetListener, NeuralValidationListener {

NeuralNet net;
long startms;
private static String filePath = "F:/ANN/project01/XOR";
private static String netFile="F:/ANN/project01/XOR/sample.snet";
private static String outputFile="F:/ANN/project01/XOR/output.txt";
private static boolean singleThreadMode = true;

public Peat() {
}

public static void main(String[] args) {

    Peat sampleNet = new Peat();
    sampleNet.initialize(filePath);
    sampleNet.start();
    sampleNet.saveNeuralNet(netFile);
    NeuralNet nnet=sampleNet.restoreNeutralNet(netFile);
    sampleNet.test(nnet);

}

private void initialize(String path) {
    /* Creates the three layers and connect them */
    LinearLayer ILayer = new LinearLayer();  // Input Layer
    SigmoidLayer HLayer = new SigmoidLayer(); // Hidden Layer
    SigmoidLayer OLayer = new SigmoidLayer(); // Output Layer
    ILayer.setRows(7); // The input pattern has 13 columns
    HLayer.setRows(3);
    OLayer.setRows(1);  // The desired pattern has 1 column
    FullSynapse synIH = new FullSynapse();
    FullSynapse synHO = new FullSynapse();
    this.connect(ILayer, synIH, HLayer);
    this.connect(HLayer, synHO, OLayer);

    /* Creates all the required input data sets */
    FileInputSynapse ITdata = this.createInputForIn(path+"/file1.txt",1,2,8); /* The input training data set */
    FileInputSynapse IVdata = this.createInputForIn(path+"/file1.txt",816,2,8); /* The input validation data set */
    FileInputSynapse DTdata = this.createInputForOut(path+"/file1.txt",1,1,1); /* The desired training data set */
    FileInputSynapse DVdata = this.createInputForOut(path+"/file1.txt",816,1,1); /* The desired validation data set */

    /* Creates and attach the input learning switch */
    LearningSwitch Ilsw = this.createSwitch(ITdata, IVdata);
    ILayer.addInputSynapse(Ilsw);

    /* Creates and attach the desired learning switch */
    LearningSwitch Dlsw = this.createSwitch(DTdata, DVdata);
    TeachingSynapse ts = new TeachingSynapse(); // The teacher of the net
    ts.setDesired(Dlsw);
    OLayer.addOutputSynapse(ts);

    /* Now we put all togheter into a NeuralNet object */
    net = new NeuralNet();
    net.addLayer(ILayer, NeuralNet.INPUT_LAYER);
    net.addLayer(HLayer, NeuralNet.HIDDEN_LAYER);
    net.addLayer(OLayer, NeuralNet.OUTPUT_LAYER);
    net.setTeacher(ts);

    /* Sets the Monitor's parameters */
    Monitor mon = net.getMonitor();
    mon.setLearningRate(0.4);
    mon.setMomentum(0.5);
    mon.setSingleThreadMode(singleThreadMode);
    mon.setTrainingPatterns(815);
    mon.setValidationPatterns(100);

    mon.setTotCicles(100);
    mon.setLearning(true);
}
/**store the neutral net*/
private void saveNeuralNet(String filename){
    try{
        if (!new File(filename).exists()){
            File file=new File(filename);
        }
        FileOutputStream stream =new FileOutputStream(filename);
        ObjectOutputStream out=new ObjectOutputStream(stream);
        out.writeObject(net);
        out.close();
    }
    catch(Exception excp){
        excp.printStackTrace();
    }
}

/**read the neural net from external file*/
private NeuralNet restoreNeutralNet(String filename){
    NeuralNet nnet=null;
    try{
        FileInputStream stream=new FileInputStream(filename);
        ObjectInputStream inp=new ObjectInputStream(stream);
        nnet=(NeuralNet)inp.readObject();
    }catch(Exception excp){
        excp.printStackTrace();
    }
    return nnet;
}

private void test(NeuralNet nnet){

    double inputArray[][]={ {9.6,1800,16,75,142.108,1.98,42,52}} ;//The test data

     if(nnet!=null){
        Layer input=nnet.getInputLayer();
        input.removeAllInputs();

        NormalizerPlugIn norm1 = new NormalizerPlugIn();
        norm1.setAdvancedSerieSelector("1");
        norm1.setMin(0);
        norm1.setMax(1);
        norm1.setDataMax(2200);
        norm1.setDataMin(1600);

        NormalizerPlugIn norm2 = new NormalizerPlugIn();
        norm2.setAdvancedSerieSelector("2");
        norm2.setMin(0);
        norm2.setMax(1);
        norm2.setDataMax(64);
        norm2.setDataMin(5);

        NormalizerPlugIn norm3 = new NormalizerPlugIn();
        norm3.setAdvancedSerieSelector("3");
        norm3.setMin(0);
        norm3.setMax(1);
        norm3.setDataMax(151);
        norm3.setDataMin(60);

        NormalizerPlugIn norm4 = new NormalizerPlugIn();
        norm4.setAdvancedSerieSelector("4");
        norm4.setMin(0);
        norm4.setMax(1);
        norm4.setDataMax(403.332);
        norm4.setDataMin(70.433);

        NormalizerPlugIn norm5 = new NormalizerPlugIn();
        norm5.setAdvancedSerieSelector("5");
        norm5.setMin(0);
        norm5.setMax(1);
        norm5.setDataMax(8.24);
        norm5.setDataMin(1.27);

        NormalizerPlugIn norm6 = new NormalizerPlugIn();
        norm6.setAdvancedSerieSelector("6");
        norm6.setMin(0);
        norm6.setMax(1);
        norm6.setDataMax(85);
        norm6.setDataMin(30);

        NormalizerPlugIn norm7 = new NormalizerPlugIn();
        norm7.setAdvancedSerieSelector("7");
        norm7.setMin(0);
        norm7.setMax(1);
        norm7.setDataMax(62);
        norm7.setDataMin(46);

        MemoryInputSynapse memInp=new MemoryInputSynapse();
        memInp.setInputArray(inputArray);
        memInp.setFirstRow(1);
        memInp.setAdvancedColumnSelector("2-8");
        memInp.addPlugIn(norm7);
        memInp.addPlugIn(norm6);
        memInp.addPlugIn(norm5);
        memInp.addPlugIn(norm4);
        memInp.addPlugIn(norm3);
        memInp.addPlugIn(norm2);
        memInp.addPlugIn(norm1);
        input.addInputSynapse(memInp);

        Layer output=nnet.getOutputLayer();
        output.removeAllOutputs();

        FileOutputSynapse fileoutput=new FileOutputSynapse();
        fileoutput.setFileName(outputFile);
        UnNormalizerOutputPlugIn unnormalize =new   UnNormalizerOutputPlugIn();
        unnormalize.setAdvancedSerieSelector("1");
        unnormalize.setInDataMin(0);
        unnormalize.setInDataMax(1);
        unnormalize.setOutDataMax(11.3);
        unnormalize.setOutDataMin(6.7);
        fileoutput.addPlugIn(unnormalize);
        output.addOutputSynapse(fileoutput);

        nnet.getMonitor().setTotCicles(1);
        nnet.getMonitor().setLearning(false);
        nnet.getMonitor().setTrainingPatterns(1);
        nnet.start();
        nnet.getMonitor().Go();
        nnet.join();
        //System.out.print("ERROR"+nnet.getMonitor().getGlobalError());
    }
}
/** Creates a FileInputSynapse */
private FileInputSynapse createInputForIn(String name, int firstRow, int firstCol, int lastCol) {
    FileInputSynapse input = new FileInputSynapse();
    input.setInputFile(new File(name));
    input.setFirstRow(firstRow);
    if (firstCol != lastCol)
        input.setAdvancedColumnSelector(firstCol+"-"+lastCol);
    else
        input.setAdvancedColumnSelector(Integer.toString(firstCol));

    // We normalize the input data in the range 0 - 1
    NormalizerPlugIn norm1 = new NormalizerPlugIn();
    NormalizerPlugIn norm2 = new NormalizerPlugIn();
    NormalizerPlugIn norm3 = new NormalizerPlugIn();
    NormalizerPlugIn norm4 = new NormalizerPlugIn();
    NormalizerPlugIn norm5 = new NormalizerPlugIn();
    NormalizerPlugIn norm6 = new NormalizerPlugIn();
    NormalizerPlugIn norm7 = new NormalizerPlugIn();

    norm1.setAdvancedSerieSelector("1");
    norm2.setAdvancedSerieSelector("2");
    norm3.setAdvancedSerieSelector("3");
    norm4.setAdvancedSerieSelector("4");
    norm5.setAdvancedSerieSelector("5");
    norm6.setAdvancedSerieSelector("6");
    norm7.setAdvancedSerieSelector("7");

    /**column 1*/
    norm1.setMin(0);
    norm1.setMax(1);
    norm1.setDataMax(2200);
    norm1.setDataMin(1600);

    /**column 2*/
    norm2.setMin(0);
    norm2.setMax(1);
    norm2.setDataMax(64);
    norm2.setDataMin(5);

    /**column 3*/
    norm3.setMin(0);
    norm3.setMax(1);
    norm3.setDataMax(151);
    norm3.setDataMin(60);

    /**column 4*/
    norm4.setMin(0);
    norm4.setMax(1);
    norm4.setDataMax(403.332);
    norm4.setDataMin(70.433);

    /**column 5*/
    norm5.setMin(0);
    norm5.setMax(1);
    norm5.setDataMax(8.24);
    norm5.setDataMin(1.27);

    /**column 6*/
    norm6.setMin(0);
    norm6.setMax(1);
    norm6.setDataMax(85);
    norm6.setDataMin(30);

    /**column 7*/
    norm7.setMin(0);
    norm7.setMax(1);
    norm7.setDataMax(62);
    norm7.setDataMin(46);

    input.addPlugIn(norm7);

    input.addPlugIn(norm6);

    input.addPlugIn(norm5);

    input.addPlugIn(norm4);

    input.addPlugIn(norm3);

    input.addPlugIn(norm2);

    input.addPlugIn(norm1);

    return input;
}

private FileInputSynapse createInputForOut(String name, int firstRow, int firstCol, int lastCol) {
    FileInputSynapse input = new FileInputSynapse();
    input.setInputFile(new File(name));
    input.setFirstRow(firstRow);
    input.setAdvancedColumnSelector(Integer.toString(firstCol));

    // We normalize the input data in the range 0 - 1
    NormalizerPlugIn norm = new NormalizerPlugIn();

    norm.setAdvancedSerieSelector("1");

    norm.setMin(0);
    norm.setMax(1);
    norm.setDataMax(11.3);
    norm.setDataMin(6.7);

    input.addPlugIn(norm);

    return input;
}

/** Connects two Layers with a Synapse */
private void connect(Layer ly1, Synapse syn, Layer ly2) {
    ly1.addOutputSynapse(syn);
    ly2.addInputSynapse(syn);
}

/* Creates a LearningSwitch and attach to it both the training and
   the desired input synapses */
private LearningSwitch createSwitch(StreamInputSynapse IT, StreamInputSynapse IV) {
    LearningSwitch lsw = new LearningSwitch();
    lsw.addTrainingSet(IT);
    lsw.addValidationSet(IV);
    return lsw;
}

private void start() {
    // Registers itself as a listener
    net.getMonitor().addNeuralNetListener(this);
    startms = System.currentTimeMillis();
    net.go();
    //Waits for all the threads to finish
    net.join();

}

/* Events */
public void netValidated(NeuralValidationEvent event) {
    // Shows the RMSE at the end of the cycle
    NeuralNet NN = (NeuralNet)event.getSource();
    System.out.println("    Validation Error: "+NN.getMonitor().getGlobalError());
}

public void cicleTerminated(NeuralNetEvent e) {
    // Prints out the cycle and the training error
    int cycle = net.getMonitor().getTotCicles() - net.getMonitor().getCurrentCicle()+1;
    if (cycle % 10 == 0) { // We validate the net every 10 cycles
        System.out.println("Cycle #"+cycle);
        System.out.println("    Training Error:   " + net.getMonitor().getGlobalError());

        // Creates a copy of the neural network
        net.getMonitor().setExporting(true);
        NeuralNet newNet = net.cloneNet();
        net.getMonitor().setExporting(false);

        // Cleans the old listeners
        // This is a fundamental action to avoid that the validating net
        // calls the cicleTerminated method of this class
        newNet.removeAllListeners();

        // Set all the parameters for the validation
        NeuralNetValidator nnv = new NeuralNetValidator(newNet);
        nnv.addValidationListener(this);
        nnv.start();  // Validates the net
    }
}

public void errorChanged(NeuralNetEvent e) {
}

public void netStarted(NeuralNetEvent e) {
}

public void netStopped(NeuralNetEvent e) {
    Monitor mon=(Monitor)e.getSource();
    if(mon.isLearning())
    {
        int epoch=mon.getTotCicles()-mon.getCurrentCicle();
        System.out.print("Epoch:"+epoch+"last RMSE="+mon.getGlobalError());
    }
    System.out.println("Stopped after "+(System.currentTimeMillis()-startms)+" ms");
}

public void netStoppedError(NeuralNetEvent e,String error) {
}

}

Help with Normalization and unnormalization ranges

LiuKejun — Wed, 03 Jun 2015 09:52:51 -0000

Hello!
I am training a FFNN by 815 sets of data and validate using 100 sets of data. In the neural net,there are 7 inputs and 1 output.The inputs are the independent variables and the output is dependent variable. So I need to train my network to predict the output, based on input data.

I have the following problem: I have 7 inputs, which have different ranges. My output data is in range started from 6.7 to 11.3. I've used linear layer for the input,sigmoid layer for the hidden layer and output layer, so I need to normalize both my input and output data.
Both in training process and validation process,I normalize the 7 kinds of inputs to (0.1,0.9).

But when I use test data to get result , I am confused about how should I unnormalize the data?
How should I get the results?
I get pretty good RMSE, around 0.19,but the results are far from expected.

Here are my JAVA codes:
In the initiate():
......
LinearLayer ILayer = new LinearLayer();
SigmoidLayer HLayer = new SigmoidLayer();
SigmoidLayer OLayer = new SigmoidLayer();
ILayer.setRows(7);
HLayer.setRows(3);
OLayer.setRows(1);
......
FileInputSynapse ITdata = this.createInput(path+"/file1.txt",1,2,8); / The input training data set /
FileInputSynapse IVdata = this.createInput(path+"/file1.txt",816,2,8); / The input validation data set /
FileInputSynapse DTdata = this.createInput(path+"/file1.txt",1,1,1); / The desired training data set /
FileInputSynapse DVdata = this.createInput(path+"/file1.txt",816,1,1);
LearningSwitch Ilsw = this.createSwitch(ITdata, IVdata);
ILayer.addInputSynapse(llsw);
LearningSwitch Dlsw = this.createSwitch(DTdata, DVdata);
TeachingSynapse ts = new TeachingSynapse(); // The teacher of the net
ts.setDesired(Dlsw);
OLayer.addOutputSynapse(ts);
......

In the test():
Layer input=nnet.getInputLayer();
input.removeAllInputs();
NormalizerPlugIn norm = new NormalizerPlugIn();
norm.setAdvancedSerieSelector("2-8");
norm.setMin(0);
norm.setMax(1);

MemoryInputSynapse memInp=new MemoryInputSynapse();
//memInp.addPlugIn(norm);
memInp.setFirstRow(1);
memInp.setAdvancedColumnSelector("2-8");
input.addInputSynapse(memInp);
memInp.setInputArray(inputArray);

Layer output=nnet.getOutputLayer();
output.removeAllOutputs();

FileOutputSynapse fileoutput=new FileOutputSynapse();
fileoutput.setFileName(outputFile);
UnNormalizerOutputPlugIn unnormalize =new UnNormalizerOutputPlugIn();
unnormalize.setAdvancedSerieSelector("1");
unnormalize.setInDataMin(-1);
unnormalize.setInDataMax(1);
unnormalize.setOutDataMax(11.3);
unnormalize.setOutDataMin(6.7);
fileoutput.addPlugIn(unnormalize);
output.addOutputSynapse(fileoutput);

Please, help me with this question, I would be very thankful!
Thanks!
With best regards,
LiuKejun.

Recent posts to Discussion

JOONE results are mot accurate?? Help with unnormalize!!

JOONE results are mot accurate?? Help with unnormalize!!

Help with Normalization and unnormalization ranges