Changeset 5024 for branches/DataAnalysis.PopulationDiversityAnalysis/HeuristicLab.Problems.DataAnalysis.Regression/3.3/Symbolic/Analyzers/FineGrainedStructuralPopulationDiversityAnalyzer.cs

Timestamp:

12/05/10 22:10:49 (14 years ago)

Author:

swinkler

Message:

Worked on population diversity for GP (#1278): Moved GeneticInformationItem class into separate file.

File:

• branches/DataAnalysis.PopulationDiversityAnalysis/HeuristicLab.Problems.DataAnalysis.Regression/3.3/Symbolic/Analyzers/FineGrainedStructuralPopulationDiversityAnalyzer.cs (modified) (3 diffs)

branches/DataAnalysis.PopulationDiversityAnalysis/HeuristicLab.Problems.DataAnalysis.Regression/3.3/Symbolic/Analyzers/FineGrainedStructuralPopulationDiversityAnalyzer.cs

@@ -43 +43 @@
   [StorableClass]
   public sealed class FineGrainedStructuralPopulationDiversityAnalyzer : SymbolicRegressionPopulationDiversityAnalyzer {
-
-    // properties: min level delts, max level delta, etc.
 
     #region Properties and Parameters
@@ -192 +190 @@
       IDictionary<string, IList<GeneticInformationItem>>[] geneticInformationItemsListsDictionaries = new IDictionary<string, IList<GeneticInformationItem>>[n];
       for (int i = 0; i < n; i++) {
-        geneticInformationItemsLists[i] = GeneticInformationItem.getGeneticInformationItems(solutions[i].Root, variableNames, MinimumLevelDelta, MaximumLevelDelta);
+        geneticInformationItemsLists[i] = GeneticInformationItem.GetGeneticInformationItems(solutions[i].Root, variableNames, MinimumLevelDelta, MaximumLevelDelta);
         geneticInformationItemsListsDictionaries[i] = GeneticInformationItem.GetDictionary(geneticInformationItemsLists[i]);
       }
@@ -233 +231 @@
     }
 
-    #region private class GeneticInformationItem
-
-    private class GeneticInformationItem {
-
-      private Type myAncestorDefinition;
-      public Type AncestorDefinition {
-        get { return myAncestorDefinition; }
-      }
-
-      private int myAncestorIndex;
-      public int AncestorIndex {
-        get { return myAncestorIndex; }
-      }
-
-      private Type myDescendantDefinition;
-      public Type DescendantDefinition {
-        get { return myDescendantDefinition; }
-      }
-
-      private int myAncestorLevel;
-      public int AncestorLevel {
-        get { return myAncestorLevel; }
-      }
-
-      private double myDescendantCoefficient;
-      public double DescendantCoefficient {
-        get { return myDescendantCoefficient; }
-      }
-
-      private double myDescendantVariableIndex;
-      public double DescendantVariableIndex {
-        get { return myDescendantVariableIndex; }
-      }
-
-      private int myDescendantTimeOffset;
-      public int DescendantTimeOffset {
-        get { return myDescendantTimeOffset; }
-      }
-
-      private SymbolicExpressionTreeNode myDescendantTreeNode;
-      public SymbolicExpressionTreeNode DescendantTreeNode {
-        get { return myDescendantTreeNode; }
-      }
-
-      private int myDescendantLevel;
-      public int DescendantLevel {
-        get { return myDescendantLevel; }
-      }
-
-      public int LevelDelta {
-        get { return (myDescendantLevel - myAncestorLevel); }
-      }
-
-      public static IList<GeneticInformationItem> CopyList(IList<GeneticInformationItem> GeneticInformationItemsList) {
-        List<GeneticInformationItem> list = new List<GeneticInformationItem>(GeneticInformationItemsList.Count);
-        list.AddRange(GeneticInformationItemsList);
-        return list;
-      }
-
-      public static string GetKey(GeneticInformationItem item) {
-        return item.AncestorDefinition.Name.ToString() + "," + item.DescendantDefinition.Name.ToString();
-      }
-
-      public static IDictionary<string, IList<GeneticInformationItem>> GetDictionary(IList<GeneticInformationItem> GeneticInformationItemsList) {
-        IDictionary<string, IList<GeneticInformationItem>> dictionary = new Dictionary<string, IList<GeneticInformationItem>>();
-        foreach (GeneticInformationItem item in GeneticInformationItemsList) {
-          string key = GetKey(item);
-          if (!dictionary.ContainsKey(key))
-            dictionary.Add(key, new List<GeneticInformationItem>());
-          dictionary[key].Add(item);
-        }
-        return dictionary;
-      }
-
-      public static IDictionary<string, IList<GeneticInformationItem>> CopyDictionary(IDictionary<string, IList<GeneticInformationItem>> Dictionary) {
-        IDictionary<string, IList<GeneticInformationItem>> copy = new Dictionary<string, IList<GeneticInformationItem>>();
-        foreach (KeyValuePair<string, IList<GeneticInformationItem>> pair in Dictionary) {
-          copy.Add(pair.Key, CopyList(pair.Value));
-        }
-        return copy;
-      }
-
-      public static GeneticInformationItem FindBestPendant(GeneticInformationItem Item, IList<GeneticInformationItem> ComparisonItems,
-          double ConstantMinimum, double ConstantMaximum, double VariableWeightSigma,
-          int MaximumTreeHeight, int MinimumTimeOffset, int MaximumTimeOffset,
-          double LevelDifferenceCoefficient, double AncestorIndexCoefficient,
-          double ConstantValueCoefficient, double VariableWeightCoefficient, double TimeOffsetCoefficient, double VariableIndexCoefficient,
-          bool AdditiveSimilarityCalculation,
-          out double BestPendantSimilarity) {
-        int maxSimilarityIndex = -1;
-        double similarity, maxSimilarity = -double.MaxValue;
-        for (int i = 0; i < ComparisonItems.Count; i++) {
-          similarity = Similarity(Item, ComparisonItems[i], ConstantMinimum, ConstantMaximum, VariableWeightSigma, MaximumTreeHeight, MinimumTimeOffset, MaximumTimeOffset,
-            LevelDifferenceCoefficient, AncestorIndexCoefficient, ConstantValueCoefficient, VariableWeightSigma, TimeOffsetCoefficient,
-            VariableWeightCoefficient, AdditiveSimilarityCalculation);
-          if (!double.IsNaN(similarity) && similarity > maxSimilarity) {
-            maxSimilarity = similarity;
-            maxSimilarityIndex = i;
-          }
-        }
-        BestPendantSimilarity = maxSimilarity;
-        if (maxSimilarityIndex >= 0)
-          return ComparisonItems[maxSimilarityIndex];
-        else
-          return null;
-      }
-
-      public static double Similarity(GeneticInformationItem Item1, GeneticInformationItem Item2,
-          double ConstantMinimum, double ConstantMaximum, double VariableWeightSigma,
-          int MaximumTreeHeight, int MinimumTimeOffset, int MaximumTimeOffset,
-          double LevelDifferenceCoefficient, double AncestorIndexCoefficient,
-          double ConstantValueCoefficient, double VariableWeightCoefficient, double TimeOffsetCoefficient, double VariableIndexCoefficient,
-          bool AdditiveSimilarityCalculation) {
-
-        if (Item1.AncestorDefinition != Item2.AncestorDefinition || Item1.DescendantDefinition != Item2.DescendantDefinition)
-          return double.NaN;
-
-        // the two items for sure have the same behavior definitions
-
-        #region initialize punishments
-
-        double punishmentContributionSum = 0;
-        double punishmentCoefficientsProduct = 1;
-
-        double ancestorIndexDifferencePunishment = 0;
-        double levelDifferencePunishment = 0;
-
-        double descendantConstantValueDifferencePunishment = 0;
-        double descendantVariableWeightDifferencePunishment = 0;
-        double descendantTimeOffsetDifferencePunishment = 0;
-        double descendantVariableIndexDifferencePunishment = 0;
-
-        #endregion
-
-        if (LevelDifferenceCoefficient > 0) {
-          levelDifferencePunishment = Item1.LevelDelta - Item2.LevelDelta;
-          if (levelDifferencePunishment < 0)
-            levelDifferencePunishment = -levelDifferencePunishment;
-          levelDifferencePunishment /= MaximumTreeHeight;
-          if (levelDifferencePunishment > 1)
-            levelDifferencePunishment = 1;
-          levelDifferencePunishment *= LevelDifferenceCoefficient;
-          punishmentContributionSum += LevelDifferenceCoefficient;
-          punishmentCoefficientsProduct *= (1 - LevelDifferenceCoefficient);
-        }
-        if (AncestorIndexCoefficient > 0) {
-          if (Item1.AncestorIndex != Item2.AncestorIndex)
-            ancestorIndexDifferencePunishment = 1;
-          else
-            ancestorIndexDifferencePunishment = 0;
-          ancestorIndexDifferencePunishment *= AncestorIndexCoefficient;
-          punishmentContributionSum += AncestorIndexCoefficient;
-          punishmentCoefficientsProduct *= (1 - AncestorIndexCoefficient);
-        }
-
-        if (Item1.DescendantTreeNode is ConstantTreeNode) {
-          if (ConstantValueCoefficient > 0) {
-            double constantValueCoefficientDifference = Math.Abs(Item1.DescendantCoefficient - Item2.DescendantCoefficient);
-            // assume uniform distribution within given minimum and maximum
-            descendantConstantValueDifferencePunishment = (constantValueCoefficientDifference / (ConstantMaximum - ConstantMinimum));
-            if (descendantConstantValueDifferencePunishment > 1)
-              descendantConstantValueDifferencePunishment = 1;
-            descendantConstantValueDifferencePunishment *= ConstantValueCoefficient;
-            punishmentContributionSum += ConstantValueCoefficient;
-            punishmentCoefficientsProduct *= (1 - ConstantValueCoefficient);
-          }
-        }
-        if(Item1.DescendantTreeNode is VariableTreeNode) {
-          if (VariableWeightCoefficient > 0) {
-            double variableWeightDifference = Math.Abs(Item1.DescendantCoefficient - Item2.DescendantCoefficient);
-            // assume normal distribution within given sigma
-            descendantVariableWeightDifferencePunishment = variableWeightDifference / (VariableWeightSigma * 4);
-            if (descendantVariableWeightDifferencePunishment > 1)
-              descendantVariableWeightDifferencePunishment = 1;
-            descendantVariableWeightDifferencePunishment *= VariableWeightCoefficient;
-            punishmentContributionSum += VariableWeightCoefficient;
-            punishmentCoefficientsProduct *= (1 - VariableWeightCoefficient);
-          }
-          if (TimeOffsetCoefficient > 0) {
-            double timeOffsetDifference = Math.Abs(Item1.DescendantTimeOffset - Item2.DescendantTimeOffset);
-            if (MaximumTimeOffset > 0)
-              descendantTimeOffsetDifferencePunishment = timeOffsetDifference / (MaximumTimeOffset - MinimumTimeOffset);
-            descendantTimeOffsetDifferencePunishment *= TimeOffsetCoefficient;
-            punishmentContributionSum += TimeOffsetCoefficient;
-            punishmentCoefficientsProduct *= (1 - TimeOffsetCoefficient);
-          }
-          if (VariableIndexCoefficient > 0) {
-            if (Item1.DescendantVariableIndex != Item2.DescendantVariableIndex)
-              descendantVariableIndexDifferencePunishment = 1;
-            else
-              descendantVariableIndexDifferencePunishment = 0;
-            descendantVariableIndexDifferencePunishment *= VariableIndexCoefficient;
-            punishmentContributionSum += VariableIndexCoefficient;
-            punishmentCoefficientsProduct *= (1 - VariableIndexCoefficient);
-          }
-        }
-
-        double result;
-
-        if (AdditiveSimilarityCalculation) {
-          double punishmentsSum = ancestorIndexDifferencePunishment + levelDifferencePunishment +
-            descendantConstantValueDifferencePunishment + descendantVariableWeightDifferencePunishment +
-            descendantTimeOffsetDifferencePunishment + descendantVariableIndexDifferencePunishment;
-          result = (1 - punishmentsSum / punishmentContributionSum);
-        } else {
-          result =
-            (1 - ancestorIndexDifferencePunishment) *
-            (1 - levelDifferencePunishment) *
-            (1 - descendantConstantValueDifferencePunishment) *
-            (1 - descendantVariableWeightDifferencePunishment) *
-            (1 - descendantTimeOffsetDifferencePunishment) *
-            (1 - descendantVariableIndexDifferencePunishment);
-          // worst possible result is (1-punishmentCoefficientsProduct), so scale linearly to [0;1]:
-          result = (result - punishmentCoefficientsProduct) / (1 - punishmentCoefficientsProduct);
-        }
-
-        if (result < 0 || result > 1)
-          throw new InvalidOperationException("ERROR in GeneticInformationItem.Similarity: An invalid similarity value (" + result.ToString() + ") has been calculated.");
-
-        return result;
-
-      }
-
-      public static IList<GeneticInformationItem> getGeneticInformationItems(SymbolicExpressionTreeNode node, List<string> variableNames,
-          int MinimumLevelDelta, int MaximumLevelDelta) {
-        // first we have to collect all items, then we filter; it is not possible to filter while collecting because the items are
-        // collected recursively and used for collecting the parents' items.
-        if (MinimumLevelDelta > MaximumLevelDelta)
-          return new List<GeneticInformationItem>();
-        IList<GeneticInformationItem> list = getGeneticInformationItems(node, variableNames, 0);
-        List<GeneticInformationItem> resultList = new List<GeneticInformationItem>();
-        foreach (GeneticInformationItem item in list)
-          if (item.LevelDelta >= MinimumLevelDelta && item.LevelDelta <= MaximumLevelDelta)
-            resultList.Add(item);
-        return resultList;
-      }
-
-      private static IList<GeneticInformationItem> getGeneticInformationItems(SymbolicExpressionTreeNode node, List<string> variableNames, int level) {
-        // Idea: collect all descendants' lists and then add new items using the retrieved ones.
-        // This should save lots of time and reduce complexity of the items retrieval process.
-        // Program roots are not considered, neither are start symbol nodes
-        if (node.Symbol is ProgramRootSymbol)
-          return getGeneticInformationItems(node.SubTrees[0], variableNames, level + 1);
-        List<GeneticInformationItem> list = new List<GeneticInformationItem>();
-        // add item for this node:
-        if (!(node.Symbol is StartSymbol)) {
-          list.Add(new GeneticInformationItem(node, variableNames, level));
-        }
-        // add items for the descendants, but prevent multiple references to descendant nodes:
-        List<SymbolicExpressionTreeNode> descendantNodes = new List<SymbolicExpressionTreeNode>();
-        for (int i = 0; i < node.SubTrees.Count; i++) {
-          IList<GeneticInformationItem> descendantItems = getGeneticInformationItems(node.SubTrees[i], variableNames, level + 1);
-          list.AddRange(descendantItems);
-          if (!(node.Symbol is StartSymbol))
-            foreach (GeneticInformationItem item in descendantItems) {
-              if (!descendantNodes.Contains(item.DescendantTreeNode)) {
-                list.Add(new GeneticInformationItem(node, item, i, level));
-                descendantNodes.Add(item.DescendantTreeNode);
-              }
-            }
-        }
-        return list;
-      }
-
-      private GeneticInformationItem (SymbolicExpressionTreeNode node, List<string> variableNames, int level) {
-        myAncestorIndex = -1;
-        VariableTreeNode variableTreeNode = node as VariableTreeNode;
-        LaggedVariableTreeNode laggedVariableTreeNode = node as LaggedVariableTreeNode;
-        ConstantTreeNode constantTreeNode = node as ConstantTreeNode;
-        myAncestorDefinition = node.Symbol.GetType();
-        myDescendantDefinition = myAncestorDefinition;
-        if (variableTreeNode != null)
-          myDescendantCoefficient = variableTreeNode.Weight;
-        else if (constantTreeNode != null)
-          myDescendantCoefficient = constantTreeNode.Value;
-        else
-          myDescendantCoefficient = double.NaN;
-        if (laggedVariableTreeNode != null)
-          myDescendantTimeOffset = laggedVariableTreeNode.Lag;
-        else
-          myDescendantTimeOffset = 0;
-        if (variableTreeNode != null)
-          myDescendantVariableIndex = variableNames.IndexOf(variableTreeNode.VariableName);
-        else
-          myDescendantVariableIndex = -1;
-        myAncestorLevel = level;
-        myDescendantLevel = level;
-        myDescendantTreeNode = node;
-      }
-
-      private GeneticInformationItem(SymbolicExpressionTreeNode parentNode, GeneticInformationItem descendantGeneticInformationItem,
-          int ancestorIndex, int parentNodeLevel) {
-        myAncestorIndex = ancestorIndex;
-        myAncestorLevel = parentNodeLevel;
-        myAncestorDefinition = parentNode.Symbol.GetType();
-        myDescendantCoefficient = descendantGeneticInformationItem.DescendantCoefficient;
-        myDescendantDefinition = descendantGeneticInformationItem.DescendantDefinition;
-        myDescendantTimeOffset = descendantGeneticInformationItem.DescendantTimeOffset;
-        myDescendantVariableIndex = descendantGeneticInformationItem.DescendantVariableIndex;
-        myDescendantLevel = descendantGeneticInformationItem.DescendantLevel;
-        myDescendantTreeNode = descendantGeneticInformationItem.DescendantTreeNode;
-      }
-
-      public override string ToString() {
-        return "ancestor: " + AncestorDefinition.Name.ToString() + ", [" + AncestorIndex + "]; "
-          + "descendant: " + DescendantDefinition.Name.ToString() + " (varIndex " + DescendantVariableIndex + ", "
-          + DescendantCoefficient + ", t-" + DescendantTimeOffset + ");"
-          + " level delta = " + DescendantLevel + "-" + AncestorLevel + " = " + LevelDelta;
-      }
-
-    }
-
-    #endregion
-
   }
+
 }