KNN_2.m

%Assignment 4 Q1 Gagan Khanijau 2011046

% clc
% clear
%%data loading and division
No_of_Samples= 1600;
No_of_features= size (teh,2)-1 ;
No_of_classes=100;
DataSet = d; %importdata('E:\datat\data.txt');
Indices_crossValidated= crossvalind('Kfold',No_of_Samples,2);
%creating traing and testing matrices

% DataSet_train = zeros(size(trh,1),No_of_features+1);
% DataSet_test = zeros(size(teh,1),No_of_features+1);
% r = randi([1 16],1,2);
% for i=1:No_of_Samples
%     ndx= mod(i,16);
%     ndx(ndx == 0) = 16;
%     if ndx ~= r(1) && ndx ~=r(2)
%         DataSet_train(i,1:No_of_features+1) = DataSet(i,1:No_of_features+1);
%     else
%         DataSet_test(i,1:No_of_features+1) = DataSet(i,1:No_of_features+1);
%     end
% end
% 
% DataSet_train(all(~DataSet_train,2), : ) = [];  
% DataSet_test( all(~DataSet_test,2), : ) = [];

DataSet_train= trh;
DataSet_test=teh;
%random initial centres k=No_of_classes
InitialRandomCentres= randperm(size(trh,1),No_of_classes);
CentreFeatures= zeros(No_of_classes,No_of_features);
% counter= 1;
for i=1:No_of_classes
      CentreFeatures(i,1:No_of_features)= DataSet_train(InitialRandomCentres(i),2:No_of_features + 1);
%       CentreFeatures(i,1:No_of_features)= DataSet_train(counter,2:No_of_features + 1);
%       counter= counter+14;
end
% CentreFeatures(1,1:4)= DataSet_train(InitialRandomCentres(1),2:5);
% CentreFeatures(2,1:4)= DataSet_train(InitialRandomCentres(2),2:5);
% CentreFeatures(3,1:4)= DataSet_train(InitialRandomCentres(3),2:5);


Train_assignedCluster= zeros(size(trh,1),1);
%stopping flag

flag= 1;

%algorithm k means
while(flag)
    flag=0;
    for i = 1:size(trh,1)
        DistanceFromCentres= zeros(No_of_classes,1);
        for j=1:No_of_classes
            DistanceFromCentres(j)= norm(DataSet_train(i,2:No_of_features + 1)- CentreFeatures(j,1:No_of_features));
        end
%         DistanceFromCentres(1)= norm(DataSet_train(i,2:5)- CentreFeatures(1,1:4));
%         DistanceFromCentres(2)= norm(DataSet_train(i,2:5)- CentreFeatures(2,1:4));
%         DistanceFromCentres(3)= norm(DataSet_train(i,2:5)- CentreFeatures(3,1:4));
        [sortedValues, sortedIndex]= sort(DistanceFromCentres,'ascend');
        if Train_assignedCluster(i)~=sortedIndex(1)
            Train_assignedCluster(i)=sortedIndex(1);
            flag=1;
        end    
    end 
    CentreFeatures= zeros(No_of_classes,No_of_features);
    PointsinCluster=zeros(No_of_classes,1);
    for j=1:size(trh,1)
        CentreFeatures(Train_assignedCluster(j),:) = CentreFeatures(Train_assignedCluster(j),:) + DataSet_train(j,2:No_of_features + 1); 
        PointsinCluster(Train_assignedCluster(j))=PointsinCluster(Train_assignedCluster(j))+1;
    end
    for j=1:No_of_classes
        CentreFeatures(j,:)= CentreFeatures(j,:)/PointsinCluster(j);
    end
%     CentreFeatures(1,:)= CentreFeatures(1,:)/PointsinCluster(1);
%     CentreFeatures(2,:)= CentreFeatures(2,:)/PointsinCluster(2);
%     CentreFeatures(3,:)= CentreFeatures(3,:)/PointsinCluster(3);
end


%%training data accuracy
TrainingFrequencyMatrix=zeros(No_of_classes);

for i=1:size(trh,1)
    TrainingFrequencyMatrix(Train_assignedCluster(i),DataSet_train(i))=TrainingFrequencyMatrix(Train_assignedCluster(i),DataSet_train(i))+1;
end

TrainingOutputLabels= zeros(size(trh,1),1);
for i=1:size(trh,1)
    [val,ndx]=max(TrainingFrequencyMatrix(Train_assignedCluster(i),:));
    TrainingOutputLabels(i)= ndx;
end

Correct=0;
for i=1:size(trh,1)
    if TrainingOutputLabels(i)==DataSet_train(i,1)
        Correct=Correct+1;
    end    
end

AccuracyTraining = Correct/size(trh,1);
Training_Error = 1- AccuracyTraining
%plotconfusion(training_data(:,1),train_outputLbls)

%%testing data accuracy 

Test_AssignedCluster= zeros(size(teh,1),1);
for i = 1:size(teh,1)
        DistanceFromCentres= zeros(No_of_classes,1);
        for j=1:No_of_classes
            DistanceFromCentres(j)= norm(DataSet_test(i,2:No_of_features + 1)- CentreFeatures(j,1:No_of_features));
        end
%         DistanceFromCentres(1)= norm(DataSet_test(i,2:5)- CentreFeatures(1,1:4));
%         DistanceFromCentres(2)= norm(DataSet_test(i,2:5)- CentreFeatures(2,1:4));
%         DistanceFromCentres(3)= norm(DataSet_test(i,2:5)- CentreFeatures(3,1:4));
        [sortedValues, sortedIndex]= sort(DistanceFromCentres,'ascend');
        Test_AssignedCluster(i)=sortedIndex(1);
end

TestingFrequencyMatrix=zeros(No_of_classes);

for i=1:size(teh,1)
    TestingFrequencyMatrix(Test_AssignedCluster(i),DataSet_test(i))=TestingFrequencyMatrix(Test_AssignedCluster(i),DataSet_test(i))+1;
end

TestingOutputLabels= zeros(size(teh,1),1);
for i=1:size(teh,1)
    [val,ndx]=max(TestingFrequencyMatrix(Test_AssignedCluster(i),:));
    TestingOutputLabels(i)= ndx;
end

Correct=0;
for i=1:size(teh,1)
    if TestingOutputLabels(i)==DataSet_test(i,1)
        Correct=Correct+1;
    end    
end

accuracytest = Correct/size(teh,1);
Testing_Error= 1-accuracytest

%plotconfusion(testing_data(:,1),test_outputLbls)