auf22.m

clear all;
close all;
clc;

% read data from files
testfile='pendigits-testing.txt';
trainfile='pendigits-training.txt';
X_train=dlmread(trainfile);
X_test=dlmread(testfile);

% Initialisierung
C=zeros(10);
means=zeros(10,16);
cm=zeros(160,16);

% for every digit
for i=0:9
    % find all vectors labeled with n
	idx=find(X_train(:,end)==i);
	cluster = X_train(idx,1:16);
    % compte mean of cluster
	means(i+1,:) = mean(cluster);
    % compute covariance matrix
    cm(16*i+1:16*(i+1),:) = cov(cluster);
end;
% add some noise
cm = cm .+ 0.00001;

%loop through testSamples
for i=1:size(X_test,1)
    % p is matrix of likelyhoods
	p = zeros(1,10);
    % compute likelyhood for every digit and put in p
	for j=0:9
		x = X_test(i,:);
		vec=x(1:16)-means(j+1,:);
		exponent=-0.5*vec* inv(cm(16*j+1:16*(j+1),:))*vec';
		normal = 1/((2*pi)^8 * sqrt(det(cm(16*j+1:16*(j+1),:))));
		p(j+1)=normal*exp(exponent);
	end;
    % compute maximum over p and update confusion matrix
	[v,idx]=max(p);
	belief = idx-1;	
	actual = x(17);
	C(belief+1,actual+1)=C(belief+1,actual+1)+1;
end;

% print confusion matrix and results
rate=trace(C)/sum(C(:));
fprintf(1,'confusion matrix\n');
disp(C);
missed=sum(C(:))-trace(C);
fprintf(1,'%d misclassified out of %d test samples\n',missed,sum(C(:)));
fprintf(1,'rate (correct): %f\n',rate);