% First a set of some funny data, sin(x) is quite nice
p = [0:0.1:2*pi]; % input
t = sin(p); % output

% Neural net with default values of gradient descent 
net1 = newff([0 2*pi],[3 1],{'logsig' 'purelin'},'trainscg'); 
net1.trainParam.epochs = 150; % these are smaller just for the demo
net1.performFcn = 'msereg';
net1.performParam.ratio = 0.995;
net1.trainParam.show = 25;
tic
net1 = train(net1,p,t);
toc

% Neural net with DE learning
net2 = newff([0 2*pi],[3 1],{'logsig' 'purelin'},'traindiffevol'); 
net2.trainParam.epochs = 150; % these are smaller just for the demo
net2.trainParam.show = 25;
net2.performFcn = 'msereg';
net2.performParam.ratio = 0.995;
net2.trainParam.popsizew = 1;
%net2.trainParam.initw = 100;
net2.trainParam.cr = 0.5;
net2.trainParam.f = 0.8;
tic
[net2, tr2] = train(net2, p, t);
toc

% Neural net with default values of gradient descent 
net3 = newff([0 2*pi],[3 1],{'logsig' 'purelin'},'trainscg'); 
net3.trainParam.epochs = 100; % these are smaller just for the demo
net3.trainParam.show = 25;
net3.performFcn = 'msereg';
net3.performParam.ratio = 0.999;
net3 = setx(net3,getx(net2));
tic
net3 = train(net3,p,t);
toc

% Test nets and plot the original and modeled data
test1 = sim(net1, p);
test2 = sim(net2, p);
test3 = sim(net3, p);

figure;
set(plot(p, t, 'k-'), 'LineWidth', 2); % original
hold;
set(plot(p, test1, 'r--'),'LineWidth', 2);
set(plot(p, test2, 'g.'), 'LineWidth', 2); % DE learning
set(plot(p, test3, 'b-.'), 'LineWidth', 2);
grid;
legend('sin(x)', 'trainscg', 'traindiffevol', 'traindiffevol + trainscg');
set(xlabel('x'),'FontSize', 12);
set(ylabel('y'),'FontSize', 12);