…rest feature selection algorithm. Once this implementation is complete, we would try to integrate the algorithm to mothur's codebase

.cproject

@@ -0,0 +1,61 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<?fileVersion 4.0.0?><cproject storage_type_id="org.eclipse.cdt.core.XmlProjectDescriptionStorage">
+	<storageModule moduleId="org.eclipse.cdt.core.settings">
+		<cconfiguration id="cdt.managedbuild.toolchain.gnu.base.841520651">
+			<storageModule buildSystemId="org.eclipse.cdt.managedbuilder.core.configurationDataProvider" id="cdt.managedbuild.toolchain.gnu.base.841520651" moduleId="org.eclipse.cdt.core.settings" name="Default">
+				<externalSettings/>
+				<extensions>
+					<extension id="org.eclipse.cdt.core.ELF" point="org.eclipse.cdt.core.BinaryParser"/>
+					<extension id="org.eclipse.cdt.core.GmakeErrorParser" point="org.eclipse.cdt.core.ErrorParser"/>
+					<extension id="org.eclipse.cdt.core.CWDLocator" point="org.eclipse.cdt.core.ErrorParser"/>
+					<extension id="org.eclipse.cdt.core.GCCErrorParser" point="org.eclipse.cdt.core.ErrorParser"/>
+					<extension id="org.eclipse.cdt.core.GASErrorParser" point="org.eclipse.cdt.core.ErrorParser"/>
+					<extension id="org.eclipse.cdt.core.GLDErrorParser" point="org.eclipse.cdt.core.ErrorParser"/>
+				</extensions>
+			</storageModule>
+			<storageModule moduleId="cdtBuildSystem" version="4.0.0">
+				<configuration buildProperties="" description="" id="cdt.managedbuild.toolchain.gnu.base.841520651" name="Default" parent="org.eclipse.cdt.build.core.emptycfg">
+					<folderInfo id="cdt.managedbuild.toolchain.gnu.base.841520651.1695811624" name="/" resourcePath="">
+						<toolChain id="cdt.managedbuild.toolchain.gnu.base.1429026183" name="cdt.managedbuild.toolchain.gnu.base" superClass="cdt.managedbuild.toolchain.gnu.base">
+							<targetPlatform archList="all" binaryParser="org.eclipse.cdt.core.ELF" id="cdt.managedbuild.target.gnu.platform.base.644038726" name="Debug Platform" osList="linux,hpux,aix,qnx" superClass="cdt.managedbuild.target.gnu.platform.base"/>
+							<builder id="cdt.managedbuild.target.gnu.builder.base.172403811" incrementalBuildTarget="mothur" keepEnvironmentInBuildfile="false" managedBuildOn="false" name="Gnu Make Builder" superClass="cdt.managedbuild.target.gnu.builder.base"/>
+							<tool id="cdt.managedbuild.tool.gnu.archiver.base.1193850959" name="GCC Archiver" superClass="cdt.managedbuild.tool.gnu.archiver.base"/>
+							<tool id="cdt.managedbuild.tool.gnu.cpp.compiler.base.2138473137" name="GCC C++ Compiler" superClass="cdt.managedbuild.tool.gnu.cpp.compiler.base">
+								<inputType id="cdt.managedbuild.tool.gnu.cpp.compiler.input.1478328097" superClass="cdt.managedbuild.tool.gnu.cpp.compiler.input"/>
+							</tool>
+							<tool id="cdt.managedbuild.tool.gnu.c.compiler.base.1034448490" name="GCC C Compiler" superClass="cdt.managedbuild.tool.gnu.c.compiler.base">
+								<inputType id="cdt.managedbuild.tool.gnu.c.compiler.input.312213090" superClass="cdt.managedbuild.tool.gnu.c.compiler.input"/>
+							</tool>
+							<tool id="cdt.managedbuild.tool.gnu.c.linker.base.1726755405" name="GCC C Linker" superClass="cdt.managedbuild.tool.gnu.c.linker.base"/>
+							<tool id="cdt.managedbuild.tool.gnu.cpp.linker.base.329465033" name="GCC C++ Linker" superClass="cdt.managedbuild.tool.gnu.cpp.linker.base">
+								<inputType id="cdt.managedbuild.tool.gnu.cpp.linker.input.307557811" superClass="cdt.managedbuild.tool.gnu.cpp.linker.input">
+									<additionalInput kind="additionalinputdependency" paths="$(USER_OBJS)"/>
+									<additionalInput kind="additionalinput" paths="$(LIBS)"/>
+								</inputType>
+							</tool>
+							<tool id="cdt.managedbuild.tool.gnu.assembler.base.286576925" name="GCC Assembler" superClass="cdt.managedbuild.tool.gnu.assembler.base">
+								<inputType id="cdt.managedbuild.tool.gnu.assembler.input.587993891" superClass="cdt.managedbuild.tool.gnu.assembler.input"/>
+							</tool>
+						</toolChain>
+					</folderInfo>
+				</configuration>
+			</storageModule>
+			<storageModule moduleId="org.eclipse.cdt.core.externalSettings"/>
+		</cconfiguration>
+	</storageModule>
+	<storageModule moduleId="scannerConfiguration">
+		<autodiscovery enabled="true" problemReportingEnabled="true" selectedProfileId=""/>
+		<scannerConfigBuildInfo instanceId="0.1309272964">
+			<autodiscovery enabled="false" problemReportingEnabled="true" selectedProfileId="org.eclipse.cdt.make.core.GCCStandardMakePerFileProfile"/>
+		</scannerConfigBuildInfo>
+	</storageModule>
+	<storageModule moduleId="cdtBuildSystem" version="4.0.0">
+		<project id="mothur.null.1905430380" name="mothur"/>
+	</storageModule>
+	<storageModule moduleId="org.eclipse.cdt.core.LanguageSettingsProviders"/>
+	<storageModule moduleId="refreshScope" versionNumber="2">
+		<configuration configurationName="Default">
+			<resource resourceType="PROJECT" workspacePath="/mothur"/>
+		</configuration>
+	</storageModule>
+</cproject>

.project

@@ -0,0 +1,27 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<projectDescription>
+	<name>mothur-RF</name>
+	<comment></comment>
+	<projects>
+	</projects>
+	<buildSpec>
+		<buildCommand>
+			<name>org.eclipse.cdt.managedbuilder.core.genmakebuilder</name>
+			<triggers>clean,full,incremental,</triggers>
+			<arguments>
+			</arguments>
+		</buildCommand>
+		<buildCommand>
+			<name>org.eclipse.cdt.managedbuilder.core.ScannerConfigBuilder</name>
+			<triggers>full,incremental,</triggers>
+			<arguments>
+			</arguments>
+		</buildCommand>
+	</buildSpec>
+	<natures>
+		<nature>org.eclipse.cdt.core.cnature</nature>
+		<nature>org.eclipse.cdt.core.ccnature</nature>
+		<nature>org.eclipse.cdt.managedbuilder.core.managedBuildNature</nature>
+		<nature>org.eclipse.cdt.managedbuilder.core.ScannerConfigNature</nature>
+	</natures>
+</projectDescription>

.settings/language.settings.xml

@@ -0,0 +1,11 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<project>
+	<configuration id="cdt.managedbuild.toolchain.gnu.base.841520651" name="Default">
+		<extension point="org.eclipse.cdt.core.LanguageSettingsProvider">
+			<provider copy-of="extension" id="org.eclipse.cdt.ui.UserLanguageSettingsProvider"/>
+			<provider class="org.eclipse.cdt.managedbuilder.language.settings.providers.GCCBuildCommandParser" id="org.eclipse.cdt.managedbuilder.core.GCCBuildCommandParser" keep-relative-paths="false" name="CDT GCC Build Output Parser" parameter="(gcc)|([gc]\+\+)|(clang)" prefer-non-shared="true"/>
+			<provider-reference id="org.eclipse.cdt.managedbuilder.core.GCCBuiltinSpecsDetector" ref="shared-provider"/>
+			<provider-reference id="org.eclipse.cdt.managedbuilder.core.MBSLanguageSettingsProvider" ref="shared-provider"/>
+		</extension>
+	</configuration>
+</project>

abstractrandomforest.cpp

@@ -1,59 +0,0 @@
-//
-//  abstractrandomforest.cpp
-//  Mothur
-//
-//  Created by Sarah Westcott on 10/1/12.
-//  Copyright (c) 2012 Schloss Lab. All rights reserved.
-//
-
-#include "abstractrandomforest.hpp"
-
-/***********************************************************************/
-AbstractRandomForest::AbstractRandomForest(const std::vector < std::vector<int> > dataSet, 
-                     const int numDecisionTrees, 
-                     const string treeSplitCriterion = "informationGain")
-: dataSet(dataSet), 
-numDecisionTrees(numDecisionTrees),
-numSamples((int)dataSet.size()),
-numFeatures((int)(dataSet[0].size() - 1)),
-globalDiscardedFeatureIndices(getGlobalDiscardedFeatureIndices()),
-globalVariableImportanceList(numFeatures, 0),
-treeSplitCriterion(treeSplitCriterion) {
-    m = MothurOut::getInstance();
-    // TODO: double check if the implemenatation of 'globalOutOfBagEstimates' is correct
-}
-
-/***********************************************************************/
-
-vector<int> AbstractRandomForest::getGlobalDiscardedFeatureIndices() {
-    try {
-        vector<int> globalDiscardedFeatureIndices;
-        
-        // calculate feature vectors
-        vector< vector<int> > featureVectors(numFeatures, vector<int>(numSamples, 0));
-        for (int i = 0; i < numSamples; i++) {
-            if (m->control_pressed) { return globalDiscardedFeatureIndices; }
-            for (int j = 0; j < numFeatures; j++) { featureVectors[j][i] = dataSet[i][j]; }
-        }
-        
-        for (int i = 0; i < featureVectors.size(); i++) {
-            if (m->control_pressed) { return globalDiscardedFeatureIndices; }
-            double standardDeviation = m->getStandardDeviation(featureVectors[i]);
-            if (standardDeviation <= 0){ globalDiscardedFeatureIndices.push_back(i); }
-        }
-        
-        if (m->debug) {
-            m->mothurOut("number of global discarded features:  " + toString(globalDiscardedFeatureIndices.size())+ "\n");
-            m->mothurOut("total features: " + toString(featureVectors.size())+ "\n");
-        }
-        
-        return globalDiscardedFeatureIndices;
-    }
-	catch(exception& e) {
-		m->errorOut(e, "AbstractRandomForest", "getGlobalDiscardedFeatureIndices");
-		exit(1);
-	} 
-}
-
-/***********************************************************************/
-

abstractrandomforest.hpp

@@ -1,67 +0,0 @@
-//
-//  abstractrandomforest.hpp
-//  rrf-fs-prototype
-//
-//  Created by Abu Zaher Faridee on 7/20/12.
-//  Copyright (c) 2012 Schloss Lab. All rights reserved.
-//
-
-#ifndef rrf_fs_prototype_abstractrandomforest_hpp
-#define rrf_fs_prototype_abstractrandomforest_hpp
-
-#include "mothurout.h"
-#include "macros.h"
-#include "abstractdecisiontree.hpp"
-
-#define DEBUG_MODE
-
-/***********************************************************************/
-
-class AbstractRandomForest{
-public:
-    // intialization with vectors
-    AbstractRandomForest(const std::vector < std::vector<int> > dataSet, 
-                       const int numDecisionTrees, 
-                       const string);
-    virtual ~AbstractRandomForest(){ }
-    virtual int populateDecisionTrees() = 0;
-    virtual int calcForrestErrorRate() = 0;
-    virtual int calcForrestVariableImportance(string) = 0;
- 
-/***********************************************************************/
-  
-protected:
-  
-    // TODO: create a better way of discarding feature
-    // currently we just set FEATURE_DISCARD_SD_THRESHOLD to 0 to solved this
-    // it can be tuned for better selection
-    // also, there might be other factors like Mean or other stuffs
-    // same would apply for createLocalDiscardedFeatureList in the TreeNode class
-  
-    // TODO: Another idea is getting an aggregated discarded feature indices after the run, from combining
-    // the local discarded feature indices
-    // this would penalize a feature, even if in global space the feature looks quite good
-    // the penalization would be averaged, so this woould unlikely to create a local optmina
-    
-    vector<int> getGlobalDiscardedFeatureIndices();
-    
-    int numDecisionTrees;
-    int numSamples;
-    int numFeatures;
-    vector< vector<int> > dataSet;
-    vector<int> globalDiscardedFeatureIndices;
-    vector<double> globalVariableImportanceList;
-    string treeSplitCriterion;
-    // This is a map of each feature to outcome count of each classes
-    // e.g. 1 => [2 7] means feature 1 has 2 outcome of 0 and 7 outcome of 1
-    map<int, vector<int> > globalOutOfBagEstimates;
-    
-    // TODO: fix this, do we use pointers?
-    vector<AbstractDecisionTree*> decisionTrees;
-    
-    MothurOut* m;
-  
-private:
-
-};
-#endif

classifysharedcommand.cpp

@@ -375,6 +375,7 @@ void ClassifySharedCommand::processSharedAndDesignData(vector<SharedRAbundVector
         
         randomForest.populateDecisionTrees();
         randomForest.calcForrestErrorRate();
+        randomForest.calcConfusionMatrix(intToTreatmentMap);
         
         map<string, string> variables; 
         variables["[filename]"] = outputDir + m->getRootName(m->getSimpleName(sharedfile));

forest.cpp

@@ -29,11 +29,13 @@ Forest::Forest(const std::vector < std::vector<int> > dataSet,
         discardHighErrorTrees(discardHighErrorTrees),
         highErrorTreeDiscardThreshold(highErrorTreeDiscardThreshold),
         optimumFeatureSubsetSelectionCriteria(optimumFeatureSubsetSelectionCriteria),
-        featureStandardDeviationThreshold(featureStandardDeviationThreshold)
+        featureStandardDeviationThreshold(featureStandardDeviationThreshold),
+		predictedClasses((int)dataSet.size(), -1)
         {
         
     m = MothurOut::getInstance();
     globalDiscardedFeatureIndices = getGlobalDiscardedFeatureIndices();
+//    calculateFScore();
     // TODO: double check if the implemenatation of 'globalOutOfBagEstimates' is correct
 }
 
@@ -72,3 +74,80 @@ vector<int> Forest::getGlobalDiscardedFeatureIndices() {
 
 /***********************************************************************/
 
+void Forest::calculateFScore() {
+    vector< vector<int> > featureVectors(numFeatures, vector<int>(numSamples, 0) );
+    vector<int> outputVector(numSamples, 0);
+    vector<int> outputClasses;
+    int numOutputClasses = 0;
+
+        // calculate feature vectors and output vector
+    for (int i = 0; i < numSamples; i++) {
+        for (int j = 0; j < numFeatures; j++) { featureVectors[j][i] = dataSet[i][j]; }
+        int outcome = dataSet[i][numFeatures];
+        outputVector[i] = outcome;
+        vector<int>::iterator it = find(outputClasses.begin(), outputClasses.end(), outcome);
+        if (it == outputClasses.end()){       // find() will return classes.end() if the element is not found
+            outputClasses.push_back(outcome);
+            numOutputClasses++;
+        }
+    }    
+    
+    for (int i = 0; i < numFeatures; i++) {
+        double fisherNumerator = 0;
+        double fisherDenominator = 0;
+        double fisherScore = 0;
+        
+        long featureSum = accumulate(featureVectors[i].begin(), featureVectors[i].end(), 0);
+        double featureMean = (double) featureSum / (double) numSamples;
+        
+        vector<int> outputClassCounts(numOutputClasses, 0);
+        vector<long> outputClassBasedSums(numOutputClasses, 0);
+        vector<double> outputClassBasedMeans(numOutputClasses, 0);
+        
+        for (int j = 0; j < numSamples; j++) {
+            outputClassCounts[outputVector[j]]++;
+            outputClassBasedSums[outputVector[j]] += featureVectors[i][j];
+        }
+        
+        for (int k = 0; k < numOutputClasses; k++) {
+            outputClassBasedMeans[k] = (double) outputClassBasedSums[k] / (double)outputClassCounts[k];
+            fisherNumerator += pow(outputClassBasedMeans[k] - featureMean, 2);
+        }
+        
+        for (int j = 0; j < numSamples; j++) {
+                // TODO: choose between weighted and un-weighted denominator calculation
+                // which one is the correct and/or perfect?
+            
+                // This is the unweighted calculation
+//            fisherDenominator += pow(featureVectors[i][j] - outputClassBasedMeans[outputVector[j]], 2);
+            
+                // this is the weighted calculation
+            fisherDenominator += (1 / ((double)outputClassCounts[outputVector[j]] - 1)) * pow(featureVectors[i][j] - outputClassBasedMeans[outputVector[j]], 2);
+        }
+        
+        if (fisherDenominator == 0) { fisherDenominator = numeric_limits<double>::min(); }
+        
+        fisherScore = fisherNumerator / fisherDenominator;
+        pair<int, double> fisherScoreRank(i, fisherScore);
+        featureRanksByFScore.push_back(fisherScoreRank);
+    }
+    
+    VariableRankDescendingSorterDouble sorter;
+    sort(featureRanksByFScore.begin(), featureRanksByFScore.end(), sorter);
+        
+    // scale the scores, the top feature will have it's value scaled to 1
+//    for (vector<pair<int, double> >::iterator it = featureRanksByFScore.begin(); it != featureRanksByFScore.end(); it++) {
+//        it->second /= featureRanksByFScore.begin()->second;
+//        if (it->second < 0) {
+//            featureRanksByFScore.erase(it, featureRanksByFScore.end());
+//            break;
+//        }
+//    }
+    
+//    for (int i = 0; i < featureRanksByFScore.size(); i++) {
+//        cout << featureRanksByFScore[i].first << "\t" << featureRanksByFScore[i].second << endl;
+//    }
+//    cout << endl;
+//    exit(1);
+}
+

forest.h

@@ -51,13 +51,17 @@ class Forest{
     // the penalization would be averaged, so this woould unlikely to create a local optmina
     
     vector<int> getGlobalDiscardedFeatureIndices();
+    void calculateFScore();
     
     int numDecisionTrees;
     int numSamples;
     int numFeatures;
     vector< vector<int> > dataSet;
     vector<int> globalDiscardedFeatureIndices;
     vector<double> globalVariableImportanceList;
+    
+    vector< pair<int, double> > featureRanksByFScore;
+    
     string treeSplitCriterion;
   
     bool doPruning;
@@ -74,6 +78,10 @@ class Forest{
     // TODO: fix this, do we use pointers?
     vector<AbstractDecisionTree*> decisionTrees;
     
+    // predictedClasses[i] denotes the class predicted by the algorithm
+    // for i'th training sample
+    vector<int> predictedClasses;
+
     MothurOut* m;
     
 private:

fs/Datasets/HumanCRC.final.subsample.shared.encoding

@@ -0,0 +1,3 @@
+adenoma     0
+cancer     1
+Healthy     2