decisiontree.hpp

   1   //
   2   //  decisiontree.hpp
   3   //  rrf-fs-prototype
   4   //
   5   //  Created by Abu Zaher Faridee on 5/28/12.
   6   //  Copyright (c) 2012 Schloss Lab. All rights reserved.
   7   //
   8
   9 #ifndef RF_DECISIONTREE_HPP
  10 #define RF_DECISIONTREE_HPP
  11
  12 #include "macros.h"
  13 #include "rftreenode.hpp"
  14 #include "abstractdecisiontree.hpp"
  15
  16 /***********************************************************************/
  17
  18 struct VariableRankDescendingSorter {
  19   bool operator() (const pair<int, int>& firstPair, const pair<int, int>& secondPair){
  20       return firstPair.second > secondPair.second;
  21   }
  22 };
  23 struct VariableRankDescendingSorterDouble {
  24     bool operator() (const pair<int, double>& firstPair, const pair<int, double>& secondPair){
  25         return firstPair.second > secondPair.second;
  26     }
  27 };
  28 /***********************************************************************/
  29
  30 class DecisionTree: public AbstractDecisionTree{
  31
  32     friend class RandomForest;
  33
  34 public:
  35
  36     DecisionTree(vector< vector<int> >& baseDataSet,
  37                  vector<int> globalDiscardedFeatureIndices,
  38                  OptimumFeatureSubsetSelector optimumFeatureSubsetSelector,
  39                  string treeSplitCriterion,
  40                  float featureStandardDeviationThreshold);
  41
  42     virtual ~DecisionTree(){ deleteTreeNodesRecursively(rootNode); }
  43
  44     int calcTreeVariableImportanceAndError(int& numCorrect, double& treeErrorRate);
  45     int evaluateSample(vector<int> testSample);
  46     int calcTreeErrorRate(int& numCorrect, double& treeErrorRate);
  47
  48     void randomlyShuffleAttribute(const vector< vector<int> >& samples,
  49                                   const int featureIndex,
  50                                   const int prevFeatureIndex,
  51                                   vector< vector<int> >& shuffledSample);
  52
  53     void purgeDataSetsFromTree() { purgeTreeNodesDataRecursively(rootNode); }
  54     int purgeTreeNodesDataRecursively(RFTreeNode* treeNode);
  55
  56     void pruneTree(double pruneAggressiveness);
  57     void pruneRecursively(RFTreeNode* treeNode, double pruneAggressiveness);
  58     void updateMisclassificationCountRecursively(RFTreeNode* treeNode, vector<int> testSample);
  59     void updateOutputClassOfNode(RFTreeNode* treeNode);
  60
  61
  62 private:
  63
  64     void buildDecisionTree();
  65     int splitRecursively(RFTreeNode* rootNode);
  66     int findAndUpdateBestFeatureToSplitOn(RFTreeNode* node);
  67     vector<int> selectFeatureSubsetRandomly(vector<int> globalDiscardedFeatureIndices, vector<int> localDiscardedFeatureIndices);
  68     int printTree(RFTreeNode* treeNode, string caption);
  69     void deleteTreeNodesRecursively(RFTreeNode* treeNode);
  70
  71     vector<int> variableImportanceList;
  72     map<int, int> outOfBagEstimates;
  73
  74     float featureStandardDeviationThreshold;
  75 };
  76
  77 #endif