DaveM@6: function featureVector = rfFeatureSelection(data, labels, numFeatures, iterMethod, numTrees, featureVector)
DaveM@4: % rfFeatureSelection(data, labels, numFeatures, iterMethod, numTrees)
DaveM@3: %
DaveM@4: % using random forests to perform feature selection for a given data set
DaveM@4: % data has size (x,y), where x is the number of labels and y, the number of
DaveM@4: % features. 
DaveM@4: % labels is the set of labels for the data
DaveM@4: % numFeatures is the dimension of the output vector (default 5)
DaveM@4: % iterMethod is the method for which the features are cut down
DaveM@5: %       * 'onePass' will simply select the top (numFeatures) features and
DaveM@5: %       report them 
DaveM@5: %       * 'cutX' will iteratively cut the bottom X percent of
DaveM@5: %       features out, and perform random forest feature selection on the
DaveM@5: %       new set, until the desired number of features has been returned
DaveM@5: %       * 'oobErr' will do something with the out-of-bag error, and return
DaveM@4: %       that in some way, but this has not been implemented yet.
DaveM@5: %       * 'featureDeltaErr' will do something with the feature importance
DaveM@5: %       prediction error, and return that in some way, but this has not
DaveM@5: %       been implemented yet. The OOBPermutedVarDeltaError property is a
DaveM@5: %       numeric array of size 1-by-Nvars containing a measure of importance
DaveM@5: %       for each predictor variable (feature). For any variable, the
DaveM@5: %       measure is the increase in prediction error if the values of that
DaveM@5: %       variable are permuted across the out-of-bag observations. This
DaveM@5: %       measure is computed for every tree, then averaged over the entire
DaveM@5: %       ensemble and divided by the standard deviation over the entire
DaveM@5: %       ensemble.
DaveM@6: % featureVector is a list of the features to use, for recursive purposes.
DaveM@3: 
DaveM@3: if(length(labels) ~= size(data,1))
DaveM@3:     error('labels and data do not match up');
DaveM@3: end
DaveM@3: 
DaveM@3: if(nargin < 2)
DaveM@3:     error('must pass data and labels into function')
DaveM@3: end
DaveM@3: if(nargin < 3)
DaveM@3:     numFeatures = 5;
DaveM@3: end
DaveM@3: if(nargin < 4)
DaveM@3:     iterMethod = 'onePass';
DaveM@3: end
DaveM@3: if(nargin < 5)
DaveM@3:     numTrees = 200;
DaveM@3: end
DaveM@6: if(nargin < 5)
DaveM@6:     featureVector = 1:size(data,2);
DaveM@6: end
DaveM@3: 
DaveM@3: 
DaveM@6: if(length(featureVector) > numFeatures)
DaveM@6:     options = statset('UseParallel', true);
DaveM@6:     b = TreeBagger(numTrees, data(:,featureVector), labels,'OOBVarImp','On',...
DaveM@6:         'SampleWithReplacement', 'Off','FBoot', 0.632,'Options', options);
DaveM@6:     [FI,I] = sort(b.OOBPermutedVarDeltaError,'descend'); 
DaveM@6:     featureVector = featureVector(I);
DaveM@3: 
DaveM@6:     if(strcmp(iterMethod,'onePass'))
DaveM@6:         disp('onePass')
DaveM@6:         featureVector = featureVector(1:numFeatures);
DaveM@6:     elseif(strcmp(iterMethod(1:3),'cut'))
DaveM@6:         disp(iterMethod)
DaveM@6:         cutPercentage = str2double(iterMethod(4:end));
DaveM@6:         cutSize = max(floor(length(featureVector)*cutPercentage/100),1);
DaveM@6:         if(length(featureVector) - cutSize < numFeatures)
DaveM@6:             cutSize = length(featureVector) - numFeatures;
DaveM@6:         end
DaveM@6:         featureVector = featureVector(1:end-cutSize);
DaveM@6:     %     data = data(:,sort(featureVector));
DaveM@6:         featureVector = rfFeatureSelection(data, labels, numFeatures, iterMethod, numTrees, featureVector);
DaveM@6:     elseif(strcmp(iterMethod,'oobErr'))
DaveM@6:         warning('This method has not been implemented yet, using onePass to return results')
DaveM@6:         featureVector = featureVector(1:numFeatures);
DaveM@6:     elseif(strcmp(iterMethod,'featureDeltaErr'))
DaveM@6:         warning('This method has not been implemented yet, using onePass to return results')
DaveM@6:         % this will use variable FI
DaveM@6:         featureVector = featureVector(1:numFeatures);
DaveM@6:     end
DaveM@3: end
DaveM@3: end