Changeset 12979 for branches/HeuristicLab.EvolutionTracking/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Tracking/SchemaDiversification/UpdateEstimatedValuesOperator.cs

Timestamp:

10/02/15 01:08:30 (9 years ago)

Author:

bburlacu

Message:

#1772:

• added parameters to the SchemaCreator for calculating things in parallel
• added parallel code to the SchemaEvaluator
• implemented quality calculation and saving of estimated values in the scope in the UpdateEstimatedValuesOperator
• small refactoring of QueryMatch (added some useful methods and made NodeInfo internal)
• updated query match unit test

File:

• branches/HeuristicLab.EvolutionTracking/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Tracking/SchemaDiversification/UpdateEstimatedValuesOperator.cs (modified) (3 diffs)

branches/HeuristicLab.EvolutionTracking/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Tracking/SchemaDiversification/UpdateEstimatedValuesOperator.cs

@@ -20 +20 @@
 #endregion
 
+using System;
 using System.Linq;
 using HeuristicLab.Common;
@@ -51 +52 @@
     }
 
+
     public UpdateEstimatedValuesOperator() {
       Parameters.Add(new LookupParameter<IRegressionProblemData>(ProblemDataParameterName));
@@ -74 +76 @@
       var interpreter = InterpreterParameter.ActualValue;
 
-      var estimatedValues = interpreter.GetSymbolicExpressionTreeValues(tree, problemData.Dataset, problemData.TrainingIndices)
-                                       .LimitToRange(estimationLimits.Lower, estimationLimits.Upper).ToArray();
+      var estimatedValues = interpreter.GetSymbolicExpressionTreeValues(tree, problemData.Dataset, problemData.TrainingIndices).ToArray();
+      var targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, problemData.TrainingIndices).ToArray();
+
+      if (estimatedValues.Length != targetValues.Length)
+        throw new ArgumentException("Number of elements in target and estimated values enumeration do not match.");
+
+      var linearScalingCalculator = new OnlineLinearScalingParameterCalculator();
+
+      for (int i = 0; i < estimatedValues.Length; ++i) {
+        var estimated = estimatedValues[i];
+        var target = targetValues[i];
+        if (!double.IsNaN(estimated) && !double.IsInfinity(estimated))
+          linearScalingCalculator.Add(estimated, target);
+      }
+      double alpha = linearScalingCalculator.Alpha;
+      double beta = linearScalingCalculator.Beta;
+      if (linearScalingCalculator.ErrorState != OnlineCalculatorError.None) {
+        alpha = 0.0;
+        beta = 1.0;
+      }
+
+      var scaled = estimatedValues.Select(x => x * beta + alpha).LimitToRange(estimationLimits.Lower, estimationLimits.Upper).ToArray();
+      OnlineCalculatorError error;
+      var r = OnlinePearsonsRCalculator.Calculate(targetValues, scaled, out error);
+      if (error != OnlineCalculatorError.None) r = 0;
+
+      var r2 = r * r;
+      if (r2 > 1.0) r2 = 1.0;
 
       var variables = ExecutionContext.Scope.Variables;
-      if (variables.ContainsKey("EstimatedValues"))
-        variables["EstimatedValues"].Value = new DoubleArray(estimatedValues);
-      else
-        variables.Add(new Core.Variable("EstimatedValues", new DoubleArray(estimatedValues)));
+      ((DoubleValue)variables["Quality"].Value).Value = r2;
+
+      if (variables.ContainsKey("EstimatedValues")) {
+        variables["EstimatedValues"].Value = new DoubleArray(scaled);
+      } else {
+        variables.Add(new Core.Variable("EstimatedValues", new DoubleArray(scaled)));
+      }
       return base.Apply();
     }