Changeset 8664 for trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic

Timestamp:

09/17/12 11:18:40 (12 years ago)

Author:

mkommend

Message:

#1951:

• Added linear scaling parameter to data analysis problems.
• Adapted interfaces, evaluators and analyzers accordingly.
• Added OnlineBoundedMeanSquaredErrorCalculator.
• Adapted symbolic regression sample unit test.

Location:

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4

Files:

• Analyzers/SymbolicDataAnalysisMultiObjectiveAnalyzer.cs (modified) (4 diffs)
• Analyzers/SymbolicDataAnalysisSingleObjectiveAnalyzer.cs (modified) (4 diffs)
• Evaluators/SymbolicDataAnalysisEvaluator.cs (modified) (5 diffs)
• Interfaces/ISymbolicDataAnalysisEvaluator.cs (modified) (1 diff)
• Interfaces/ISymbolicDataAnalysisMultiObjectiveAnalyzer.cs (modified) (1 diff)
• Interfaces/ISymbolicDataAnalysisSingleObjectiveAnalyzer.cs (modified) (1 diff)
• SymbolicDataAnalysisModel.cs (modified) (2 diffs)
• SymbolicDataAnalysisProblem.cs (modified) (10 diffs)

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Analyzers/SymbolicDataAnalysisMultiObjectiveAnalyzer.cs

@@ -20 +20 @@
 #endregion
 
-using System.Collections.Generic;
-using System.Linq;
 using HeuristicLab.Common;
 using HeuristicLab.Core;
 using HeuristicLab.Data;
-using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
-using HeuristicLab.Operators;
 using HeuristicLab.Parameters;
 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
-using HeuristicLab.Optimization;
 
 namespace HeuristicLab.Problems.DataAnalysis.Symbolic {
@@ -39 +34 @@
     private const string QualitiesParameterName = "Qualities";
     private const string MaximizationParameterName = "Maximization";
+    private const string ApplyLinearScalingParameterName = "ApplyLinearScaling";
     #region parameter properties
     public IScopeTreeLookupParameter<DoubleArray> QualitiesParameter {
@@ -45 +41 @@
     public ILookupParameter<BoolArray> MaximizationParameter {
       get { return (ILookupParameter<BoolArray>)Parameters[MaximizationParameterName]; }
+    }
+    public ILookupParameter<BoolValue> ApplyLinearScalingParameter {
+      get { return (ILookupParameter<BoolValue>)Parameters[ApplyLinearScalingParameterName]; }
     }
     #endregion
@@ -64 +63 @@
       Parameters.Add(new ScopeTreeLookupParameter<DoubleArray>(QualitiesParameterName, "The qualities of the trees that should be analyzed."));
       Parameters.Add(new LookupParameter<BoolArray>(MaximizationParameterName, "The directions of optimization for each dimension."));
+      Parameters.Add(new LookupParameter<BoolValue>(ApplyLinearScalingParameterName, "Flag that indicates if the individual should be linearly scaled before evaluating."));
+    }
+
+    [StorableHook(HookType.AfterDeserialization)]
+    private void AfterDeserialization() {
+      if (Parameters.ContainsKey(ApplyLinearScalingParameterName) && Parameters[ApplyLinearScalingParameterName] is LookupParameter<BoolValue>)
+        Parameters.Remove(ApplyLinearScalingParameterName);
+      if (!Parameters.ContainsKey(ApplyLinearScalingParameterName))
+        Parameters.Add(new LookupParameter<BoolValue>(ApplyLinearScalingParameterName, "Flag that indicates if the individual should be linearly scaled before evaluating."));
     }
   }

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Analyzers/SymbolicDataAnalysisSingleObjectiveAnalyzer.cs

@@ -20 +20 @@
 #endregion
 
-using System.Collections.Generic;
-using System.Linq;
 using HeuristicLab.Common;
 using HeuristicLab.Core;
 using HeuristicLab.Data;
-using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
-using HeuristicLab.Operators;
 using HeuristicLab.Parameters;
 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
-using HeuristicLab.Optimization;
 
 namespace HeuristicLab.Problems.DataAnalysis.Symbolic {
@@ -39 +34 @@
     private const string QualityParameterName = "Quality";
     private const string MaximizationParameterName = "Maximization";
+    private const string ApplyLinearScalingParameterName = "ApplyLinearScaling";
+
     #region parameter properties
     public IScopeTreeLookupParameter<DoubleValue> QualityParameter {
@@ -45 +42 @@
     public ILookupParameter<BoolValue> MaximizationParameter {
       get { return (ILookupParameter<BoolValue>)Parameters[MaximizationParameterName]; }
+    }
+    public ILookupParameter<BoolValue> ApplyLinearScalingParameter {
+      get { return (ILookupParameter<BoolValue>)Parameters[ApplyLinearScalingParameterName]; }
     }
     #endregion
@@ -64 +64 @@
       Parameters.Add(new ScopeTreeLookupParameter<DoubleValue>(QualityParameterName, "The qualities of the trees that should be analyzed."));
       Parameters.Add(new LookupParameter<BoolValue>(MaximizationParameterName, "The direction of optimization."));
+      Parameters.Add(new LookupParameter<BoolValue>(ApplyLinearScalingParameterName, "Flag that indicates if the individual should be linearly scaled before evaluating."));
+    }
+
+    [StorableHook(HookType.AfterDeserialization)]
+    private void AfterDeserialization() {
+      if (Parameters.ContainsKey(ApplyLinearScalingParameterName) && !(Parameters[ApplyLinearScalingParameterName] is LookupParameter<BoolValue>))
+        Parameters.Remove(ApplyLinearScalingParameterName);
+      if (!Parameters.ContainsKey(ApplyLinearScalingParameterName))
+        Parameters.Add(new LookupParameter<BoolValue>(ApplyLinearScalingParameterName, "Flag that indicates if the individual should be linearly scaled before evaluating."));
     }
   }

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Evaluators/SymbolicDataAnalysisEvaluator.cs

@@ -44 +44 @@
     private const string EvaluationPartitionParameterName = "EvaluationPartition";
     private const string RelativeNumberOfEvaluatedSamplesParameterName = "RelativeNumberOfEvaluatedSamples";
+    private const string ApplyLinearScalingParameterName = "ApplyLinearScaling";
 
     public override bool CanChangeName { get { return false; } }
@@ -70 +71 @@
       get { return (IValueLookupParameter<PercentValue>)Parameters[RelativeNumberOfEvaluatedSamplesParameterName]; }
     }
+    public ILookupParameter<BoolValue> ApplyLinearScalingParameter {
+      get { return (ILookupParameter<BoolValue>)Parameters[ApplyLinearScalingParameterName]; }
+    }
     #endregion
 
@@ -87 +91 @@
       Parameters.Add(new ValueLookupParameter<DoubleLimit>(EstimationLimitsParameterName, "The upper and lower limit that should be used as cut off value for the output values of symbolic data analysis trees."));
       Parameters.Add(new ValueLookupParameter<PercentValue>(RelativeNumberOfEvaluatedSamplesParameterName, "The relative number of samples of the dataset partition, which should be randomly chosen for evaluation between the start and end index."));
+      Parameters.Add(new LookupParameter<BoolValue>(ApplyLinearScalingParameterName, "Flag that indicates if the individual should be linearly scaled before evaluating."));
+    }
+
+    [StorableHook(HookType.AfterDeserialization)]
+    private void AfterDeserialization() {
+      if (Parameters.ContainsKey(ApplyLinearScalingParameterName) && !(Parameters[ApplyLinearScalingParameterName] is LookupParameter<BoolValue>))
+        Parameters.Remove(ApplyLinearScalingParameterName);
+      if (!Parameters.ContainsKey(ApplyLinearScalingParameterName))
+        Parameters.Add(new LookupParameter<BoolValue>(ApplyLinearScalingParameterName, "Flag that indicates if the individual should be linearly scaled before evaluating."));
     }
 
@@ -94 +107 @@
 
     protected IEnumerable<int> GenerateRowsToEvaluate(double percentageOfRows) {
-
-
       IEnumerable<int> rows;
       int samplesStart = EvaluationPartitionParameter.ActualValue.Start;
@@ -115 +126 @@
       return rows.Where(i => i < testPartitionStart || testPartitionEnd <= i);
     }
+
+    [ThreadStatic]
+    private static double[] cache;
+    protected static void CalculateWithScaling(IEnumerable<double> targetValues, IEnumerable<double> estimatedValues,
+      double lowerEstimationLimit, double upperEstimationLimit,
+      IOnlineCalculator calculator, int maxRows) {
+      if (cache == null || cache.GetLength(0) < maxRows) {
+        cache = new double[maxRows];
+      }
+
+      //calculate linear scaling
+      //the static methods of the calculator could not be used as it performs a check if the enumerators have an equal amount of elements
+      //this is not true if the cache is used
+      int i = 0;
+      var linearScalingCalculator = new OnlineLinearScalingParameterCalculator();
+      var targetValuesEnumerator = targetValues.GetEnumerator();
+      var estimatedValuesEnumerator = estimatedValues.GetEnumerator();
+      while (targetValuesEnumerator.MoveNext() & estimatedValuesEnumerator.MoveNext()) {
+        double target = targetValuesEnumerator.Current;
+        double estimated = estimatedValuesEnumerator.Current;
+        cache[i] = estimated;
+        if (!double.IsNaN(estimated) && !double.IsInfinity(estimated))
+          linearScalingCalculator.Add(estimated, target);
+        i++;
+      }
+      if (linearScalingCalculator.ErrorState == OnlineCalculatorError.None && (targetValuesEnumerator.MoveNext() || estimatedValuesEnumerator.MoveNext()))
+        throw new ArgumentException("Number of elements in target and estimated values enumeration do not match.");
+
+      double alpha = linearScalingCalculator.Alpha;
+      double beta = linearScalingCalculator.Beta;
+      if (linearScalingCalculator.ErrorState != OnlineCalculatorError.None) {
+        alpha = 0.0;
+        beta = 1.0;
+      }
+
+      //calculate the quality by using the passed online calculator
+      targetValuesEnumerator = targetValues.GetEnumerator();
+      var scaledBoundedEstimatedValuesEnumerator = Enumerable.Range(0, i).Select(x => cache[x] * beta + alpha)
+        .LimitToRange(lowerEstimationLimit, upperEstimationLimit).GetEnumerator();
+
+      while (targetValuesEnumerator.MoveNext() & scaledBoundedEstimatedValuesEnumerator.MoveNext()) {
+        calculator.Add(targetValuesEnumerator.Current, scaledBoundedEstimatedValuesEnumerator.Current);
+      }
+    }
   }
 }

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Interfaces/ISymbolicDataAnalysisEvaluator.cs

@@ -29 +29 @@
     IValueLookupParameter<IntRange> EvaluationPartitionParameter { get; }
     IValueLookupParameter<PercentValue> RelativeNumberOfEvaluatedSamplesParameter { get; }
+    ILookupParameter<BoolValue> ApplyLinearScalingParameter { get; }
 
     IValueLookupParameter<T> ProblemDataParameter { get; }

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Interfaces/ISymbolicDataAnalysisMultiObjectiveAnalyzer.cs

@@ -21 +21 @@
 using HeuristicLab.Core;
 using HeuristicLab.Data;
-using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
-using HeuristicLab.Optimization;
-using HeuristicLab.Parameters;
 namespace HeuristicLab.Problems.DataAnalysis.Symbolic {
   public interface ISymbolicDataAnalysisMultiObjectiveAnalyzer : ISymbolicDataAnalysisAnalyzer {
     IScopeTreeLookupParameter<DoubleArray> QualitiesParameter { get; }
     ILookupParameter<BoolArray> MaximizationParameter { get; }
+    ILookupParameter<BoolValue> ApplyLinearScalingParameter { get; }
 
   }

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Interfaces/ISymbolicDataAnalysisSingleObjectiveAnalyzer.cs

@@ -28 +28 @@
     IScopeTreeLookupParameter<DoubleValue> QualityParameter { get; }
     ILookupParameter<BoolValue> MaximizationParameter { get; }
+    ILookupParameter<BoolValue> ApplyLinearScalingParameter { get; }
   }
 }

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/SymbolicDataAnalysisModel.cs

@@ -20 +20 @@
 #endregion
 
+using System;
 using System.Drawing;
 using HeuristicLab.Common;
@@ -66 +67 @@
       this.interpreter = interpreter;
     }
+
+    #region Scaling
+    public static void Scale(ISymbolicDataAnalysisModel model, IDataAnalysisProblemData problemData, string targetVariable) {
+      var dataset = problemData.Dataset;
+      var rows = problemData.TrainingIndices;
+      var estimatedValues = model.Interpreter.GetSymbolicExpressionTreeValues(model.SymbolicExpressionTree, dataset, rows);
+      var targetValues = dataset.GetDoubleValues(targetVariable, rows);
+
+      var linearScalingCalculator = new OnlineLinearScalingParameterCalculator();
+      var targetValuesEnumerator = targetValues.GetEnumerator();
+      var estimatedValuesEnumerator = estimatedValues.GetEnumerator();
+      while (targetValuesEnumerator.MoveNext() & estimatedValuesEnumerator.MoveNext()) {
+        double target = targetValuesEnumerator.Current;
+        double estimated = estimatedValuesEnumerator.Current;
+        if (!double.IsNaN(estimated) && !double.IsInfinity(estimated))
+          linearScalingCalculator.Add(estimated, target);
+      }
+      if (linearScalingCalculator.ErrorState == OnlineCalculatorError.None && (targetValuesEnumerator.MoveNext() || estimatedValuesEnumerator.MoveNext()))
+        throw new ArgumentException("Number of elements in target and estimated values enumeration do not match.");
+
+      double alpha = linearScalingCalculator.Alpha;
+      double beta = linearScalingCalculator.Beta;
+      if (linearScalingCalculator.ErrorState != OnlineCalculatorError.None) return;
+
+      ConstantTreeNode alphaTreeNode = null;
+      ConstantTreeNode betaTreeNode = null;
+      // check if model has been scaled previously by analyzing the structure of the tree
+      var startNode = model.SymbolicExpressionTree.Root.GetSubtree(0);
+      if (startNode.GetSubtree(0).Symbol is Addition) {
+        var addNode = startNode.GetSubtree(0);
+        if (addNode.SubtreeCount == 2 && addNode.GetSubtree(0).Symbol is Multiplication && addNode.GetSubtree(1).Symbol is Constant) {
+          alphaTreeNode = addNode.GetSubtree(1) as ConstantTreeNode;
+          var mulNode = addNode.GetSubtree(0);
+          if (mulNode.SubtreeCount == 2 && mulNode.GetSubtree(1).Symbol is Constant) {
+            betaTreeNode = mulNode.GetSubtree(1) as ConstantTreeNode;
+          }
+        }
+      }
+      // if tree structure matches the structure necessary for linear scaling then reuse the existing tree nodes
+      if (alphaTreeNode != null && betaTreeNode != null) {
+        betaTreeNode.Value *= beta;
+        alphaTreeNode.Value *= beta;
+        alphaTreeNode.Value += alpha;
+      } else {
+        var mainBranch = startNode.GetSubtree(0);
+        startNode.RemoveSubtree(0);
+        var scaledMainBranch = MakeSum(MakeProduct(mainBranch, beta), alpha);
+        startNode.AddSubtree(scaledMainBranch);
+      }
+    }
+
+    private static ISymbolicExpressionTreeNode MakeSum(ISymbolicExpressionTreeNode treeNode, double alpha) {
+      if (alpha.IsAlmost(0.0)) {
+        return treeNode;
+      } else {
+        var addition = new Addition();
+        var node = addition.CreateTreeNode();
+        var alphaConst = MakeConstant(alpha);
+        node.AddSubtree(treeNode);
+        node.AddSubtree(alphaConst);
+        return node;
+      }
+    }
+
+    private static ISymbolicExpressionTreeNode MakeProduct(ISymbolicExpressionTreeNode treeNode, double beta) {
+      if (beta.IsAlmost(1.0)) {
+        return treeNode;
+      } else {
+        var multipliciation = new Multiplication();
+        var node = multipliciation.CreateTreeNode();
+        var betaConst = MakeConstant(beta);
+        node.AddSubtree(treeNode);
+        node.AddSubtree(betaConst);
+        return node;
+      }
+    }
+
+    private static ISymbolicExpressionTreeNode MakeConstant(double c) {
+      var node = (ConstantTreeNode)(new Constant()).CreateTreeNode();
+      node.Value = c;
+      return node;
+    }
+    #endregion
   }
 }

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/SymbolicDataAnalysisProblem.cs

@@ -53 +53 @@
     private const string FitnessCalculationPartitionParameterName = "FitnessCalculationPartition";
     private const string ValidationPartitionParameterName = "ValidationPartition";
+    private const string ApplyLinearScalingParameterName = "ApplyLinearScaling";
 
     private const string ProblemDataParameterDescription = "";
@@ -64 +65 @@
     private const string FitnessCalculationPartitionParameterDescription = "The partition of the problem data training partition, that should be used to calculate the fitness of an individual.";
     private const string ValidationPartitionParameterDescription = "The partition of the problem data training partition, that should be used to select the best model from (optional).";
+    private const string ApplyLinearScalingParameterDescription = "Flag that indicates if the individual should be linearly scaled before evaluating.";
     #endregion
 
@@ -100 +102 @@
       get { return (IFixedValueParameter<IntRange>)Parameters[ValidationPartitionParameterName]; }
     }
+    public IFixedValueParameter<BoolValue> ApplyLinearScalingParameter {
+      get { return (IFixedValueParameter<BoolValue>)Parameters[ApplyLinearScalingParameterName]; }
+    }
     #endregion
 
@@ -144 +149 @@
     public IntRange ValidationPartition {
       get { return ValidationPartitionParameter.Value; }
+    }
+    public BoolValue ApplyLinearScaling {
+      get { return ApplyLinearScalingParameter.Value; }
     }
     #endregion
@@ -151 +159 @@
     [StorableHook(HookType.AfterDeserialization)]
     private void AfterDeserialization() {
+      if (!Parameters.ContainsKey(ApplyLinearScalingParameterName)) {
+        Parameters.Add(new FixedValueParameter<BoolValue>(ApplyLinearScalingParameterName, ApplyLinearScalingParameterDescription, new BoolValue(false)));
+        ApplyLinearScalingParameter.Hidden = true;
+      }
+
       RegisterEventHandlers();
     }
@@ -170 +183 @@
       Parameters.Add(new FixedValueParameter<IntRange>(ValidationPartitionParameterName, ValidationPartitionParameterDescription));
       Parameters.Add(new FixedValueParameter<PercentValue>(RelativeNumberOfEvaluatedSamplesParameterName, RelativeNumberOfEvaluatedSamplesParameterDescription, new PercentValue(1)));
+      Parameters.Add(new FixedValueParameter<BoolValue>(ApplyLinearScalingParameterName, ApplyLinearScalingParameterDescription, new BoolValue(false)));
 
       SymbolicExpressionTreeInterpreterParameter.Hidden = true;
       MaximumFunctionArgumentsParameter.Hidden = true;
       MaximumFunctionDefinitionsParameter.Hidden = true;
+      ApplyLinearScalingParameter.Hidden = true;
 
       SymbolicExpressionTreeGrammar = new TypeCoherentExpressionGrammar();
@@ -274 +289 @@
 
       foreach (var op in operators.OfType<ISymbolicExpressionTreeGrammarBasedOperator>()) {
-        op.SymbolicExpressionTreeGrammarParameter.ActualName = SymbolicExpressionTreeGrammarParameterName;
+        op.SymbolicExpressionTreeGrammarParameter.ActualName = SymbolicExpressionTreeGrammarParameter.Name;
       }
       foreach (var op in operators.OfType<ISymbolicExpressionTreeSizeConstraintOperator>()) {
-        op.MaximumSymbolicExpressionTreeDepthParameter.ActualName = MaximumSymbolicExpressionTreeDepthParameterName;
-        op.MaximumSymbolicExpressionTreeLengthParameter.ActualName = MaximumSymbolicExpressionTreeLengthParameterName;
+        op.MaximumSymbolicExpressionTreeDepthParameter.ActualName = MaximumSymbolicExpressionTreeDepthParameter.Name;
+        op.MaximumSymbolicExpressionTreeLengthParameter.ActualName = MaximumSymbolicExpressionTreeLengthParameter.Name;
       }
       foreach (var op in operators.OfType<ISymbolicExpressionTreeArchitectureAlteringOperator>()) {
-        op.MaximumFunctionArgumentsParameter.ActualName = MaximumFunctionArgumentsParameterName;
-        op.MaximumFunctionDefinitionsParameter.ActualName = MaximumFunctionDefinitionsParameterName;
+        op.MaximumFunctionArgumentsParameter.ActualName = MaximumFunctionArgumentsParameter.Name;
+        op.MaximumFunctionDefinitionsParameter.ActualName = MaximumFunctionDefinitionsParameter.Name;
       }
       foreach (var op in operators.OfType<ISymbolicDataAnalysisEvaluator<T>>()) {
@@ -289 +304 @@
         op.EvaluationPartitionParameter.ActualName = FitnessCalculationPartitionParameter.Name;
         op.RelativeNumberOfEvaluatedSamplesParameter.ActualName = RelativeNumberOfEvaluatedSamplesParameter.Name;
+        op.ApplyLinearScalingParameter.ActualName = ApplyLinearScalingParameter.Name;
       }
       foreach (var op in operators.OfType<ISymbolicExpressionTreeCrossover>()) {
@@ -300 +316 @@
         op.SymbolicExpressionTreeParameter.ActualName = SolutionCreator.SymbolicExpressionTreeParameter.ActualName;
       }
+      foreach (var op in operators.OfType<ISymbolicDataAnalysisSingleObjectiveAnalyzer>()) {
+        op.ApplyLinearScalingParameter.ActualName = ApplyLinearScalingParameter.Name;
+      }
+      foreach (var op in operators.OfType<ISymbolicDataAnalysisMultiObjectiveAnalyzer>()) {
+        op.ApplyLinearScalingParameter.ActualName = ApplyLinearScalingParameter.Name;
+      }
       foreach (var op in operators.OfType<ISymbolicDataAnalysisAnalyzer>()) {
         op.SymbolicExpressionTreeParameter.ActualName = SolutionCreator.SymbolicExpressionTreeParameter.ActualName;
@@ -308 +330 @@
       }
       foreach (var op in operators.OfType<ISymbolicDataAnalysisInterpreterOperator>()) {
-        op.SymbolicDataAnalysisTreeInterpreterParameter.ActualName = SymbolicExpressionTreeInterpreterParameterName;
+        op.SymbolicDataAnalysisTreeInterpreterParameter.ActualName = SymbolicExpressionTreeInterpreterParameter.Name;
       }
       foreach (var op in operators.OfType<ISymbolicDataAnalysisExpressionCrossover<T>>()) {
-        op.EvaluationPartitionParameter.ActualName = FitnessCalculationPartitionParameterName;
+        op.EvaluationPartitionParameter.ActualName = FitnessCalculationPartitionParameter.Name;
         op.ProblemDataParameter.ActualName = ProblemDataParameter.Name;
         op.EvaluationPartitionParameter.ActualName = FitnessCalculationPartitionParameter.Name;