Recreating the dataset from the MFS Paper by Bayne

main.m

function main()

    % LOAD IRIS
    % dataset was modified by changing the class name from string to a
    % number
    iris = load('iris.csv', ','); 
    
    % LOAD SONAR
    sonar = load('sonar.csv', ',');

    % NN & KNN TESTING
    display(['Iris NN    ==> ' num2str(averageTestKNN(100, iris, 80, 1))]);
    display(['Iris kNN 2 ==> ' num2str(averageTestKNN(100, iris, 80, 2))]);
    display(['Iris kNN 3 ==> ' num2str(averageTestKNN(100, iris, 80, 3))]);
    display(['Iris kNN 4 ==> ' num2str(averageTestKNN(100, iris, 80, 4))]);
    display(['Iris kNN 5 ==> ' num2str(averageTestKNN(100, iris, 80, 5))]);

    display(['Sonar NN    ==> ' num2str(averageTestKNN(100, sonar, 80, 1))]);
    display(['Sonar kNN 2 ==> ' num2str(averageTestKNN(100, sonar, 80, 2))]);
    display(['Sonar kNN 3 ==> ' num2str(averageTestKNN(100, sonar, 80, 3))]);
    display(['Sonar kNN 4 ==> ' num2str(averageTestKNN(100, sonar, 80, 4))]);
    display(['Sonar kNN 5 ==> ' num2str(averageTestKNN(100, sonar, 80, 5))]);
    
    
    % MFS IRIS
    display(['Iris mfs1 1  ==> ' num2str(averageTestMFS(10, iris, 80, 1, 'mfs1'))]);
    display(['Iris mfs1 2 ==> ' num2str(averageTestMFS(10, iris, 80, 2, 'mfs1'))]);
    display(['Iris mfs1 3 ==> ' num2str(averageTestMFS(10, iris, 80, 3, 'mfs1'))]);
    display(['Iris mfs1 4 ==> ' num2str(averageTestMFS(10, iris, 80, 4, 'mfs1'))]);
    
    display(['Iris mfs2 1  ==> ' num2str(averageTestMFS(10, iris, 80, 1, 'mfs2'))]);
    display(['Iris mfs2 2 ==> ' num2str(averageTestMFS(10, iris, 80, 2, 'mfs2'))]);
    display(['Iris mfs2 3 ==> ' num2str(averageTestMFS(10, iris, 80, 3, 'mfs2'))]);
    display(['Iris mfs2 4 ==> ' num2str(averageTestMFS(10, iris, 80, 4, 'mfs2'))]);
    
    
    % MFS SONAR
    display(['Sonar mfs1 6  ==> ' num2str(averageTestMFS(10, sonar, 80, 6, 'mfs1'))]);
    display(['Sonar mfs1 12 ==> ' num2str(averageTestMFS(10, sonar, 80, 12, 'mfs1'))]);
    display(['Sonar mfs1 18 ==> ' num2str(averageTestMFS(10, sonar, 80, 18, 'mfs1'))]);
    display(['Sonar mfs1 24 ==> ' num2str(averageTestMFS(10, sonar, 80, 24, 'mfs1'))]);
    display(['Sonar mfs1 30 ==> ' num2str(averageTestMFS(10, sonar, 80, 30, 'mfs1'))]);
    display(['Sonar mfs1 36 ==> ' num2str(averageTestMFS(10, sonar, 80, 36, 'mfs1'))]);
    display(['Sonar mfs1 42 ==> ' num2str(averageTestMFS(10, sonar, 80, 42, 'mfs1'))]);
    display(['Sonar mfs1 48 ==> ' num2str(averageTestMFS(10, sonar, 80, 48, 'mfs1'))]);
    display(['Sonar mfs1 54 ==> ' num2str(averageTestMFS(10, sonar, 80, 54, 'mfs1'))]);
    display(['Sonar mfs1 60 ==> ' num2str(averageTestMFS(10, sonar, 80, 60, 'mfs1'))]);
    
    display(['Sonar mfs2 6  ==> ' num2str(averageTestMFS(10, sonar, 80, 6, 'mfs2'))]);
    display(['Sonar mfs2 12 ==> ' num2str(averageTestMFS(10, sonar, 80, 12, 'mfs2'))]);
    display(['Sonar mfs2 18 ==> ' num2str(averageTestMFS(10, sonar, 80, 18, 'mfs2'))]);
    display(['Sonar mfs2 24 ==> ' num2str(averageTestMFS(10, sonar, 80, 24, 'mfs2'))]);
    display(['Sonar mfs2 30 ==> ' num2str(averageTestMFS(10, sonar, 80, 30, 'mfs2'))]);
    display(['Sonar mfs2 36 ==> ' num2str(averageTestMFS(10, sonar, 80, 36, 'mfs2'))]);
    display(['Sonar mfs2 42 ==> ' num2str(averageTestMFS(10, sonar, 80, 42, 'mfs2'))]);
    display(['Sonar mfs2 48 ==> ' num2str(averageTestMFS(10, sonar, 80, 48, 'mfs2'))]);
    display(['Sonar mfs2 54 ==> ' num2str(averageTestMFS(10, sonar, 80, 54, 'mfs2'))]);
    display(['Sonar mfs2 60 ==> ' num2str(averageTestMFS(10, sonar, 80, 60, 'mfs2'))]);
 
end

% Creates two different matrices one for testing and one for trainData
function [train, test] = randomizeDataset(data, percentage)
    shuffledArray = data(randperm(size(data,1)),:);
    
    split = round(size(data,1)*(percentage/100));
    train = shuffledArray(1:split, :);
    test = shuffledArray(split+1:size(data,1), :);
end 


function average = averageTestKNN(numberOfTests, data, percentage, n)
     result = zeros(numberOfTests,1);
    
     for i = 1:1:numberOfTests
        [trainData,testingData] = randomizeDataset(data, percentage);
        result(i,1) = 100 - knn(testingData, trainData, n);
     end
     
     
     average = mean(result);
end


function average = averageTestMFS(numberOfTests, data, percentage, features, type)
     result = zeros(numberOfTests,1);
    
     for i = 1:1:numberOfTests
        [trainData,testingData] = randomizeDataset(data, percentage);
        result(i,1) = 100 - mfs(testingData, trainData, features, type);
     end
    
     average = mean(result);
end




function error = knn(testingData, trainData, n)
    result =  zeros(size(testingData, 1),1);
    for i = 1:1:size(testingData, 1)
        if(testingData(i,size(testingData,2)) == identify(trainData, testingData(i,:), n));
            result(i) = 100; 
        else
            result(i) = 0;
        end
    end
    % Get overall error
    error = mean(result);
end


function error = mfs(testingData, trainData, features, type)
    % Function to easily check the performance of mfs1 & mfs2 
    % The number of features can be choosen as necessary
    result =  zeros(size(testingData, 1),1);
    for i = 1:1:size(testingData, 1)
        if (strcmp(type,'mfs1'))
            identification = identifymfs1(trainData, testingData(i,:), features);
        elseif (strcmp(type,'mfs2'))
            identification = identifymfs2(trainData, testingData(i,:), features);
        else
            display('Type has to be either mfs1 or mfs2');
        end
            
        if(testingData(i,size(testingData,2)) == identification);
            result(i) = 100; 
        else
            result(i) = 0;
        end
    end
    % Get overall error
    error = mean(result);
end


function class = identifymfs1(training, item, features)
     % Wrapper class for mfs1 (with replacement)
    class = identifymfs(training, item, true, features);
end     

function class = identifymfs2(training, item, features)
     % Wrapper class for mfs2 (without replacement)
    class = identifymfs(training, item, false, features);
end

function class = identifymfs(training, item, replacement, features)
    % features = 4; % to get data this variable was manually changed :)
    runthroughs = 10; % how many times it does voting on it
    identity = zeros(runthroughs, 1);
    
    for i = 1:1:runthroughs
           [pickedFeatures, itemFeatures] = pickFeature(training, item, features, replacement);
           identity(i, 1) = identify(pickedFeatures, itemFeatures, 1); % k = 1
    end
    
    class = mode(identity);
end


function [dataPicked, itemPicked] = pickFeature(data, item, n, replacement)
    % This functions picks random features, but leaves the identification
    % column (last one) in tact
    
    numberOfColumns = size(data,2);
    pickedColumns = numberOfColumns;
    
    % loop through as many features(columns) to select
    for i = 1:1:n
        
        if (replacement) 
            % With replacement just choose a random value form the columns
            % name (that also explain the negative one, as you can't choose
            % the last value
            value = ceil((numberOfColumns-1)*rand());
        else
            % find intersection from previous one
            chooseFrom = setdiff(1:1:numberOfColumns, pickedColumns);

            % pick a random item from the ones left over
            value = chooseFrom(ceil(length(chooseFrom)*rand()));
        end 
        
        % this line isn't as efficient as it could be, as it creates a new
        % array in memory, but should be fine since the array is so small 
        pickedColumns(end+1) = value;
    end
    
    
    % sort it, so the identification number stays at the end
    pickedColumns = sort(pickedColumns);
    
    % filter by the correct columns
    dataPicked = data(:, pickedColumns);
    itemPicked = item(:, pickedColumns);
end 


function class = identify(training, item, k)
    % get the distances
    distData = dist(training, item);
    
    % sort them
    distData = sortrows(distData,1);
    
    % find the most common element
    class = mode(distData(1:k,2));
end


function d = dist(data, item)
    %get the size
    w = size(data);
    
    % create empty array for storage
    result = zeros(w(1),2);
    
    % loop through each row  
     for i = 1:1:w(1)
         
         % set label, to the class type
         result(i,2) = data(i, w(2));
         
         % loop through all items in row - except for last one which is the
         % class name
         for j = 1:1:w(2)-1
            temp = data(i,j) - item(1,j);
            result(i,1) = result(i,1) + temp^2;
         end 
       
         result(i,1) = sqrt(result(i,1));
     end 

    d = result;
end