Free cookie consent management tool by TermsFeed Policy Generator

Ignore:
Timestamp:
03/11/15 14:07:50 (9 years ago)
Author:
bburlacu
Message:

#2359: Implemented improvements

Location:
trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic.Classification/3.4
Files:
2 edited

Legend:

Unmodified
Added
Removed
  • trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic.Classification/3.4/SymbolicClassificationPruningAnalyzer.cs

    r12012 r12189  
    4343    public SymbolicClassificationPruningAnalyzer() {
    4444      Parameters.Add(new ValueParameter<SymbolicDataAnalysisSolutionImpactValuesCalculator>(ImpactValuesCalculatorParameterName, "The impact values calculator", new SymbolicClassificationSolutionImpactValuesCalculator()));
    45       Parameters.Add(new ValueParameter<SymbolicDataAnalysisExpressionPruningOperator>(PruningOperatorParameterName, "The operator used to prune trees", new SymbolicClassificationPruningOperator()));
     45      Parameters.Add(new ValueParameter<SymbolicDataAnalysisExpressionPruningOperator>(PruningOperatorParameterName, "The operator used to prune trees", new SymbolicClassificationPruningOperator(new SymbolicClassificationSolutionImpactValuesCalculator())));
    4646    }
    4747  }
  • trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic.Classification/3.4/SymbolicClassificationPruningOperator.cs

    r12012 r12189  
    2222#endregion
    2323
     24using System.Collections.Generic;
    2425using System.Linq;
    2526using HeuristicLab.Common;
    2627using HeuristicLab.Core;
     28using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
    2729using HeuristicLab.Parameters;
    2830using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
     
    3234  [Item("SymbolicClassificationPruningOperator", "An operator which prunes symbolic classificaton trees.")]
    3335  public class SymbolicClassificationPruningOperator : SymbolicDataAnalysisExpressionPruningOperator {
    34     private const string ImpactValuesCalculatorParameterName = "ImpactValuesCalculator";
    3536    private const string ModelCreatorParameterName = "ModelCreator";
    3637
     
    5253    protected SymbolicClassificationPruningOperator(bool deserializing) : base(deserializing) { }
    5354
    54     public SymbolicClassificationPruningOperator() {
    55       Parameters.Add(new ValueParameter<ISymbolicDataAnalysisSolutionImpactValuesCalculator>(ImpactValuesCalculatorParameterName, new SymbolicClassificationSolutionImpactValuesCalculator()));
     55    public SymbolicClassificationPruningOperator(ISymbolicDataAnalysisSolutionImpactValuesCalculator impactValuesCalculator)
     56      : base(impactValuesCalculator) {
    5657      Parameters.Add(new LookupParameter<ISymbolicClassificationModelCreator>(ModelCreatorParameterName));
    5758    }
    5859
    59     protected override ISymbolicDataAnalysisModel CreateModel() {
    60       var model = ModelCreatorParameter.ActualValue.CreateSymbolicClassificationModel(SymbolicExpressionTree, Interpreter, EstimationLimits.Lower, EstimationLimits.Upper);
    61       var problemData = (IClassificationProblemData)ProblemData;
    62       var rows = problemData.TrainingIndices;
    63       model.RecalculateModelParameters(problemData, rows);
     60    protected override ISymbolicDataAnalysisModel CreateModel(ISymbolicExpressionTree tree, ISymbolicDataAnalysisExpressionTreeInterpreter interpreter, IDataAnalysisProblemData problemData, DoubleLimit estimationLimits) {
     61      var model = ModelCreatorParameter.ActualValue.CreateSymbolicClassificationModel(tree, interpreter, estimationLimits.Lower, estimationLimits.Upper);
     62      var classificationProblemData = (IClassificationProblemData)problemData;
     63      var rows = classificationProblemData.TrainingIndices;
     64      model.RecalculateModelParameters(classificationProblemData, rows);
    6465      return model;
    6566    }
     
    6970      var classificationProblemData = (IClassificationProblemData)ProblemData;
    7071      var trainingIndices = Enumerable.Range(FitnessCalculationPartition.Start, FitnessCalculationPartition.Size);
    71       var estimatedValues = classificationModel.GetEstimatedClassValues(ProblemData.Dataset, trainingIndices);
    72       var targetValues = ProblemData.Dataset.GetDoubleValues(classificationProblemData.TargetVariable, trainingIndices);
     72
     73      return Evaluate(classificationModel, classificationProblemData, trainingIndices);
     74    }
     75
     76    private static double Evaluate(IClassificationModel model, IClassificationProblemData problemData, IEnumerable<int> rows) {
     77      var estimatedValues = model.GetEstimatedClassValues(problemData.Dataset, rows);
     78      var targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
    7379      OnlineCalculatorError errorState;
    7480      var quality = OnlineAccuracyCalculator.Calculate(targetValues, estimatedValues, out errorState);
     
    7682      return quality;
    7783    }
     84
     85    public static ISymbolicExpressionTree Prune(ISymbolicExpressionTree tree, ISymbolicClassificationModelCreator modelCreator,
     86      SymbolicClassificationSolutionImpactValuesCalculator impactValuesCalculator, ISymbolicDataAnalysisExpressionTreeInterpreter interpreter,
     87      IClassificationProblemData problemData, DoubleLimit estimationLimits, IEnumerable<int> rows,
     88      double nodeImpactThreshold = 0.0, bool pruneOnlyZeroImpactNodes = false) {
     89      var clonedTree = (ISymbolicExpressionTree)tree.Clone();
     90      var model = modelCreator.CreateSymbolicClassificationModel(clonedTree, interpreter, estimationLimits.Lower, estimationLimits.Upper);
     91
     92      var nodes = clonedTree.IterateNodesPrefix().ToList();
     93      double quality = Evaluate(model, problemData, rows);
     94
     95      for (int i = 0; i < nodes.Count; ++i) {
     96        var node = nodes[i];
     97        if (node is ConstantTreeNode) continue;
     98
     99        double impactValue, replacementValue;
     100        impactValuesCalculator.CalculateImpactAndReplacementValues(model, node, problemData, rows, out impactValue, out replacementValue, quality);
     101
     102        if (pruneOnlyZeroImpactNodes) {
     103          if (!impactValue.IsAlmost(0.0)) continue;
     104        } else if (nodeImpactThreshold < impactValue) {
     105          continue;
     106        }
     107
     108        var constantNode = (ConstantTreeNode)node.Grammar.GetSymbol("Constant").CreateTreeNode();
     109        constantNode.Value = replacementValue;
     110
     111        ReplaceWithConstant(node, constantNode);
     112        i += node.GetLength() - 1; // skip subtrees under the node that was folded
     113
     114        quality -= impactValue;
     115      }
     116      return model.SymbolicExpressionTree;
     117    }
    78118  }
    79119}
Note: See TracChangeset for help on using the changeset viewer.