clear; close all;
%%%%% source code for homework6 %%%%%
%% 1. load the data %%%%%%%%%%%%
fprintf(1,'\nMNIST data loading...\n');
load MNIST_data;
whos

%% 2. Specify the training and testing data involved in this task. %%%%%%%%%%%%
target_digits = [1 3 5 7] ; % 1:10;
if length(target_digits)<10
    [train_index, test_index] = deal(cell(1, length(target_digits)));
    % look for the index of the training data
    for i = 1 : length(target_digits)
        train_index{i} = find(trainlabel == target_digits(i));
        test_index{i}  = find(testlabel  == target_digits(i));
    end
    train_index = sort(cat(2, train_index{:}));
    test_index  = sort(cat(2, test_index{:}));

    traindata = traindata(:,train_index);
    testdata = testdata(:,test_index);
    % groundtruth labels
    trainlabel = trainlabel(train_index);
    testlabel = testlabel(test_index);
end
gn = size(traindata,2);
tn = size(testdata,2);
fprintf(1,'learning to classify digits [%s] \n', num2str(target_digits));
fprintf(1,' %d examples for training, %d for testing \n', gn, tn);

%% 3. classification with minimum distance classifiers %%%%%%%%%%%%
tic;
fprintf(1,'\nLearning minimum-distance classifier ...\n');
[centers, ulabels] = mdist_learn(traindata, trainlabel);
[clabel, err_rate] = mdist_classify(centers, ulabels, testdata, testlabel);
fprintf(1,'\n the error rate is %.3f%%\n', err_rate*100);
toc;

%% 4. 1NN classification with Euclidean metric %%%%%%%%%%%%
tic;
fprintf(1,'\nClassification with 1-NN ...\n');
% final correct rate
err_rate = NN_euclidean(traindata, trainlabel, testdata, testlabel);
fprintf(1,'\n the error rate is %.3f%%\n', err_rate*100);
toc;


%% 5. k-NN classification with L3 metric %%%%%%%%%%%%


%% 6. linear perceptron with subsampled training data %%%%%%%%%%%%
randid = randperm(gn); randid = sort(randid(1:10000));
traindata = traindata(:, randid);
trainlabel = trainlabel(randid);
%%% you linear perceptron training code here