%
% BPXORB.M: XOR Problem (HW 13 part 3)
% Batch mode training
%
% Author: G. Dempsey, Bradley University Electrical Engineering
% Last code modification 2/21/09
% modification required for new version (6.0) of NN Toolbox
%
% A network, consisting of 2 inputs fed to a two-layer network,
%   is trained with backpropagation on a 4 pattern problem.
%   Layer 1 consists of 2 log-sigmoid neurons, while layer 2
%   is made up of 1 log-sigmoid neuron.
%
%===================
clc;                        % clear console
clear;                      % clear variables
close all;                  % close all figure windows

% Define input vectors.
Y1 = [0 0]';
Y2 = [0 1]';
Y3 = [1 0]';
Y4 = [1 1]';

disp('display input and target patterns') 
P = [Y1 Y2 Y3 Y4]

% Define the associated targets.
T = [0 1 1 0];
T = [0.15 0.85 0.85 0.15]	% overwrite ideal targets with practical values
%
%
% INITIALIZE NETWORK ARCHITECTURE
%================================

% Set input vector size R, layer sizes S1 & S2

S1 = 2;         % two hidden neurons
S2 = 1;         % one output neuron
R=2;            % 2 components of input vector (documentation only)

% initialize the network with single command
% Initialize weights and biases. Nyguyen/Widrow method is the default
% function newff for Version 5.0 Toolbox: enclosed in ***********
%****************************************************************
%net=newff([0 1;0 1],[S1,S2],{'logsig','logsig'},'traingd');
%****************************************************************

% initialize the network with single command
% function newff for Version 6 Toolbox: enclosed in ***********
%*********************************************************************
net=newff(P,T,S1,{'logsig','logsig'},'traingd');
net.divideFcn = '';                 % don't divide input data into test and validation data
net.inputs{1}.processFcns={};       % don't perform advanced preprocessing
net.outputs{2}.processFcns={};      % don't perform advanced postprocessing
%net.trainParam.showWindow = false; % this will disable TRAIN GUI
%net.trainParam.showCommandLine = true; % this will show training status in command window
%**********************************************************************


% override the initialized weight/biases from newff to match HW #13
net.IW{1}=[-6.8 2.8; 5.9 -5.1];     % input to hidden layer weights
W10=net.IW{1};						% save copy
net.b{1}=[5.3 -0.2]';				% hidden neuron biases
B10=net.b{1};						% save copy
net.LW{2}=[-2.5 0.8];				% hidden to output layer weights
W20=net.LW{2};						% save copy
net.b{2}=[1.4];						% output neuron bias
B20=net.b{2};						% save copy

% TRAINING PARAMETERS
net.trainParam.show=50;
net.trainParam.lr=0.5;
net.trainParam.epochs=1000;
net.trainParam.goal=0.02;
net.performFcn='sse';

% Now train the ANN
[net,tr]=train(net,P,T);

% SUMMARIZE RESULTS
%==================
fprintf('\nINITIAL NETWORK VALUES:\n')
W10
B10
W20
B20

fprintf('\nFINAL NETWORK VALUES:\n')
W1=net.IW{1}	% input to hidden layer weights
B1=net.b{1}		% hidden neuron biases
W2=net.LW{2}	% hidden to output layer weights
B2=net.b{2}		% output neuron bias

fprintf('\nNeural Network Outputs:\n')
A=sim(net,P)

fprintf('Trained for %.0f epochs.\n',max(tr.epoch))
fprintf('Sum squared error goal was %g.\n',net.trainParam.goal)
fprintf('Final sum squared error is %g.\n',tr.perf(max(tr.epoch+1)));
fprintf('Trained network operates: ');
if tr.perf(max(tr.epoch+1))< net.trainParam.goal
  disp('It works!')
else
  disp('It Failed!')
  end