Examples of Svm

com.nr.ci.Svm
Support Vector Machines Copyright (C) Numerical Recipes Software 1986-2007 Java translation Copyright (C) Huang Wen Hui 2012 @author hwh
org.encog.ml.svm.SVM
This is a network that is backed by one or more Support Vector Machines (SVM). It is designed to function very similarly to an Encog neural network, and is largely interchangeable with an Encog neural network. The support vector machine supports several types. Regression is used when you want the network to predict a value, given the input. Function approximation is a good example of regression. Classification is used when you want the SVM to group the input data into one or more classes. Support Vector Machines typically have a single output. Neural networks can have multiple output neurons. To get around this issue, this class will create multiple SVM's if there is more than one output specified. Because a SVM is trained quite differently from a neural network, none of the neural network training classes will work. This class must be trained using SVMTrain.

Examples of com.nr.ci.Svm

      data[i][1]=-1.0-(a+b);
    }
    
    // Linear kernel
    Svmlinkernel linkernel=new Svmlinkernel(data,y);
    Svm linsvm=new Svm(linkernel);
    lambda=10;
    k=0;
    do {
      test=linsvm.relax(lambda,omega);
//      System.out.printf(test);
      k++;
    } while (test > 1.e-3 && k < 100);
    int nerror=0;
    for (i=0;i<M;i++) {
//      if (i%10 == 0) System.out.printf((y[i]==1.0) << " %f\n", (linsvm.predict(i) >= 1.0));
//      if ((y[i] == 1.0) != (linsvm.predict(i) >= 1.0))
//        System.out.printf(data[i][0] << " %f\n", data[i][1] << " %f\n", y[i] << " %f\n", linsvm.predict(i));
      nerror += ((y[i]==1.0) != (linsvm.predict(i) >= 0.0) ? 1 : 0);
    }
    System.out.printf("Errors: %d\n", nerror);


    // Polynomial kernel
    Svmpolykernel polykernel=new Svmpolykernel(data,y,1.0,1.0,2.0);
    Svm polysvm=new Svm(polykernel);
    lambda=10;
    k=0;
    do {
      test=polysvm.relax(lambda,omega);
//      System.out.printf(test);
      k++;
    } while (test > 1.e-3 && k < 100);
    nerror=0;
    for (i=0;i<M;i++) {
      nerror += ((y[i]==1.0) != (polysvm.predict(i) >= 0.0) ? 1 : 0);
    }
    System.out.printf("Errors: %d\n", nerror);


    // Gaussian kernel
    Svmgausskernel gausskernel=new Svmgausskernel(data,y,1.0);
    Svm gausssvm=new Svm(gausskernel);
    lambda=10;
    k=0;
    do {
      test=gausssvm.relax(lambda,omega);
//      System.out.printf(test);
      k++;
    } while (test > 1.e-3 && k < 100);
    nerror=0;
    for (i=0;i<M;i++) {
      nerror += ((y[i]==1.0) != (gausssvm.predict(i) >= 0.0) ? 1 : 0);
    }
    System.out.printf("Errors: %d\n", nerror);


    
    // Need to add tests for harder test case and resolve issue that the two
    // support vectors give an erroneous indication for two of the kernels above


    // Example similar to the book
    Normaldev ndev=new Normaldev(0.0,0.5,17);
    for (j=0;j<4;j++) {   // Four quadrants
      for (i=0;i<M/4;i++) {
        k=(M/4)*j+i;
        if (j == 0) {
          y[k]=1.0;
          data[k][0]=1.0+ndev.dev();
          data[k][1]=1.0+ndev.dev();
        } else if (j == 1) {
          y[k]=-1.0;
          data[k][0]=-1.0+ndev.dev();
          data[k][1]=1.0+ndev.dev();
        } else if (j == 2) {
          y[k]=1.0;
          data[k][0]=-1.0+ndev.dev();
          data[k][1]=-1.0+ndev.dev();
        } else {
          y[k]=-1.0;
          data[k][0]=1.0+ndev.dev();
          data[k][1]=-1.0+ndev.dev();
        }
      }
    }
        
    // Linear kernel
    Svmlinkernel linkernel2=new Svmlinkernel(data,y);
    Svm linsvm2=new Svm(linkernel2);
    System.out.printf("Errors: ");
    for (lambda=0.001;lambda<10000;lambda *= 10) {
      k=0;
      do {
        test=linsvm2.relax(lambda,omega);
//        System.out.printf(test);
        k++;
      } while (test > 1.e-3 && k < 100);
      nerror=0;
      for (i=0;i<M;i++) {
        nerror += ((y[i]==1.0) != (linsvm2.predict(i) >= 0.0) ? 1 : 0);
      }
      System.out.printf("%d ",nerror);
      // Test new data
      nerror=0;
      for (j=0;j<4;j++) {   // Four quadrants
        for (i=0;i<M/4;i++) {
          if (j == 0) {
            yy=1.0;
            x[0]=1.0+ndev.dev();
            x[1]=1.0+ndev.dev();
          } else if (j == 1) {
            yy=-1.0;
            x[0]=-1.0+ndev.dev();
            x[1]=1.0+ndev.dev();
          } else if (j == 2) {
            yy=1.0;
            x[0]=-1.0+ndev.dev();
            x[1]=-1.0+ndev.dev();
          } else {
            yy=-1.0;
            x[0]=1.0+ndev.dev();
            x[1]=-1.0+ndev.dev();
          }
          nerror += ((yy==1.0) != (linsvm2.predict(x) >= 0.0) ? 1 : 0);
        }
      }
      System.out.printf("%d    ",nerror);
    }
    System.out.println();


    // Polynomial kernel
    Svmpolykernel polykernel2 = new Svmpolykernel(data,y,1.0,1.0,4.0);
    Svm polysvm2=new Svm(polykernel2);
    System.out.printf("Errors: ");
    for (lambda=0.001;lambda<10000;lambda *= 10) {
      k=0;
      do {
        test=polysvm2.relax(lambda,omega);
//        System.out.printf(test);
        k++;
      } while (test > 1.e-3 && k < 100);
      // Test training set
      nerror=0;
      for (i=0;i<M;i++) {
        nerror += ((y[i]==1.0) != (polysvm2.predict(i) >= 0.0) ? 1 : 0);
      }
      System.out.printf("%d ",nerror);
      // Test new data
      nerror=0;
      for (j=0;j<4;j++) {   // Four quadrants
        for (i=0;i<M/4;i++) {
          if (j == 0) {
            yy=1.0;
            x[0]=1.0+ndev.dev();
            x[1]=1.0+ndev.dev();
          } else if (j == 1) {
            yy=-1.0;
            x[0]=-1.0+ndev.dev();
            x[1]=1.0+ndev.dev();
          } else if (j == 2) {
            yy=1.0;
            x[0]=-1.0+ndev.dev();
            x[1]=-1.0+ndev.dev();
          } else {
            yy=-1.0;
            x[0]=1.0+ndev.dev();
            x[1]=-1.0+ndev.dev();
          }
          nerror += ((yy==1.0) != (polysvm2.predict(x) >= 0.0) ? 1 : 0);
        }
      }
      System.out.printf("%d    ",nerror);
    }
    System.out.println();


    // Gaussian kernel
    Svmgausskernel gausskernel2=new Svmgausskernel(data,y,1.0);
    Svm gausssvm2=new Svm(gausskernel2);
    System.out.printf("Errors: ");
    for (lambda=0.001;lambda<10000;lambda *= 10) {
      k=0;
      do {
        test=gausssvm2.relax(lambda,omega);
//        System.out.printf(test);
        k++;
      } while (test > 1.e-3 && k < 100);
      nerror=0;
      for (i=0;i<M;i++) {
        nerror += ((y[i]==1.0) != (gausssvm2.predict(i) >= 0.0) ? 1 : 0);
      }
      System.out.printf("%d ",nerror);
      // Test new data
      nerror=0;
      for (j=0;j<4;j++) {   // Four quadrants
        for (i=0;i<M/4;i++) {
          if (j == 0) {
            yy=1.0;
            x[0]=1.0+ndev.dev();
            x[1]=1.0+ndev.dev();
          } else if (j == 1) {
            yy=-1.0;
            x[0]=-1.0+ndev.dev();
            x[1]=1.0+ndev.dev();
          } else if (j == 2) {
            yy=1.0;
            x[0]=-1.0+ndev.dev();
            x[1]=-1.0+ndev.dev();
          } else {
            yy=-1.0;
            x[0]=1.0+ndev.dev();
            x[1]=-1.0+ndev.dev();
          }
          nerror += ((yy==1.0) != (gausssvm2.predict(x) >= 0.0) ? 1 : 0);
        }
      }
      System.out.printf("%d    ",nerror);
    }
    System.out.println();

View Full Code Here

Examples of org.encog.ml.svm.SVM

  public final File SERIAL_FILENAME = TEMP_DIR.createFile("encogtest.ser");
  
  private SVM create()
  {
    MLDataSet training = new BasicMLDataSet(XOR.XOR_INPUT,XOR.XOR_IDEAL);
    SVM result = new SVM(2,SVMType.EpsilonSupportVectorRegression,KernelType.RadialBasisFunction);
    final SVMTrain train = new SVMTrain(result, training);
    train.iteration();
    return result;
  }

View Full Code Here

Examples of org.encog.ml.svm.SVM

    return result;
  }
  
  public void testPersistEG()
  {
    SVM network = create();


    EncogDirectoryPersistence.saveObject((EG_FILENAME), network);
    SVM network2 = (SVM)EncogDirectoryPersistence.loadObject((EG_FILENAME));
    validate(network2);
  }

View Full Code Here

Examples of org.encog.ml.svm.SVM

    validate(network2);
  }
  
  public void testPersistSerial() throws IOException, ClassNotFoundException
  {
    SVM network = create();
    
    SerializeObject.save(SERIAL_FILENAME, network);
    SVM network2 = (SVM)SerializeObject.load(SERIAL_FILENAME);
        
    validate(network2);
  }

View Full Code Here

Examples of org.encog.ml.svm.SVM

  @Override
  public final MLMethod generate() {
    if (this.outputNeurons != 1) {
      throw new PatternError("A SVM may only have one output.");
    }
    final SVM network = new SVM(this.inputNeurons, this.svmType,
        this.kernelType);
    return network;
  }

View Full Code Here

Examples of org.encog.ml.svm.SVM

    return temp.process(this.normalizedSunspots);
  }
  
  public SVM createNetwork()
  {
    SVM network = new SVM(WINDOW_SIZE,true);
    return network;
  }

View Full Code Here

Examples of org.encog.ml.svm.SVM

  }
  
  public void run()
  {
    normalizeSunspots(0.1,0.9);
    SVM network = createNetwork();
    MLDataSet training = generateTraining();
    train(network,training);
    predict(network);
    
  }

View Full Code Here

Examples of org.encog.ml.svm.SVM

  public void finishTraining() {
    stop();
  }


  private SVM generateSVM() {
    final SVM svm = new SVM(this.modelSVM.getInputCount(),
        this.modelSVM.getSVMType(), this.modelSVM.getKernelType());
    return svm;
  }

View Full Code Here

Examples of org.encog.ml.svm.SVM

   * This method creates, and trains, a SVM with the best const and gamma.
   * @return The best SVM. 
   */
  @Override
  public MLMethod getMethod() {
    final SVM result = generateSVM();
    result.getParams().C = this.bestConst;
    result.getParams().gamma = this.bestGamma;
    result.setModel(svm.svm_train(this.problem, result.getParams()));
    return result;
  }

View Full Code Here

Examples of org.encog.ml.svm.SVM

  public Object requestNextTask() {
    if (this.done || getShouldStop()) {
      return null;
    }


    final SVM svm = generateSVM();


    // advance
    this.currentConst += this.constStep;
    if (this.currentConst > this.constEnd) {
      this.currentConst = this.constBegin;

View Full Code Here

0 1 2

TOP

All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.