clc
NNTWARN OFF % closr all warnings of neural network

% expecting output
T=[481 463 447 471 459 444 440 415 420 474 493 508 512 541 547 548 545 535 531 535 524 568 542 526];
T = T -mean(T);
T = T / norm(T);
% input
P=1:24;
%P = P - mean(P);
%P = P / max(abs(P));



%size of input layer
[size1,temp]=size(P);

%size of output layer
[size3,temp]=size(T);

% statistics method: regression
UseS= 0
if UseS
	n = 12;
	p=polyfit(P,T,n);
	plot(P,T,'+',P,polyval(p,P));
   pause
end

UseNN=1
if (UseNN)

	%size of hidden layer
	size2 = 8;

	%initial value
%  [W1,B1]=nwtan(size2,size1)	
   [W1,B1]=rands(size2,size1)	
   [W2,B2]=rands(size3,size2);
%  [W2,B2]=nwtan(size3,size2)
%  [W1,B1,W2,B2]=initff(P,size2,'logsig',size3,'logsig');


	disp_fqre = 200; %display frequence
	max_epoch = 10000;
	err_goal = 0.05; % error goal
	lr = 0.01; % learning rate

	TP=[disp_fqre max_epoch err_goal lr];
%	[W1,B1,W2,B2,epochs,error] = trainbpx(W1,B1,'logsig',W2,B2,'logsig',P,T,TP);
	[W1,B1,W2,B2,epochs,error] = trainbpx(W1,B1,'tansig',W2,B2,'tansig',P,T,TP);
%	[W1,B1,W2,B2,epochs,error] = trainbpx(W1,B1,'tansig',W2,B2,'purelin',P,T,TP);

	%test
	% x is test input
	%A1 = tansig(W1*x,B1);
	%A2 is output
	%A2 = purelin(W2*A1,B2);
   %draw error
   figure
	ploterr(error);
end