%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% CSC411: Machine Learning and Data Mining 
%     Tutorial 4 (Febuary 9th): Neural Network Toolbox 
% TA: Rui Yan
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

clear all

%Step 1: generate training dataset
x=0:.05:2;
y=humps(x);

P=x;
T=y;

plot(P,T,'+');

%Step 2: define the structure of the network
%- choose the network structure
%- choose activation functions
%- initialize network paramters: weights and bias

%net = newff(PR,[S1 S2....SNL],{TF1 TF2....TFNL},BTF)
%PR: min and max values for R input elements
%Si: size of the ith layer
%TFi: avtivation(or transfer function) of the ith layer, default is
%'transig'
%BTF: network training function, default is 'trainlm'
net = newff([0 2], [10,1],{'tansig', 'purelin'},'trainlm');

%Step 3: Define parameters associated with the training algorithms
%-error goal, maximum number of epochs(iterations)

net.trainParam.epochs = 1500;
net.trainParam.goal =1e-5;
net.trainParam.show=5;

%Step 4: Call the training algorithm
%net=train(net,P,T)
%net: initial MLPN (multiple perceptron network)
%P: input vector
%T: output vector
net=train(net,P, T);

pause(5);
%Step 5: Simulate the output of the neural network with the measured input
%data and Compare with the validaiton data

%a = sim(net, P);
%net: final MLPN
%P: input vector
a=sim(net,P);
 
plot(P,a-P,'-',P,T,'+', P, a,'o');