% ---------------------------------------------

%

% Simple classification of hand written digits

% using a least squares classifier

%

% Johan Dahlin ([email protected])

% 2013-03-19

%

% ---------------------------------------------

clear all;

% Import the training and test data

% The Zip code data is available from:

% http://www-stat.stanford.edu/~tibs/ElemStatLearn/

zipTraining = importdata('zip.train');

zipTesting = importdata('zip.test');

% Find the classes and the data

classTraining=zipTraining(:,1);

classTesting=zipTesting(:,1);

dataTraining=zipTraining(:,2:end);

dataTesting=zipTesting(:,2:end);

%Train the classifier

% Construct the vector of 1-of-K coding scheme

classTrainingCoding=zeros(length(classTraining),10);

for ii=1:length(classTraining)

classTrainingCoding(ii,classTraining(ii) 1)=1;

end

% Estimate the parameters

beta=classTrainingCoding\dataTraining;

%Classify the test data

correct=0;

for ii=1:length(classTesting)

feature(ii,:)=beta*dataTesting(ii,:)';

[~,classified(ii)]=max(feature(ii,:));

classified(ii)=classified(ii)-1;

correct=correct (classified(ii)==classTesting(ii));

end

% Estimate the classification error

1-correct/length(classTesting)

% 38.22% mis-classification rate