Free cookie consent management tool by TermsFeed Policy Generator

Changeset 7105


Ignore:
Timestamp:
11/30/11 17:21:30 (13 years ago)
Author:
bburlacu
Message:

#1682: Implemented the ProbabilisticFunctionalCrossover.

File:
1 edited

Legend:

Unmodified
Added
Removed

  • branches/gp-crossover/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Crossovers/SymbolicDataAnalysisExpressionProbabilisticFunctionalCrossover.cs

    r7089 r7105  
    5959    public static ISymbolicExpressionTree Cross(IRandom random, ISymbolicExpressionTree parent0, ISymbolicExpressionTree parent1,
    6060                                                ISymbolicDataAnalysisExpressionTreeInterpreter interpreter, T problemData, IEnumerable<int> rows, int maxDepth, int maxLength) {
    61       List<CutPoint> crossoverPoints0 = new List<CutPoint>();
     61      var crossoverPoints0 = new List<CutPoint>();
    6262      parent0.Root.ForEachNodePostfix((n) => {
    6363        if (n.Subtrees.Any() && n != parent0.Root)
    … …  
    6565            crossoverPoints0.Add(new CutPoint(n, child));
    6666      });
    67       CutPoint crossoverPoint0 = crossoverPoints0[random.Next(crossoverPoints0.Count)];
     67      var crossoverPoint0 = crossoverPoints0[random.Next(crossoverPoints0.Count)];
    6868      int level = parent0.Root.GetBranchLevel(crossoverPoint0.Child);
    6969      int length = parent0.Root.GetLength() - crossoverPoint0.Child.GetLength();
    7070
    71       List<ISymbolicExpressionTreeNode> allowedBranches = new List<ISymbolicExpressionTreeNode>();
     71      var allowedBranches = new List<ISymbolicExpressionTreeNode>();
    7272      parent1.Root.ForEachNodePostfix((n) => {
    7373        if (n.Subtrees.Any() && n != parent1.Root)
    … …  
    8888      var rootSymbol = new ProgramRootSymbol();
    8989      var startSymbol = new StartSymbol();
    90       var tree0 = CreateTreeFromNode(random, crossoverPoint0.Child, rootSymbol, startSymbol);
     90      var tree0 = CreateTreeFromNode(random, crossoverPoint0.Child, rootSymbol, startSymbol); // this will change crossoverPoint0.Child.Parent
    9191      IEnumerable<double> estimatedValues0 = interpreter.GetSymbolicExpressionTreeValues(tree0, dataset, rows);
    9292      double min0 = estimatedValues0.Min();
    9393      double max0 = estimatedValues0.Max();
     94      crossoverPoint0.Child.Parent = crossoverPoint0.Parent; // restore correct parent
    9495
    95       List<double> weights = new List<double>();
     96      var weights = new List<double>();
    9697      foreach (var node in allowedBranches) {
     98        var parent = node.Parent;
    9799        var tree1 = CreateTreeFromNode(random, node, rootSymbol, startSymbol);
    98100        IEnumerable<double> estimatedValues1 = interpreter.GetSymbolicExpressionTreeValues(tree1, dataset, rows);
    99101        double min1 = estimatedValues1.Min();
    100102        double max1 = estimatedValues1.Max();
    101 
    102         double behavioralDistance = (Math.Abs(min0 - min1) + Math.Abs(max0 - max1)) / 2;
    103 
     103        double behavioralDistance = (Math.Abs(min0 - min1) + Math.Abs(max0 - max1)) / 2; // this can be NaN of Infinity because some trees are crazy like exp(exp(exp(...))), we correct that below
    104104        weights.Add(behavioralDistance);
     105        node.Parent = parent; // restore correct node parent
    105106      }
    106107
    107       ISymbolicExpressionTreeNode selectedBranch = SelectRandomBranch(random, allowedBranches, weights);
     108      // remove branches with an infinite or NaN behavioral distance
     109      for (int i = 0; i != weights.Count; ++i)
     110        if (Double.IsNaN(weights[i]) || Double.IsInfinity(weights[i])) {
     111          weights.RemoveAt(i);
     112          allowedBranches.RemoveAt(i);
     113        }
     114
     115      // check if there are any allowed branches left
     116      if (allowedBranches.Count == 0)
     117        return parent0;
     118
     119      ISymbolicExpressionTreeNode selectedBranch;
     120      double sum = weights.Sum();
     121      if (sum == 0.0)
     122        selectedBranch = allowedBranches[0]; // just return the first, since we don't care (all weights are zero)
     123      else {
     124        // transform similarity distances into probabilities by normalizing and inverting the values
     125        for (int i = 0; i != weights.Count; ++i)
     126          weights[i] = (1 - weights[i] / sum);
     127
     128        selectedBranch = SelectRandomBranch(random, allowedBranches, weights);
     129      }
    108130      swap(crossoverPoint0, selectedBranch);
    109 
    110131      return parent0;
    111132    }
    … …  
    128149
    129150    private static ISymbolicExpressionTreeNode SelectRandomBranch(IRandom random, List<ISymbolicExpressionTreeNode> nodes, List<double> weights) {
    130       // transform similarity distances into probabilities by normalizing and inverting the values
    131       double sum = weights.Sum();
    132       for (int i = 0; i != weights.Count; ++i)
    133         weights[i] = (1 - weights[i] / sum);
    134151      double r = weights.Sum() * random.NextDouble();
    135152      for (int i = 0; i != nodes.Count; ++i) {
    … …  
    138155        r -= weights[i];
    139156      }
    140       return null;
     157      return nodes.Last();
    141158    }
    142159  }
Note: See TracChangeset for help on using the changeset viewer.