%~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
%
%  Data fitting DEMO of neural networks with matrix inputs.
%
%  Author: Povilas Daniu餴s, paralax@hacker.lt
%  http://ai.hacker.lt - lithuanian site about Artificial Intelligence.
%
%  TODO: weighted MSE minimization, maximal likelihood method, multiple
%  activation function support.
%  ----------------------------------------------------------------------

clear all


alpha = 0.9;       % inertia
eta = 0.005;       % inital learning rate
epsilon = 0.03;    % needed MSE
epsilon1 = 0.001;  % minimal descent (stopping criteria) - all iterations in this case
neurones = 12;    % su 20 neveikia :) 
n = 3;
numEpochs = 30;
earlyStop = 5;      
[a,D] = textread('c:\sunspot1947-1991.txt','%s %f');
%D = load('c:\laser1.txt');
D = (D - mean(D))/std(D);
point = 200;


for i=1:length(D) - 10
    if (i <= point)
        data.training(i).mat = [ D(i+1), D(i+2) D(i+3); D(i+4),  D(i+5), D(i+6); D(i+7),  D(i+8), D(i+9); ];
        data.target(i) = D(i+10);
    else
        data1.training(i-point).mat = [ D(i+1), D(i+2) D(i+3); D(i+4),  D(i+5), D(i+6); D(i+7),  D(i+8), D(i+9); ];
        data1.target(i-point) = D(i+10);
    end
end

e = mNN_device(neurones,size(data.training(1).mat),alpha,eta,epsilon,epsilon1,earlyStop);
e_elm = ELM_train(e,data); 

s_elml = mNN_sim(e_elm,data);
s_elm = mNN_sim(e_elm,data1);



plot(data1.target,'r-'); hold on; plot(s_elm,'b-');

mse_learn = sum((data.target - s_elml).^2) / length(data.target)
mse_test = sum((data1.target - s_elm).^2) / length(data1.target)

%hit_rate = sum(data.target .* s_elm > 0) / length(data.target)