% Classification of handwritten digits
% 18th Machine Learning Summer School
%
% Author: Ruben Martinez-Cantin <rmcantin@unizar.es>
%
clear all, close all
seed = 943; randn('seed',seed), rand('seed',seed)

% If you can work with more digits, you can download the full datasets
% from Sam Roweis website http://cs.nyu.edu/~roweis/data.html
load mnist_0to2;

% Select training and testing sets
% We assume binary classification digit0 = +1 / others = -1
trainset = [train0(1:1000,:); train1(1:400,:); train2(1:400,:)];        
ytr = [ones(1000,1); -ones(800,1)];

testset = [test0(1:500,:); test1(1:200,:); test2(1:200,:)];
yte = [ones(500,1); -ones(400,1)];

% MATLAB only knows how to do operations with type "double" but the data is
% stored as "uint8". This transforms uint8 -> double 
xtrain = mat2gray(trainset);
xtest = mat2gray(testset);

% Dimensionality reduction using PCA
nPCAComp = 16;
[xtr,v,m] = reduceusingpca(xtrain,nPCAComp);
xte = project2pca(xtest,v,m);

% Set up the GP
% TODO: Modify these lines
%--------------------------------------------------------------------------cov = 
hyp0.cov  = 
mean = 
hyp0.mean = 
lik = 
hyp0.lik =
inf = 
%--------------------------------------------------------------------------

% Classify
Ncg = 50;                                   % number of conjugate gradient steps
hyp = minimize(hyp0,'gp', -Ncg, inf, mean, cov, lik, xtr, ytr);     % opt hypers
[ymu, ys2] = gp(hyp, inf, mean, cov, lik, xtr, ytr, xte);              % predict

y = 2*[ymu>0]-1;
fprintf(1,'Classification results\n');
fprintf(1,'%2.2f %% success rate\n',sum(y==yte) / length(yte)*100);