KM+LS.m

clc;clear all;close all;

load('Train.mat');
load('GroundTruth.mat');
load('Test.mat');
load('TrueClass.mat');

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%% FINDING CENTERS
N = 60000;
K = 100;
ClosestCentroids = zeros(1,N);
centres = datasample(Train,K,1,'Replace',false);
centres_new = zeros(K,785);
Iteration = 1;
while(true)
for i=1:N
    [value index] = min(sum(sqrt((bsxfun(@minus,Train(i,:),centres)).^2),2));
    ClosestCentroids(i) = index;
end

for i=1:K
    centres_new(i,:)= mean(Train(i == ClosestCentroids,:));
end
a = corrcoef(centres,centres_new);
if(a(2,1)>0.99)
    break;
else
    centres = centres_new;
end
Iteration = Iteration + 1;
end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%% FINDING VARIANCES
d_max = 0;
for i = 1:K-1
    my_dist(i) = sum((centres_new(i,:)-centres_new(i+1,:)).^2);
    if  my_dist(i) > d_max
            d_max = my_dist(i);
    else
    end
end

% Normalization method from the slides
spread = (d_max/sqrt(2*K))*ones(K,1);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%% FINDING WEIGHTS
centres_new = centres_new';
load('Train.mat');
load('GroundTruth.mat');
load('Test.mat');
load('TrueClass.mat');

w = rand(10,K)./2  - 0.25;
g = 0;
for epoch = 1:1
    for m = 1:length(Train)
        x = Train(m,:);
        d = GroundTruth;

        % Gaussian RBF Kernel, or RBF Kernel: https://en.wikipedia.org/wiki/Radial_basis_function_kernel
        % Applying Kernel on the Test Data        
        for i = 1:K
            g(i,m) = exp(-(x-centres_new(:,i)')*(x-centres_new(:,i)')'/(2*spread(i)^2));
        end
    end
    
    % Implementation of the Least Squares algorithm    
    w=pinv(g)'*d;
end

    % Applying Kernel on the Test Data
for m = 1:length(Test)
    x = Test(m,:);

    for i = 1:K
        g_test(i,m) = exp(-(x-centres_new(:,i)')*(x-centres_new(:,i)')'/(2*spread(i)^2));
    end
end

% Predicting the values and finding Classification Rate.
PREDICTIONS = (g_test'*w);
for z = 1:length(Test)
    [val, col] = max(PREDICTIONS(z,:));
    predict(z,:) = col-1;
end
CR = (sum(predict==TrueClass)/length(TrueClass))*100

% Visualizations
figure;
plot(1:epoch,CR,'b--o');
title('Recognition Curve');
xlabel('Number of epochs');
ylabel('Classification Rate (%)');