batch_digits.m

clustering_code ,Clustering Through Ranking On Manifolds Version 0.2 Copyright by Markus Breit

function batch_Digits

close all;

dbstop if error;

load Digits_Data

[num_ex,num_labels] = size(True_Y);

for i = 1:num_labels
    ind_data(i).labels = find(True_Y(:,i) == 1);
end


lrn_par = Set_Default_Learning_Paramters;
% Optimization for both ALPHA and SIGMA together
lrn_par.OPT_S_A = 1; % set to 1 to optimize for both alpha and sigma

%Clustering Example
Y = []; %Set to null for clustering
model = LG_Cluster(Data_X,lrn_par)


lrn_par.OPT_S_A = 0;
model2 = LG_Cluster(Data_X,lrn_par);

fprintf(1,'sigma %f alpha %f clusters %i\n',model.my_sigma,model.my_alpha,model.num_classes);

save digits_model;

True_Y=True_Y(:,[1 3 4 2]);

errors = abs(model.Y - True_Y);
err_ind = find(sum(abs(model.Y - True_Y)')'>0);
num_errors = length(err_ind);
err_rate = num_errors / num_ex

errors = abs(model2.Y - True_Y);
err_ind = find(sum(abs(model2.Y - True_Y)')'>0);
num_errors = length(err_ind);
err_rate = num_errors / num_ex

t10000 = 5;

pix=16;
line=[];
nifty_plot=[];
for i=1:model.num_classes
    pic = 256 * ((reshape(Data_X(model.ind_clust(i),:),16,16))' + 1)/2; % representant for the class
    line=[line,zeros(pix,1),pic];
    if mod(i,4)==0
        nifty_plot=[nifty_plot; zeros(1,length(line)); line];
        line=[];
    end;
end;
if mod(i,4)~=0
        while size(line,2) < size(nifty_plot,2)
            line=[line,zeros(pix,1),zeros(pix,pix)];
        end;
        nifty_plot=[nifty_plot; zeros(1,length(line)); line];
end;
figure;colormap(gray);image(nifty_plot);
axis off;
title('Class Representants');

nifty_plot=[];
% plot class outliers outliers
for i = [1 3 4 2]
    line= 256 * ((reshape(Data_X(model.ind_clust(i),:),16,16))' + 1)/2; % representant for the class
    line=[line,zeros(pix,1),256*ones(pix,2),zeros(pix,1)];
    for j = 1:8
        ex_dig = 256 * ((reshape(Data_X(model.Class_Outlier(i).ind_sort(j),:),16,16))' + 1)/2;
        line=[line,zeros(pix,1),ex_dig];
        %figure;image(ex_dig);
        t77 = 77;
    end
    line = [line,zeros(pix,1)];
    nifty_plot=[nifty_plot; zeros(1,length(line)); line ];
    t77 = 0;
end
figure;colormap(gray);image(nifty_plot);
axis off;
title('Class Outliers');



% plot overall outliers
nifty_plot=[];
ii=1;
for i = 1:5
    line=[];
    for j=1:5
        ex_dig = 256 * ((reshape(Data_X(model.ind_sort_mean_dist(ii),:),16,16))' + 1)/2;
        line=[line,zeros(pix,1),ex_dig];
        ii=ii+1;
    end;
    %figure;image(ex_dig);
    line = [line,zeros(pix,1)];
    nifty_plot=[nifty_plot; zeros(1,length(line)); line ];
    t77 = 77;
end
figure;colormap(gray);image(nifty_plot);
axis off;
title('Worst overall outliers');


if 1==2
    fprintf(1,'Handling unseen data...\n');
    load Digits_Data_us
    [un_class] = Classify_New_Data(Data_X_us,model);
    True_Y_us=True_Y_us(:,[1 3 4 2]);
    err_ind = find(sum(abs(un_class.Y - True_Y_us)')'>0);
    num_errors = length(err_ind);
    err_rate = num_errors / size(Data_X_us,1)
end;

if 1==2
    fprintf(1,'Handling unseen data 2...\n');
    load Digits_Data_us
    [un_class] = Classify_New_Data(Data_X_us,model2);
    True_Y_us=True_Y_us(:,[1 3 4 2]);
    err_ind = find(sum(abs(un_class.Y - True_Y_us)')'>0);
    num_errors = length(err_ind);
    err_rate = num_errors / size(Data_X_us,1)
end;

t10000 = 5;

[i,labels]=max(model.F_orig,[],2);

figure;
hold on;
for i=1:model.num_classes
    ind_x=find(labels==i);
    plot_values = model.Class_Outlier(i).plot_values;
    plot_value2 = model2.Class_Outlier(i).plot_values;
    plot(plot_values(ind_x,:),'r*'); %plot the points in each cluster
    plot(plot_value2(ind_x,:),'b+'); %plot the points in each cluster
end;
legend('Optimizing Alpha+Sigma','Optimizing Sigma only')
title('Class outliers');

legendpos=[0,2,2,0];
for i=1:model.num_classes
    plot_values = model.Class_Outlier(i).plot_values;
    plot_value2 = model2.Class_Outlier(i).plot_values;
    figure;
    hold on;
    plot(plot_values,'r*'); %plot the points in each cluster
    plot(plot_value2,'b+'); %plot the points in each cluster
    legend('Optimizing Alpha+Sigma','Optimizing Sigma only',legendpos(i))
    mytitle=sprintf('Class outliers for class %i',i);
    title(mytitle);
end;

% err_rate =
% 
%     0.0362
% 
% 
% err_rate =
% 
%     0.0388