Free cookie consent management tool by TermsFeed Policy Generator

Changeset 12934


Ignore:
Timestamp:
09/02/15 17:08:29 (9 years ago)
Author:
gkronber
Message:

#2385 added a boolean "CreateSolution" parameter for support vector machine algorithms and added model error/accuracy metrics as algorithm results (to allow grid search without creating solutions)

Location:
trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/SupportVectorMachine
Files:
2 edited

Legend:

Unmodified
Added
Removed
  • trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/SupportVectorMachine/SupportVectorClassification.cs

    r12509 r12934  
    4646    private const string GammaParameterName = "Gamma";
    4747    private const string DegreeParameterName = "Degree";
     48    private const string CreateSolutionParameterName = "CreateSolution";
    4849
    4950    #region parameter properties
     
    6566    public IValueParameter<IntValue> DegreeParameter {
    6667      get { return (IValueParameter<IntValue>)Parameters[DegreeParameterName]; }
     68    }
     69    public IFixedValueParameter<BoolValue> CreateSolutionParameter {
     70      get { return (IFixedValueParameter<BoolValue>)Parameters[CreateSolutionParameterName]; }
    6771    }
    6872    #endregion
     
    8791    public IntValue Degree {
    8892      get { return DegreeParameter.Value; }
     93    }
     94    public bool CreateSolution {
     95      get { return CreateSolutionParameter.Value.Value; }
     96      set { CreateSolutionParameter.Value.Value = value; }
    8997    }
    9098    #endregion
     
    112120      Parameters.Add(new ValueParameter<DoubleValue>(GammaParameterName, "The value of the gamma parameter in the kernel function.", new DoubleValue(1.0)));
    113121      Parameters.Add(new ValueParameter<IntValue>(DegreeParameterName, "The degree parameter for the polynomial kernel function.", new IntValue(3)));
     122      Parameters.Add(new FixedValueParameter<BoolValue>(CreateSolutionParameterName, "Flag that indicates if a solution should be produced at the end of the run", new BoolValue(true)));
     123      Parameters[CreateSolutionParameterName].Hidden = true;
    114124    }
    115125    [StorableHook(HookType.AfterDeserialization)]
    116126    private void AfterDeserialization() {
    117127      #region backwards compatibility (change with 3.4)
    118       if (!Parameters.ContainsKey(DegreeParameterName))
    119         Parameters.Add(new ValueParameter<IntValue>(DegreeParameterName, "The degree parameter for the polynomial kernel function.", new IntValue(3)));
     128      if (!Parameters.ContainsKey(DegreeParameterName)) {
     129        Parameters.Add(new ValueParameter<IntValue>(DegreeParameterName,
     130          "The degree parameter for the polynomial kernel function.", new IntValue(3)));
     131      }
     132      if (!Parameters.ContainsKey(CreateSolutionParameterName)) {
     133        Parameters.Add(new FixedValueParameter<BoolValue>(CreateSolutionParameterName,
     134          "Flag that indicates if a solution should be produced at the end of the run", new BoolValue(true)));
     135        Parameters[CreateSolutionParameterName].Hidden = true;
     136      }
    120137      #endregion
    121138    }
     
    129146      IClassificationProblemData problemData = Problem.ProblemData;
    130147      IEnumerable<string> selectedInputVariables = problemData.AllowedInputVariables;
    131       double trainingAccuracy, testAccuracy;
    132148      int nSv;
    133       var solution = CreateSupportVectorClassificationSolution(problemData, selectedInputVariables,
    134         SvmType.Value, KernelType.Value, Cost.Value, Nu.Value, Gamma.Value, Degree.Value,
    135         out trainingAccuracy, out testAccuracy, out nSv);
    136 
    137       Results.Add(new Result("Support vector classification solution", "The support vector classification solution.", solution));
    138       Results.Add(new Result("Training accuracy", "The accuracy of the SVR solution on the training partition.", new DoubleValue(trainingAccuracy)));
    139       Results.Add(new Result("Test accuracy", "The accuracy of the SVR solution on the test partition.", new DoubleValue(testAccuracy)));
    140       Results.Add(new Result("Number of support vectors", "The number of support vectors of the SVR solution.", new IntValue(nSv)));
     149      ISupportVectorMachineModel model;
     150
     151      Run(problemData, selectedInputVariables, GetSvmType(SvmType.Value), GetKernelType(KernelType.Value), Cost.Value, Nu.Value, Gamma.Value, Degree.Value, out model, out nSv);
     152
     153      if (CreateSolution) {
     154        var solution = new SupportVectorClassificationSolution((SupportVectorMachineModel)model, (IClassificationProblemData)problemData.Clone());
     155        Results.Add(new Result("Support vector classification solution", "The support vector classification solution.",
     156          solution));
     157      }
     158
     159      {
     160        // calculate classification metrics
     161        // calculate regression model metrics
     162        var ds = problemData.Dataset;
     163        var trainRows = problemData.TrainingIndices;
     164        var testRows = problemData.TestIndices;
     165        var yTrain = ds.GetDoubleValues(problemData.TargetVariable, trainRows);
     166        var yTest = ds.GetDoubleValues(problemData.TargetVariable, testRows);
     167        var yPredTrain = model.GetEstimatedClassValues(ds, trainRows);
     168        var yPredTest = model.GetEstimatedClassValues(ds, testRows);
     169
     170        OnlineCalculatorError error;
     171        var trainAccuracy = OnlineAccuracyCalculator.Calculate(yPredTrain, yTrain, out error);
     172        if (error != OnlineCalculatorError.None) trainAccuracy = double.MaxValue;
     173        var testAccuracy = OnlineAccuracyCalculator.Calculate(yPredTest, yTest, out error);
     174        if (error != OnlineCalculatorError.None) testAccuracy = double.MaxValue;
     175
     176        Results.Add(new Result("Accuracy (training)", "The mean of squared errors of the SVR solution on the training partition.", new DoubleValue(trainAccuracy)));
     177        Results.Add(new Result("Accuracy (test)", "The mean of squared errors of the SVR solution on the test partition.", new DoubleValue(testAccuracy)));
     178
     179        Results.Add(new Result("Number of support vectors", "The number of support vectors of the SVR solution.",
     180          new IntValue(nSv)));
     181      }
    141182    }
    142183
     
    147188    }
    148189
     190    // BackwardsCompatibility3.4
     191    #region Backwards compatible code, remove with 3.5
    149192    public static SupportVectorClassificationSolution CreateSupportVectorClassificationSolution(IClassificationProblemData problemData, IEnumerable<string> allowedInputVariables,
    150193      int svmType, int kernelType, double cost, double nu, double gamma, int degree, out double trainingAccuracy, out double testAccuracy, out int nSv) {
     194
     195      ISupportVectorMachineModel model;
     196      Run(problemData, allowedInputVariables, svmType, kernelType, cost, nu, gamma, degree, out model, out nSv);
     197      var solution = new SupportVectorClassificationSolution((SupportVectorMachineModel)model, (IClassificationProblemData)problemData.Clone());
     198
     199      trainingAccuracy = solution.TrainingAccuracy;
     200      testAccuracy = solution.TestAccuracy;
     201
     202      return solution;
     203    }
     204
     205    #endregion
     206
     207    public static void Run(IClassificationProblemData problemData, IEnumerable<string> allowedInputVariables,
     208      int svmType, int kernelType, double cost, double nu, double gamma, int degree,
     209      out ISupportVectorMachineModel model, out int nSv) {
    151210      var dataset = problemData.Dataset;
    152211      string targetVariable = problemData.TargetVariable;
     
    154213
    155214      //extract SVM parameters from scope and set them
    156       svm_parameter parameter = new svm_parameter();
    157       parameter.svm_type = svmType;
    158       parameter.kernel_type = kernelType;
    159       parameter.C = cost;
    160       parameter.nu = nu;
    161       parameter.gamma = gamma;
    162       parameter.cache_size = 500;
    163       parameter.probability = 0;
    164       parameter.eps = 0.001;
    165       parameter.degree = degree;
    166       parameter.shrinking = 1;
    167       parameter.coef0 = 0;
     215      svm_parameter parameter = new svm_parameter {
     216        svm_type = svmType,
     217        kernel_type = kernelType,
     218        C = cost,
     219        nu = nu,
     220        gamma = gamma,
     221        cache_size = 500,
     222        probability = 0,
     223        eps = 0.001,
     224        degree = degree,
     225        shrinking = 1,
     226        coef0 = 0
     227      };
    168228
    169229      var weightLabels = new List<int>();
     
    182242      parameter.weight = weights.ToArray();
    183243
    184 
    185244      svm_problem problem = SupportVectorMachineUtil.CreateSvmProblem(dataset, targetVariable, allowedInputVariables, rows);
    186245      RangeTransform rangeTransform = RangeTransform.Compute(problem);
    187246      svm_problem scaledProblem = rangeTransform.Scale(problem);
    188247      var svmModel = svm.svm_train(scaledProblem, parameter);
    189       var model = new SupportVectorMachineModel(svmModel, rangeTransform, targetVariable, allowedInputVariables, problemData.ClassValues);
    190       var solution = new SupportVectorClassificationSolution(model, (IClassificationProblemData)problemData.Clone());
    191 
    192248      nSv = svmModel.SV.Length;
    193       trainingAccuracy = solution.TrainingAccuracy;
    194       testAccuracy = solution.TestAccuracy;
    195 
    196       return solution;
     249
     250      model = new SupportVectorMachineModel(svmModel, rangeTransform, targetVariable, allowedInputVariables, problemData.ClassValues);
    197251    }
    198252
  • trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4/SupportVectorMachine/SupportVectorRegression.cs

    r12509 r12934  
    4747    private const string EpsilonParameterName = "Epsilon";
    4848    private const string DegreeParameterName = "Degree";
     49    private const string CreateSolutionParameterName = "CreateSolution";
    4950
    5051    #region parameter properties
     
    6970    public IValueParameter<IntValue> DegreeParameter {
    7071      get { return (IValueParameter<IntValue>)Parameters[DegreeParameterName]; }
     72    }
     73    public IFixedValueParameter<BoolValue> CreateSolutionParameter {
     74      get { return (IFixedValueParameter<BoolValue>)Parameters[CreateSolutionParameterName]; }
    7175    }
    7276    #endregion
     
    9498    public IntValue Degree {
    9599      get { return DegreeParameter.Value; }
     100    }
     101    public bool CreateSolution {
     102      get { return CreateSolutionParameter.Value.Value; }
     103      set { CreateSolutionParameter.Value.Value = value; }
    96104    }
    97105    #endregion
     
    120128      Parameters.Add(new ValueParameter<DoubleValue>(EpsilonParameterName, "The value of the epsilon parameter for epsilon-SVR.", new DoubleValue(0.1)));
    121129      Parameters.Add(new ValueParameter<IntValue>(DegreeParameterName, "The degree parameter for the polynomial kernel function.", new IntValue(3)));
     130      Parameters.Add(new FixedValueParameter<BoolValue>(CreateSolutionParameterName, "Flag that indicates if a solution should be produced at the end of the run", new BoolValue(true)));
     131      Parameters[CreateSolutionParameterName].Hidden = true;
    122132    }
    123133    [StorableHook(HookType.AfterDeserialization)]
    124134    private void AfterDeserialization() {
    125135      #region backwards compatibility (change with 3.4)
    126       if (!Parameters.ContainsKey(DegreeParameterName))
    127         Parameters.Add(new ValueParameter<IntValue>(DegreeParameterName, "The degree parameter for the polynomial kernel function.", new IntValue(3)));
     136
     137      if (!Parameters.ContainsKey(DegreeParameterName)) {
     138        Parameters.Add(new ValueParameter<IntValue>(DegreeParameterName,
     139          "The degree parameter for the polynomial kernel function.", new IntValue(3)));
     140      }
     141      if (!Parameters.ContainsKey(CreateSolutionParameterName)) {
     142        Parameters.Add(new FixedValueParameter<BoolValue>(CreateSolutionParameterName, "Flag that indicates if a solution should be produced at the end of the run", new BoolValue(true)));
     143        Parameters[CreateSolutionParameterName].Hidden = true;
     144      }
    128145      #endregion
    129146    }
     
    137154      IRegressionProblemData problemData = Problem.ProblemData;
    138155      IEnumerable<string> selectedInputVariables = problemData.AllowedInputVariables;
    139       double trainR2, testR2;
    140156      int nSv;
    141       var solution = CreateSupportVectorRegressionSolution(problemData, selectedInputVariables, SvmType.Value,
    142         KernelType.Value, Cost.Value, Nu.Value, Gamma.Value, Epsilon.Value, Degree.Value,
    143         out trainR2, out testR2, out nSv);
    144 
    145       Results.Add(new Result("Support vector regression solution", "The support vector regression solution.", solution));
    146       Results.Add(new Result("Training R²", "The Pearson's R² of the SVR solution on the training partition.", new DoubleValue(trainR2)));
    147       Results.Add(new Result("Test R²", "The Pearson's R² of the SVR solution on the test partition.", new DoubleValue(testR2)));
     157      ISupportVectorMachineModel model;
     158      Run(problemData, selectedInputVariables, SvmType.Value, KernelType.Value, Cost.Value, Nu.Value, Gamma.Value, Epsilon.Value, Degree.Value, out model, out nSv);
     159
     160      if (CreateSolution) {
     161        var solution = new SupportVectorRegressionSolution((SupportVectorMachineModel)model, (IRegressionProblemData)problemData.Clone());
     162        Results.Add(new Result("Support vector regression solution", "The support vector regression solution.", solution));
     163      }
     164
    148165      Results.Add(new Result("Number of support vectors", "The number of support vectors of the SVR solution.", new IntValue(nSv)));
    149     }
    150 
    151     public static SupportVectorRegressionSolution CreateSupportVectorRegressionSolution(IRegressionProblemData problemData, IEnumerable<string> allowedInputVariables,
     166
     167
     168      {
     169        // calculate regression model metrics
     170        var ds = problemData.Dataset;
     171        var trainRows = problemData.TrainingIndices;
     172        var testRows = problemData.TestIndices;
     173        var yTrain = ds.GetDoubleValues(problemData.TargetVariable, trainRows);
     174        var yTest = ds.GetDoubleValues(problemData.TargetVariable, testRows);
     175        var yPredTrain = model.GetEstimatedValues(ds, trainRows).ToArray();
     176        var yPredTest = model.GetEstimatedValues(ds, testRows).ToArray();
     177
     178        OnlineCalculatorError error;
     179        var trainMse = OnlineMeanSquaredErrorCalculator.Calculate(yPredTrain, yTrain, out error);
     180        if (error != OnlineCalculatorError.None) trainMse = double.MaxValue;
     181        var testMse = OnlineMeanSquaredErrorCalculator.Calculate(yPredTest, yTest, out error);
     182        if (error != OnlineCalculatorError.None) testMse = double.MaxValue;
     183
     184        Results.Add(new Result("Mean squared error (training)", "The mean of squared errors of the SVR solution on the training partition.", new DoubleValue(trainMse)));
     185        Results.Add(new Result("Mean squared error (test)", "The mean of squared errors of the SVR solution on the test partition.", new DoubleValue(testMse)));
     186
     187
     188        var trainMae = OnlineMeanAbsoluteErrorCalculator.Calculate(yPredTrain, yTrain, out error);
     189        if (error != OnlineCalculatorError.None) trainMae = double.MaxValue;
     190        var testMae = OnlineMeanAbsoluteErrorCalculator.Calculate(yPredTest, yTest, out error);
     191        if (error != OnlineCalculatorError.None) testMae = double.MaxValue;
     192
     193        Results.Add(new Result("Mean absolute error (training)", "The mean of absolute errors of the SVR solution on the training partition.", new DoubleValue(trainMae)));
     194        Results.Add(new Result("Mean absolute error (test)", "The mean of absolute errors of the SVR solution on the test partition.", new DoubleValue(testMae)));
     195
     196
     197        var trainRelErr = OnlineMeanAbsolutePercentageErrorCalculator.Calculate(yPredTrain, yTrain, out error);
     198        if (error != OnlineCalculatorError.None) trainRelErr = double.MaxValue;
     199        var testRelErr = OnlineMeanAbsolutePercentageErrorCalculator.Calculate(yPredTest, yTest, out error);
     200        if (error != OnlineCalculatorError.None) testRelErr = double.MaxValue;
     201
     202        Results.Add(new Result("Average relative error (training)", "The mean of relative errors of the SVR solution on the training partition.", new DoubleValue(trainRelErr)));
     203        Results.Add(new Result("Average relative error (test)", "The mean of relative errors of the SVR solution on the test partition.", new DoubleValue(testRelErr)));
     204      }
     205    }
     206
     207    // BackwardsCompatibility3.4
     208    #region Backwards compatible code, remove with 3.5
     209    // for compatibility with old API
     210    public static SupportVectorRegressionSolution CreateSupportVectorRegressionSolution(
     211      IRegressionProblemData problemData, IEnumerable<string> allowedInputVariables,
    152212      string svmType, string kernelType, double cost, double nu, double gamma, double epsilon, int degree,
    153213      out double trainingR2, out double testR2, out int nSv) {
     214      ISupportVectorMachineModel model;
     215      Run(problemData, allowedInputVariables, svmType, kernelType, cost, nu, gamma, epsilon, degree, out model, out nSv);
     216
     217      var solution = new SupportVectorRegressionSolution((SupportVectorMachineModel)model, (IRegressionProblemData)problemData.Clone());
     218      trainingR2 = solution.TrainingRSquared;
     219      testR2 = solution.TestRSquared;
     220      return solution;
     221    }
     222    #endregion
     223
     224    public static void Run(IRegressionProblemData problemData, IEnumerable<string> allowedInputVariables,
     225      string svmType, string kernelType, double cost, double nu, double gamma, double epsilon, int degree,
     226      out ISupportVectorMachineModel model, out int nSv) {
    154227      var dataset = problemData.Dataset;
    155228      string targetVariable = problemData.TargetVariable;
     
    157230
    158231      //extract SVM parameters from scope and set them
    159       svm_parameter parameter = new svm_parameter();
    160       parameter.svm_type = GetSvmType(svmType);
    161       parameter.kernel_type = GetKernelType(kernelType);
    162       parameter.C = cost;
    163       parameter.nu = nu;
    164       parameter.gamma = gamma;
    165       parameter.p = epsilon;
    166       parameter.cache_size = 500;
    167       parameter.probability = 0;
    168       parameter.eps = 0.001;
    169       parameter.degree = degree;
    170       parameter.shrinking = 1;
    171       parameter.coef0 = 0;
    172 
    173 
     232      svm_parameter parameter = new svm_parameter {
     233        svm_type = GetSvmType(svmType),
     234        kernel_type = GetKernelType(kernelType),
     235        C = cost,
     236        nu = nu,
     237        gamma = gamma,
     238        p = epsilon,
     239        cache_size = 500,
     240        probability = 0,
     241        eps = 0.001,
     242        degree = degree,
     243        shrinking = 1,
     244        coef0 = 0
     245      };
    174246
    175247      svm_problem problem = SupportVectorMachineUtil.CreateSvmProblem(dataset, targetVariable, allowedInputVariables, rows);
     
    178250      var svmModel = svm.svm_train(scaledProblem, parameter);
    179251      nSv = svmModel.SV.Length;
    180       var model = new SupportVectorMachineModel(svmModel, rangeTransform, targetVariable, allowedInputVariables);
    181       var solution = new SupportVectorRegressionSolution(model, (IRegressionProblemData)problemData.Clone());
    182       trainingR2 = solution.TrainingRSquared;
    183       testR2 = solution.TestRSquared;
    184       return solution;
     252
     253      model = new SupportVectorMachineModel(svmModel, rangeTransform, targetVariable, allowedInputVariables);
    185254    }
    186255
Note: See TracChangeset for help on using the changeset viewer.