Changeset 8276 for branches/DatasetFeatureCorrelation/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Classification

Timestamp:

07/10/12 15:26:13 (12 years ago)

Author:

sforsten

Message:

#1292:

• merged r8034:8179 from trunk
• added BackgroundWorker
• added ProgressBar
• added SpearmansRankCorrelationCoefficientCalculator
• corrected bug in HoeffdingsDependenceCalculator
• made some changes in the GUI

Location:

branches/DatasetFeatureCorrelation/HeuristicLab.Problems.DataAnalysis

Files:

• . (modified) (1 prop)
• 3.4/Implementation/Classification/ClassificationEnsembleSolution.cs (modified) (10 diffs)
• 3.4/Implementation/Classification/ClassificationProblemData.cs (modified) (1 diff)
• 3.4/Implementation/Classification/ClassificationSolution.cs (modified) (2 diffs)
• 3.4/Implementation/Classification/ClassificationSolutionBase.cs (modified) (1 diff)
• 3.4/Implementation/Classification/DiscriminantFunctionClassificationSolution.cs (modified) (2 diffs)
• 3.4/Implementation/Classification/DiscriminantFunctionClassificationSolutionBase.cs (modified) (3 diffs)
• 3.4/Implementation/Classification/ThresholdCalculators/AccuracyMaximizationThresholdCalculator.cs (modified) (2 diffs)

branches/DatasetFeatureCorrelation/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Classification/ClassificationEnsembleSolution.cs

@@ -37 +37 @@
   [Creatable("Data Analysis - Ensembles")]
   public sealed class ClassificationEnsembleSolution : ClassificationSolution, IClassificationEnsembleSolution {
+    private readonly Dictionary<int, double> trainingEvaluationCache = new Dictionary<int, double>();
+    private readonly Dictionary<int, double> testEvaluationCache = new Dictionary<int, double>();
+
     public new IClassificationEnsembleModel Model {
       get { return (IClassificationEnsembleModel)base.Model; }
@@ -85 +88 @@
       }
 
+      trainingEvaluationCache = new Dictionary<int, double>(original.ProblemData.TrainingIndices.Count());
+      testEvaluationCache = new Dictionary<int, double>(original.ProblemData.TestIndices.Count());
+
       classificationSolutions = cloner.Clone(original.classificationSolutions);
       RegisterClassificationSolutionsEventHandler();
@@ -128 +134 @@
       }
 
+      trainingEvaluationCache = new Dictionary<int, double>(problemData.TrainingIndices.Count());
+      testEvaluationCache = new Dictionary<int, double>(problemData.TestIndices.Count());
+
       RegisterClassificationSolutionsEventHandler();
       classificationSolutions.AddRange(solutions);
@@ -148 +157 @@
     public override IEnumerable<double> EstimatedTrainingClassValues {
       get {
-        var rows = ProblemData.TrainingIndizes;
-        var estimatedValuesEnumerators = (from model in Model.Models
-                                          select new { Model = model, EstimatedValuesEnumerator = model.GetEstimatedClassValues(ProblemData.Dataset, rows).GetEnumerator() })
-                                         .ToList();
-        var rowsEnumerator = rows.GetEnumerator();
-        // aggregate to make sure that MoveNext is called for all enumerators 
-        while (rowsEnumerator.MoveNext() & estimatedValuesEnumerators.Select(en => en.EstimatedValuesEnumerator.MoveNext()).Aggregate(true, (acc, b) => acc & b)) {
-          int currentRow = rowsEnumerator.Current;
-
-          var selectedEnumerators = from pair in estimatedValuesEnumerators
-                                    where RowIsTrainingForModel(currentRow, pair.Model) && !RowIsTestForModel(currentRow, pair.Model)
-                                    select pair.EstimatedValuesEnumerator;
-          yield return AggregateEstimatedClassValues(selectedEnumerators.Select(x => x.Current));
+        var rows = ProblemData.TrainingIndices;
+        var rowsToEvaluate = rows.Except(trainingEvaluationCache.Keys);
+        var rowsEnumerator = rowsToEvaluate.GetEnumerator();
+        var valuesEnumerator = GetEstimatedValues(rowsToEvaluate, (r, m) => RowIsTrainingForModel(r, m) && !RowIsTestForModel(r, m)).GetEnumerator();
+
+        while (rowsEnumerator.MoveNext() & valuesEnumerator.MoveNext()) {
+          trainingEvaluationCache.Add(rowsEnumerator.Current, valuesEnumerator.Current);
         }
+
+        return rows.Select(row => trainingEvaluationCache[row]);
       }
     }
@@ -167 +172 @@
     public override IEnumerable<double> EstimatedTestClassValues {
       get {
-        var rows = ProblemData.TestIndizes;
-        var estimatedValuesEnumerators = (from model in Model.Models
-                                          select new { Model = model, EstimatedValuesEnumerator = model.GetEstimatedClassValues(ProblemData.Dataset, rows).GetEnumerator() })
-                                         .ToList();
-        var rowsEnumerator = ProblemData.TestIndizes.GetEnumerator();
-        // aggregate to make sure that MoveNext is called for all enumerators 
-        while (rowsEnumerator.MoveNext() & estimatedValuesEnumerators.Select(en => en.EstimatedValuesEnumerator.MoveNext()).Aggregate(true, (acc, b) => acc & b)) {
-          int currentRow = rowsEnumerator.Current;
-
-          var selectedEnumerators = from pair in estimatedValuesEnumerators
-                                    where RowIsTestForModel(currentRow, pair.Model)
-                                    select pair.EstimatedValuesEnumerator;
-
-          yield return AggregateEstimatedClassValues(selectedEnumerators.Select(x => x.Current));
+        var rows = ProblemData.TestIndices;
+        var rowsToEvaluate = rows.Except(testEvaluationCache.Keys);
+        var rowsEnumerator = rowsToEvaluate.GetEnumerator();
+        var valuesEnumerator = GetEstimatedValues(rowsToEvaluate, RowIsTestForModel).GetEnumerator();
+
+        while (rowsEnumerator.MoveNext() & valuesEnumerator.MoveNext()) {
+          testEvaluationCache.Add(rowsEnumerator.Current, valuesEnumerator.Current);
         }
+
+        return rows.Select(row => testEvaluationCache[row]);
+      }
+    }
+
+    private IEnumerable<double> GetEstimatedValues(IEnumerable<int> rows, Func<int, IClassificationModel, bool> modelSelectionPredicate) {
+      var estimatedValuesEnumerators = (from model in Model.Models
+                                        select new { Model = model, EstimatedValuesEnumerator = model.GetEstimatedClassValues(ProblemData.Dataset, rows).GetEnumerator() })
+                                       .ToList();
+      var rowsEnumerator = rows.GetEnumerator();
+      // aggregate to make sure that MoveNext is called for all enumerators 
+      while (rowsEnumerator.MoveNext() & estimatedValuesEnumerators.Select(en => en.EstimatedValuesEnumerator.MoveNext()).Aggregate(true, (acc, b) => acc & b)) {
+        int currentRow = rowsEnumerator.Current;
+
+        var selectedEnumerators = from pair in estimatedValuesEnumerators
+                                  where modelSelectionPredicate(currentRow, pair.Model)
+                                  select pair.EstimatedValuesEnumerator;
+
+        yield return AggregateEstimatedClassValues(selectedEnumerators.Select(x => x.Current));
       }
     }
@@ -196 +213 @@
 
     public override IEnumerable<double> GetEstimatedClassValues(IEnumerable<int> rows) {
-      return from xs in GetEstimatedClassValueVectors(ProblemData.Dataset, rows)
-             select AggregateEstimatedClassValues(xs);
+      var rowsToEvaluate = rows.Except(evaluationCache.Keys);
+      var rowsEnumerator = rowsToEvaluate.GetEnumerator();
+      var valuesEnumerator = (from xs in GetEstimatedClassValueVectors(ProblemData.Dataset, rowsToEvaluate)
+                              select AggregateEstimatedClassValues(xs))
+                             .GetEnumerator();
+
+      while (rowsEnumerator.MoveNext() & valuesEnumerator.MoveNext()) {
+        evaluationCache.Add(rowsEnumerator.Current, valuesEnumerator.Current);
+      }
+
+      return rows.Select(row => evaluationCache[row]);
     }
 
@@ -223 +249 @@
 
     protected override void OnProblemDataChanged() {
+      trainingEvaluationCache.Clear();
+      testEvaluationCache.Clear();
+      evaluationCache.Clear();
+
       IClassificationProblemData problemData = new ClassificationProblemData(ProblemData.Dataset,
                                                                      ProblemData.AllowedInputVariables,
@@ -251 +281 @@
     public void AddClassificationSolutions(IEnumerable<IClassificationSolution> solutions) {
       classificationSolutions.AddRange(solutions);
+
+      trainingEvaluationCache.Clear();
+      testEvaluationCache.Clear();
+      evaluationCache.Clear();
     }
     public void RemoveClassificationSolutions(IEnumerable<IClassificationSolution> solutions) {
       classificationSolutions.RemoveRange(solutions);
+
+      trainingEvaluationCache.Clear();
+      testEvaluationCache.Clear();
+      evaluationCache.Clear();
     }
 
@@ -275 +313 @@
       trainingPartitions[solution.Model] = solution.ProblemData.TrainingPartition;
       testPartitions[solution.Model] = solution.ProblemData.TestPartition;
+
+      trainingEvaluationCache.Clear();
+      testEvaluationCache.Clear();
+      evaluationCache.Clear();
     }
 
@@ -282 +324 @@
       trainingPartitions.Remove(solution.Model);
       testPartitions.Remove(solution.Model);
+
+      trainingEvaluationCache.Clear();
+      testEvaluationCache.Clear();
+      evaluationCache.Clear();
     }
   }

branches/DatasetFeatureCorrelation/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Classification/ClassificationProblemData.cs

@@ -207 +207 @@
 
     #region parameter properties
-    public ConstrainedValueParameter<StringValue> TargetVariableParameter {
-      get { return (ConstrainedValueParameter<StringValue>)Parameters[TargetVariableParameterName]; }
+    public IConstrainedValueParameter<StringValue> TargetVariableParameter {
+      get { return (IConstrainedValueParameter<StringValue>)Parameters[TargetVariableParameterName]; }
     }
     public IFixedValueParameter<StringMatrix> ClassNamesParameter {

branches/DatasetFeatureCorrelation/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Classification/ClassificationSolution.cs

@@ -44 +44 @@
     public ClassificationSolution(IClassificationModel model, IClassificationProblemData problemData)
       : base(model, problemData) {
-      evaluationCache = new Dictionary<int, double>();
+      evaluationCache = new Dictionary<int, double>(problemData.Dataset.Rows);
     }
 
@@ -51 +51 @@
     }
     public override IEnumerable<double> EstimatedTrainingClassValues {
-      get { return GetEstimatedClassValues(ProblemData.TrainingIndizes); }
+      get { return GetEstimatedClassValues(ProblemData.TrainingIndices); }
     }
     public override IEnumerable<double> EstimatedTestClassValues {
-      get { return GetEstimatedClassValues(ProblemData.TestIndizes); }
+      get { return GetEstimatedClassValues(ProblemData.TestIndices); }
     }
 

branches/DatasetFeatureCorrelation/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Classification/ClassificationSolutionBase.cs

@@ -87 +87 @@
     protected void CalculateResults() {
       double[] estimatedTrainingClassValues = EstimatedTrainingClassValues.ToArray(); // cache values
-      double[] originalTrainingClassValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndizes).ToArray();
+      double[] originalTrainingClassValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndices).ToArray();
       double[] estimatedTestClassValues = EstimatedTestClassValues.ToArray(); // cache values
-      double[] originalTestClassValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TestIndizes).ToArray();
+      double[] originalTestClassValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TestIndices).ToArray();
 
       OnlineCalculatorError errorState;

branches/DatasetFeatureCorrelation/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Classification/DiscriminantFunctionClassificationSolution.cs

@@ -59 +59 @@
     }
     public override IEnumerable<double> EstimatedTrainingClassValues {
-      get { return GetEstimatedClassValues(ProblemData.TrainingIndizes); }
+      get { return GetEstimatedClassValues(ProblemData.TrainingIndices); }
     }
     public override IEnumerable<double> EstimatedTestClassValues {
-      get { return GetEstimatedClassValues(ProblemData.TestIndizes); }
+      get { return GetEstimatedClassValues(ProblemData.TestIndices); }
     }
 
@@ -82 +82 @@
     }
     public override IEnumerable<double> EstimatedTrainingValues {
-      get { return GetEstimatedValues(ProblemData.TrainingIndizes); }
+      get { return GetEstimatedValues(ProblemData.TrainingIndices); }
     }
     public override IEnumerable<double> EstimatedTestValues {
-      get { return GetEstimatedValues(ProblemData.TestIndizes); }
+      get { return GetEstimatedValues(ProblemData.TestIndices); }
     }
 

branches/DatasetFeatureCorrelation/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Classification/DiscriminantFunctionClassificationSolutionBase.cs

@@ -103 +103 @@
     protected void CalculateRegressionResults() {
       double[] estimatedTrainingValues = EstimatedTrainingValues.ToArray(); // cache values
-      double[] originalTrainingValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndizes).ToArray();
+      double[] originalTrainingValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndices).ToArray();
       double[] estimatedTestValues = EstimatedTestValues.ToArray(); // cache values
-      double[] originalTestValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TestIndizes).ToArray();
+      double[] originalTestValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TestIndices).ToArray();
 
       OnlineCalculatorError errorState;
@@ -140 +140 @@
       double[] classValues;
       double[] thresholds;
-      var targetClassValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndizes);
+      var targetClassValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndices);
       AccuracyMaximizationThresholdCalculator.CalculateThresholds(ProblemData, EstimatedTrainingValues, targetClassValues, out classValues, out thresholds);
 
@@ -149 +149 @@
       double[] classValues;
       double[] thresholds;
-      var targetClassValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndizes);
+      var targetClassValues = ProblemData.Dataset.GetDoubleValues(ProblemData.TargetVariable, ProblemData.TrainingIndices);
       NormalDistributionCutPointsThresholdCalculator.CalculateThresholds(ProblemData, EstimatedTrainingValues, targetClassValues, out classValues, out thresholds);
 

branches/DatasetFeatureCorrelation/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Classification/ThresholdCalculators/AccuracyMaximizationThresholdCalculator.cs

@@ -54 +54 @@
     public static void CalculateThresholds(IClassificationProblemData problemData, IEnumerable<double> estimatedValues, IEnumerable<double> targetClassValues, out double[] classValues, out double[] thresholds) {
       int slices = 100;
+      double minThresholdInc = 10e-5; // necessary to prevent infinite loop when maxEstimated - minEstimated is effectively zero (constant model)
       List<double> estimatedValuesList = estimatedValues.ToList();
       double maxEstimatedValue = estimatedValuesList.Max();
       double minEstimatedValue = estimatedValuesList.Min();
-      double thresholdIncrement = (maxEstimatedValue - minEstimatedValue) / slices;
+      double thresholdIncrement = Math.Max((maxEstimatedValue - minEstimatedValue) / slices, minThresholdInc);
       var estimatedAndTargetValuePairs =
         estimatedValuesList.Zip(targetClassValues, (x, y) => new { EstimatedValue = x, TargetClassValue = y })
@@ -70 +71 @@
 
       // incrementally calculate accuracy of all possible thresholds
-      int[,] confusionMatrix = new int[nClasses, nClasses];
-
       for (int i = 1; i < thresholds.Length; i++) {
         double lowerThreshold = thresholds[i - 1];