Package de.bwaldvogel.liblinear

Examples of de.bwaldvogel.liblinear.Parameter

900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
      throw new LiblinearException("The liblinear bias value is not numerical value. ", e);
    }
  }

  public Parameter getLiblinearParameters() throws MaltChainedException {
    Parameter param = new Parameter(SolverType.MCSVM_CS, 0.1, 0.1);
    String type = liblinearOptions.get("s");
   
    if (type.equals("0")) {
      param.setSolverType(SolverType.L2R_LR);
    } else if (type.equals("1")) {
      param.setSolverType(SolverType.L2R_L2LOSS_SVC_DUAL);
    } else if (type.equals("2")) {
      param.setSolverType(SolverType.L2R_L2LOSS_SVC);
    } else if (type.equals("3")) {
      param.setSolverType(SolverType.L2R_L1LOSS_SVC_DUAL);
    } else if (type.equals("4")) {
      param.setSolverType(SolverType.MCSVM_CS);
    } else if (type.equals("5")) {
      param.setSolverType(SolverType.L1R_L2LOSS_SVC); 
    } else if (type.equals("6")) {
      param.setSolverType(SolverType.L1R_LR); 
    } else {
      throw new LiblinearException("The liblinear type (-s) is not an integer value between 0 and 4. ");
    }
    try {
      param.setC(Double.valueOf(liblinearOptions.get("c")).doubleValue());
    } catch (NumberFormatException e) {
      throw new LiblinearException("The liblinear cost (-c) value is not numerical value. ", e);
    }
    try {
      param.setEps(Double.valueOf(liblinearOptions.get("e")).doubleValue());
    } catch (NumberFormatException e) {
      throw new LiblinearException("The liblinear epsilon (-e) value is not numerical value. ", e);
    }
    return param;
  }
View Full Code Here
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
    }
    double[] wmodel = null;
    int[] labels = null;
    int nr_class = 0;
    int nr_feature = 0;
    Parameter parameter = getLiblinearParameters();
    try { 
      Problem problem = readProblem(getInstanceInputStreamReader(".ins"));
      boolean res = checkProblem(problem);
      if (res == false) {
        throw new LibException("Abort (The number of training instances * the number of classes) > "+Integer.MAX_VALUE+" and this is not supported by LibLinear. ");
      }
      if (configLogger.isInfoEnabled()) {
        owner.getGuide().getConfiguration().getConfigLogger().info("- Train a parser model using LibLinear.\n");
      }
      final PrintStream out = System.out;
      final PrintStream err = System.err;
      System.setOut(NoPrintStream.NO_PRINTSTREAM);
      System.setErr(NoPrintStream.NO_PRINTSTREAM);
      Model model = Linear.train(problem, parameter);
      System.setOut(err);
      System.setOut(out);
      problem = null;
      wmodel = model.getFeatureWeights();
      labels = model.getLabels();
      nr_class = model.getNrClass();
      nr_feature = model.getNrFeature();
      if (!saveInstanceFiles) {
        getFile(".ins").delete();
      }
    } catch (OutOfMemoryError e) {
      throw new LibException("Out of memory. Please increase the Java heap size (-Xmx<size>). ", e);
    } catch (IllegalArgumentException e) {
      throw new LibException("The Liblinear learner was not able to redirect Standard Error stream. ", e);
    } catch (SecurityException e) {
      throw new LibException("The Liblinear learner cannot remove the instance file. ", e);
    } catch (NegativeArraySizeException e) {
      throw new LibException("(The number of training instances * the number of classes) > "+Integer.MAX_VALUE+" and this is not supported by LibLinear.", e);
    }
   
    if (configLogger.isInfoEnabled()) {
      configLogger.info("- Optimize the memory usage\n");
    }
    MaltLiblinearModel xmodel = null;
    try {
//      System.out.println("Nr Features:" +  nr_feature);
//      System.out.println("nr_class:" + nr_class);
//      System.out.println("wmodel.length:" + wmodel.length);   
      double[][] wmatrix = convert2(wmodel, nr_class, nr_feature);
      xmodel = new MaltLiblinearModel(labels, nr_class, wmatrix.length, wmatrix, parameter.getSolverType());
      if (configLogger.isInfoEnabled()) {
        configLogger.info("- Save the Liblinear model "+getFile(".moo").getName()+"\n");
      }
    } catch (OutOfMemoryError e) {
      throw new LibException("Out of memory. Please increase the Java heap size (-Xmx<size>). ", e);
View Full Code Here
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
    }
    return true;
  }
 
  private Parameter getLiblinearParameters() throws MaltChainedException {
    Parameter param = new Parameter(SolverType.MCSVM_CS, 0.1, 0.1);
    String type = libOptions.get("s");
   
    if (type.equals("0")) {
      param.setSolverType(SolverType.L2R_LR);
    } else if (type.equals("1")) {
      param.setSolverType(SolverType.L2R_L2LOSS_SVC_DUAL);
    } else if (type.equals("2")) {
      param.setSolverType(SolverType.L2R_L2LOSS_SVC);
    } else if (type.equals("3")) {
      param.setSolverType(SolverType.L2R_L1LOSS_SVC_DUAL);
    } else if (type.equals("4")) {
      param.setSolverType(SolverType.MCSVM_CS);
    } else if (type.equals("5")) {
      param.setSolverType(SolverType.L1R_L2LOSS_SVC); 
    } else if (type.equals("6")) {
      param.setSolverType(SolverType.L1R_LR); 
    } else if (type.equals("7")) {
      param.setSolverType(SolverType.L2R_LR_DUAL); 
    } else {
      throw new LibException("The liblinear type (-s) is not an integer value between 0 and 4. ");
    }
    try {
      param.setC(Double.valueOf(libOptions.get("c")).doubleValue());
    } catch (NumberFormatException e) {
      throw new LibException("The liblinear cost (-c) value is not numerical value. ", e);
    }
    try {
      param.setEps(Double.valueOf(libOptions.get("e")).doubleValue());
    } catch (NumberFormatException e) {
      throw new LibException("The liblinear epsilon (-e) value is not numerical value. ", e);
    }
    return param;
  }
View Full Code Here
78
79
80
81
82
83
84
    ps[0] = 0.1;
    cs = new double[1];
    cs[0] = 1;
    eps = 0.1;
   
    params = new Parameter(solver, cs[0], eps);
  }
View Full Code Here
110
111
112
113
114
115
116
117
118
119
120
121
122
  public Parameter getParams() {
    return params;
  }
 
  public Parameter getParamsCopy() {
    Parameter p2 = new Parameter(params.getSolverType(), params.getC(), params.getEps());
    if (weights != null) {
      p2.setWeights(params.getWeights(), params.getWeightLabels());
    }   
    p2.setEps(params.getEps());
   
    return p2;
  }
View Full Code Here
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
    Problem trainProb = null;
    Problem testProb = null;

    Map<Double, Double> cc = null;

    Parameter linearParams = params.getParamsCopy();

    /*
     * We set the weights here, because LibLINEAR crashes when there are weights for labels that are in the total set, but not in the train set.
     * However in the case of cross-validation for parameter optimisation we cannot make sure that this does not happen, since the CV is done by the LibLINEAR CV method.
     * In the future we should implement our own CV. Moreover, we introduce a slight bias in this way, since the weights are computed over train+test in the CV case.
     *
     */
    if (params.isDoCrossValidation()) {
      target = prob.y;
      cc = EvaluationUtils.computeClassCounts(EvaluationUtils.doubles2target(prob.y));

      /*
      if (params.isDoWeightLabels()) {
        linearParams.setWeights(EvaluationUtils.computeWeights(EvaluationUtils.doubles2target(prob.y)), EvaluationUtils.computeWeightLabels(EvaluationUtils.doubles2target(prob.y)));
      }
       */

    } else {
      trainProb = createProblemTrainSplit(prob, params.getSplitFraction());
      testProb  = createProblemTestSplit(prob, params.getSplitFraction());
      target = testProb.y;

      cc = EvaluationUtils.computeClassCounts(EvaluationUtils.doubles2target(trainProb.y));     

      /*
      if (params.isDoWeightLabels()) {
        linearParams.setWeights(EvaluationUtils.computeWeights(EvaluationUtils.doubles2target(trainProb.y)), EvaluationUtils.computeWeightLabels(EvaluationUtils.doubles2target(trainProb.y)));
      }
       */
    }

    if (params.isDoWeightLabels()) {
      List<Double> wl = new ArrayList<Double>();
      List<Integer> wll = new ArrayList<Integer>();
      for (int i = 0; i < params.getWeightLabels().length; i++) {
        if (cc.containsKey((double) params.getWeightLabels()[i])) {
          wll.add(params.getWeightLabels()[i]);
          wl.add(params.getWeights()[i]);
        }
      }
      linearParams.setWeights(EvaluationUtils.target2Doubles(wl), EvaluationUtils.target2Integers(wll));
    }


    double score = 0, bestScore = 0, bestC = 0, bestP = 0;


    for (double p : params.getPs()) {
      linearParams.setP(p);
      for (double c : params.getCs()) {
        linearParams.setC(c);

        if (params.isDoCrossValidation()) {
          prediction = crossValidate(prob, linearParams, params.getNumFolds());

        } else {
          prediction = testLinearModel(new LibLINEARModel(Linear.train(trainProb, linearParams)), testProb.x);
        }

        score = params.getEvalFunction().computeScore(target, prediction);

        if (bestC == 0 || params.getEvalFunction().isBetter(score, bestScore)) {
          bestC = c;
          bestP = p;
          bestScore = score;
        } 
      }
    }

    linearParams.setC(bestC); 
    System.out.println("Training model for C: " + bestC + " and P (SVR only): " + bestP);
    return new LibLINEARModel(Linear.train(prob, linearParams));
  }
View Full Code Here
689
690
691
692
693
694
695
696
697
698
699
700
701
     *
     * @return the configured svm_parameter object
     */
    protected Parameter getParameters() {

        Parameter parameter = new Parameter(m_SolverType, m_Cost, m_eps);

        if (m_Weight.length > 0) {
            parameter.setWeights(m_Weight, m_WeightLabel);
        }

        return parameter;
    }
View Full Code Here
TOP

Related Classes of de.bwaldvogel.liblinear.Parameter

  • org.data2semantics.proppred.learners.liblinear.LibLINEAR
  • org.data2semantics.proppred.learners.liblinear.LibLINEARParameters
  • org.maltparserx.ml.lib.LibLinear
  • org.maltparserx.ml.liblinear.Liblinear
  • weka.classifiers.functions.LibLINEAR
    • Copyright © 2018 www.massapicom. All rights reserved.
    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.