% This program calls the DeepReLU function and trains the
%network max_epoch times. It enters the training data into the trained 
%network and displays the output. We verify the adequacy of the training by 
%comparing the output and correct output.

clear all
close all
clc

X = zeros(5, 5, 5);

% The input set consists of five 5x5 pixel squares.
%0: white pixel; 1: black pixel.
X(:, :, 1) = [0 1 1 0 0;
              0 0 1 0 0;
              0 0 1 0 0;
              0 0 1 0 0;
              0 1 1 1 0]; %1
          
X(:, :, 2) = [1 1 1 1 0;
              0 0 0 0 1;
              0 1 1 1 0;
              1 0 0 0 0;
              1 1 1 1 1]; %2
          
X(:, :, 3) = [1 1 1 1 0;
              0 0 0 0 1;
              0 1 1 1 0;
              0 0 0 0 1;
              1 1 1 1 0]; %3
          
X(:, :, 4) = [0 0 0 1 0;
              0 0 1 1 0;
              0 1 0 1 0;
              1 1 1 1 1;
              0 0 0 1 0]; %4
          
X(:, :, 5) = [1 1 1 1 1;
              1 0 0 0 0;
              1 1 1 1 0;
              0 0 0 0 1;
              1 1 1 1 0]; %5
      
% Desired outputs mapped via one-hot encoding (or 1-of-N encoding):
D = [1 0 0 0 0;     %1
     0 1 0 0 0;     %2
     0 0 1 0 0;     %3
     0 0 0 1 0;     %4
     0 0 0 0 1];    %5
 

% Weights initialization:
W1 = 2*rand(20, 25) - 1;
W2 = 2*rand(20, 20) - 1;
W3 = 2*rand(20, 20) - 1;
W4 = 2*rand( 5, 20) - 1;

% Training process:
max_epoch = 10000;
for epoch = 1:max_epoch
    [W1, W2, W3, W4] = DeepReLU(W1, W2, W3, W4, X, D);
end


% Inference:
N = size(X,3);
y = zeros(N,size(D,2));
for k = 1:N
    x = reshape(X(:, :, k), 25, 1);
    v1 = W1*x;
    y1 = ReLU(v1);
    v2 = W2*y1;
    y2 = ReLU(v2);
    v3 = W3*y2;
    y3 = ReLU(v3);
    v = W4*y3;
    y(k,:) = Softmax(v);
end

disp('Results:');
disp('   [desired]:');
disp(D);
disp('   [network_output]:');
disp(y)

%% Showing images:
% for i = 1:N
%     compareImages(X(:,:,i), y(i,:));
% end