DaveM@2: clearvars;
DaveM@2: load('AdobeStratified.mat');
DaveM@2: morefeatures = true;
DaveM@2: idxvar = (1:1450);
DaveM@2: count = 1;
DaveM@2: featuredata = struct('IdxVar', [], 'FeatureNamesRanked', {}, 'FeatureImportance', [], 'OOBError', [], 'LastOOBError', [], 'EMClusters', [], 'AIC', [], 'PreviousAIC', []); 
DaveM@2: 
DaveM@2: while(morefeatures)
DaveM@2:     DataTrain = DataTrain(:, idxvar);
DaveM@2:     FeatureNames = FeatureNames(idxvar);
DaveM@2:     idxvar = (1:length(FeatureNames));
DaveM@2:     fprintf('\n Growing a Random Forest of 200 trees using %i features\n',length(idxvar))
DaveM@2:     
DaveM@2:     rng(1945,'twister')
DaveM@2:     tic
DaveM@2:     options = statset('UseParallel', true);
DaveM@2:     b = TreeBagger(200, DataTrain, LabelsTrain,'OOBVarImp','On', 'SampleWithReplacement', 'Off', 'FBoot', 0.632, 'Options', options);
DaveM@2:     toc
DaveM@2:     
DaveM@2:     oobErr = oobError(b);
DaveM@2:     LastoobErr = oobErr(end);
DaveM@2:     
DaveM@2:     fprintf('\n The cumulative OOB Error at 200 trees is %f\n', LastoobErr);
DaveM@2:     
DaveM@2:     Indices = crossvalind('Kfold', size(DataTrain, 1), 10);
DaveM@2:     
DaveM@2:     AICInitial = 1e16;
DaveM@2:     AICNext = -1e16;
DaveM@2:     AICAvg = zeros(10, 1);
DaveM@2:     NumClusters = 1;
DaveM@2:     
DaveM@2:     while(AICNext <= AICInitial)
DaveM@2:         
DaveM@2:         if(NumClusters ~= 1)
DaveM@2:             AICInitial = AICNext; 
DaveM@2:         end
DaveM@2:         NumClusters = NumClusters + 1;
DaveM@2:         
DaveM@2:         fprintf('\n Performing EM using 10 fold CV and %i clusters and %i features\n', NumClusters, length(idxvar))
DaveM@2:         
DaveM@2:         for i = 1:10
DaveM@2:            
DaveM@2:             emidx = (Indices == i); emidx = ~emidx;
DaveM@2:             
DaveM@2:             EMDataTrain = DataTrain(emidx, :);
DaveM@2:             GMModelCV = fitgmdist(EMDataTrain, NumClusters, 'RegularizationValue', 1e-5);
DaveM@2:             AICAvg(i) = GMModelCV.AIC;
DaveM@2:         end
DaveM@2:         
DaveM@2:         AICNext = mean(AICAvg);
DaveM@2:         fprintf('The average AIC was %f\n', AICNext);
DaveM@2:     end
DaveM@2:     
DaveM@2:     FI = b.OOBPermutedVarDeltaError;
DaveM@2:     
DaveM@2:     [FI,I]=sort(FI, 'descend');
DaveM@2:     idxvar = idxvar(I);
DaveM@2:     FeatureNamesRanked = FeatureNames(I);
DaveM@2:     
DaveM@2:     featuredata(count).IdxVar = idxvar;
DaveM@2:     featuredata(count).FeatureNamesRanked = FeatureNamesRanked;
DaveM@2:     featuredata(count).FeatureImportance = FI;
DaveM@2:     featuredata(count).OOBError = oobErr;
DaveM@2:     featuredata(count).LastOOBError = LastoobErr;
DaveM@2:     featuredata(count).EMClusters = NumClusters;
DaveM@2:     featuredata(count).AIC = AICNext;
DaveM@2:     featuredata(count).PreviousAIC = AICInitial;    
DaveM@2:     
DaveM@2:     idxRemove = round((length(idxvar) / 100)* 1);
DaveM@2:     fprintf('\n %i features will be removed.\n', idxRemove)
DaveM@2:     idxRemove = (length(idxvar) - idxRemove);
DaveM@2:     idxvar = idxvar(1:idxRemove);
DaveM@2:     count = count + 1;
DaveM@2:     
DaveM@2:     save('Results1Percent.mat', 'featuredata');
DaveM@2:     
DaveM@2:     if(length(idxvar) == 2)
DaveM@2:         morefeatures = false;
DaveM@2:     end
DaveM@2: end