Prev: voronoi cells in bounded surface or volume
Next: Fast Matrix Exponential
From: Christine on 3 May 2010 07:01
I have simplify the actual problem. I am using narxsp to train my problem. Whenever I use d1=[0 1] and d2=[0 1], there is an error occurred when executing the conversion from narxsp to the parallel form. It says, "network contains a zero-delay loop".

If I want to change my d1 and d2 (delay) value. How should I represent the data? If I just change the d1 and d2, the training result is not good.

Besides, can explain d1 and d2? I just know that it is a delay loop. But, How does it actually functions?

%---------------------------------------------------------------------------
P = [0 1 1 0 -1 -1 0 1 1 0 -1];
T = [0 1 2 2 1 0 1 2 1 0 1];
PT = [P;T];
d1=[0 1];
d2=[0 1];
net = newnarxsp(P,T,d1,d2,5,{'tansig','purelin'});

[net,tr,Y,e] = train(net,PT,T);

net = sp2narx(net); % convert the narxsp to parallel form
Y = sim(net,P);
%---------------------------------------------------------------------------

The Error:
??? Error using ==> network.sim at 169
Network contains a zero-delay loop.
From: Christine on 3 May 2010 08:23
I realise that there is no error if I changed d1 and d2 equal to [1 2]. But, I want to use d1 and d2 equal to [0 1].

Why can't I convert without any error using d1 = d2 = [0 1] ?
%---------------------------------------------------------------------------
% No error
P = [0 1 1 0 -1 -1 0 1 1 0 -1];
T = [0 1 2 2 1 0 1 2 1 0 1];
PT = [P;T];
d1=[1 2];
d2=[1 2];
net = newnarxsp(P,T,d1,d2,5,{'tansig','purelin'});

[net,tr,Y,e] = train(net,PT,T);

net = sp2narx(net); % convert the narxsp to parallel form
Y = sim(net,P);
%---------------------------------------------------------------------------
From: Christine on 3 May 2010 12:40
help pls.
 | 
Pages: 1
Prev: voronoi cells in bounded surface or volume
Next: Fast Matrix Exponential