randomforest.cpp

   1 //
   2 //  randomforest.cpp
   3 //  Mothur
   4 //
   5 //  Created by Sarah Westcott on 10/2/12.
   6 //  Copyright (c) 2012 Schloss Lab. All rights reserved.
   7 //
   8
   9 #include "randomforest.hpp"
  10
  11 /***********************************************************************/
  12
  13 RandomForest::RandomForest(const vector <vector<int> > dataSet,const int numDecisionTrees,
  14              const string treeSplitCriterion = "informationGain") : AbstractRandomForest(dataSet, numDecisionTrees, treeSplitCriterion) {
  15     m = MothurOut::getInstance();
  16 }
  17
  18 /***********************************************************************/
  19 // DONE
  20 int RandomForest::calcForrestErrorRate() {
  21     try {
  22         int numCorrect = 0;
  23         for (map<int, vector<int> >::iterator it = globalOutOfBagEstimates.begin(); it != globalOutOfBagEstimates.end(); it++) {
  24
  25             if (m->control_pressed) { return 0; }
  26
  27             int indexOfSample = it->first;
  28             vector<int> predictedOutComes = it->second;
  29             vector<int>::iterator maxPredictedOutComeIterator = max_element(predictedOutComes.begin(), predictedOutComes.end());
  30             int majorityVotedOutcome = (int)(maxPredictedOutComeIterator - predictedOutComes.begin());
  31             int realOutcome = dataSet[indexOfSample][numFeatures];
  32
  33             if (majorityVotedOutcome == realOutcome) { numCorrect++; }
  34         }
  35
  36         // TODO: save or return forrestErrorRate for future use;
  37         double forrestErrorRate = 1 - ((double)numCorrect / (double)globalOutOfBagEstimates.size());
  38
  39         m->mothurOut("numCorrect = " + toString(numCorrect)+ "\n");
  40         m->mothurOut("forrestErrorRate = " + toString(forrestErrorRate)+ "\n");
  41
  42         return 0;
  43     }
  44         catch(exception& e) {
  45                 m->errorOut(e, "RandomForest", "calcForrestErrorRate");
  46                 exit(1);
  47         }
  48 }
  49
  50 /***********************************************************************/
  51 // DONE
  52 int RandomForest::calcForrestVariableImportance(string filename) {
  53     try {
  54
  55     // TODO: need to add try/catch operators to fix this
  56     // follow the link: http://en.wikipedia.org/wiki/Dynamic_cast
  57         //if you are going to dynamically cast, aren't you undoing the advantage of abstraction. Why abstract at all?
  58         //could cause maintenance issues later if other types of Abstract decison trees are created that cannot be cast as a decision tree.
  59     for (int i = 0; i < decisionTrees.size(); i++) {
  60         if (m->control_pressed) { return 0; }
  61         DecisionTree* decisionTree = dynamic_cast<DecisionTree*>(decisionTrees[i]);
  62
  63         for (int j = 0; j < numFeatures; j++) {
  64             globalVariableImportanceList[j] += (double)decisionTree->variableImportanceList[j];
  65         }
  66     }
  67
  68     for (int i = 0;  i < numFeatures; i++) {
  69         cout << "[" << i << ',' << globalVariableImportanceList[i] << "], ";
  70         globalVariableImportanceList[i] /= (double)numDecisionTrees;
  71     }
  72
  73     vector< vector<double> > globalVariableRanks;
  74     for (int i = 0; i < globalVariableImportanceList.size(); i++) {
  75         if (globalVariableImportanceList[i] > 0) {
  76             vector<double> globalVariableRank(2, 0);
  77             globalVariableRank[0] = i; globalVariableRank[1] = globalVariableImportanceList[i];
  78             globalVariableRanks.push_back(globalVariableRank);
  79         }
  80     }
  81
  82     VariableRankDescendingSorterDouble variableRankDescendingSorter;
  83     sort(globalVariableRanks.begin(), globalVariableRanks.end(), variableRankDescendingSorter);
  84         ofstream out;
  85         m->openOutputFile(filename, out);
  86         out <<"OTU\tRank\n";
  87         for (int i = 0; i < globalVariableRanks.size(); i++) {
  88             out << m->currentBinLabels[(int)globalVariableRanks[i][0]] << '\t' << globalVariableImportanceList[globalVariableRanks[i][0]] << endl;
  89         }
  90         out.close();
  91         return 0;
  92     }
  93         catch(exception& e) {
  94                 m->errorOut(e, "RandomForest", "calcForrestVariableImportance");
  95                 exit(1);
  96         }
  97 }
  98 /***********************************************************************/
  99 // DONE
 100 int RandomForest::populateDecisionTrees() {
 101     try {
 102
 103         for (int i = 0; i < numDecisionTrees; i++) {
 104             if (m->control_pressed) { return 0; }
 105             if (((i+1) % 10) == 0) {  m->mothurOut("Creating " + toString(i+1) + " (th) Decision tree\n");  }
 106             // TODO: need to first fix if we are going to use pointer based system or anything else
 107             DecisionTree* decisionTree = new DecisionTree(dataSet, globalDiscardedFeatureIndices, OptimumFeatureSubsetSelector("log2"), treeSplitCriterion);
 108             decisionTree->calcTreeVariableImportanceAndError();
 109             if (m->control_pressed) { return 0; }
 110             updateGlobalOutOfBagEstimates(decisionTree);
 111             if (m->control_pressed) { return 0; }
 112             decisionTree->purgeDataSetsFromTree();
 113             if (m->control_pressed) { return 0; }
 114             decisionTrees.push_back(decisionTree);
 115         }
 116
 117         if (m->debug) {
 118             // m->mothurOut("globalOutOfBagEstimates = " + toStringVectorMap(globalOutOfBagEstimates)+ "\n");
 119         }
 120
 121         return 0;
 122     }
 123     catch(exception& e) {
 124         m->errorOut(e, "RandomForest", "populateDecisionTrees");
 125         exit(1);
 126     }
 127 }
 128 /***********************************************************************/
 129 // TODO: need to finalize bettween reference and pointer for DecisionTree [partially solved]
 130 // TODO: make this pure virtual in superclass
 131 // DONE
 132 int RandomForest::updateGlobalOutOfBagEstimates(DecisionTree* decisionTree) {
 133     try {
 134         for (map<int, int>::iterator it = decisionTree->outOfBagEstimates.begin(); it != decisionTree->outOfBagEstimates.end(); it++) {
 135
 136             if (m->control_pressed) { return 0; }
 137
 138             int indexOfSample = it->first;
 139             int predictedOutcomeOfSample = it->second;
 140
 141             if (globalOutOfBagEstimates.count(indexOfSample) == 0) {
 142                 globalOutOfBagEstimates[indexOfSample] = vector<int>(decisionTree->numOutputClasses, 0);
 143             };
 144
 145             globalOutOfBagEstimates[indexOfSample][predictedOutcomeOfSample] += 1;
 146         }
 147         return 0;
 148     }
 149     catch(exception& e) {
 150         m->errorOut(e, "RandomForest", "updateGlobalOutOfBagEstimates");
 151         exit(1);
 152     }
 153 }
 154 /***********************************************************************/
 155
 156