It is my first time with neural network Kohonen.

I try predict time series price copper with neural network Kohonen. In file testsom.m is erroron wzor=[wzor,predc]( The following error occurred converting from double to struct: Error using struct Conversion to struct from double is not posiible.

function result=testsom(szereg,pocz,p,ile)

%p = 10; %pocz = 1140; kon = pocz+p*ile; wy = []; good = []; bladd = 0; bladg = 0; cp = 8;

sp = 10; sk = 5;

rozxx = 5; rozxy = 5; rozxy = 5; rozyy = 5; wylicz = szereg(1:pocz);

for ii=pocz:p:(kon-p)

    wzor = szereg(1:ii);
    model = start(wzor,sk,sp,rozxx,rozxy,rozxy,rozyy);    

    od = ii;
    do = ii + p;

    for i=od:do-1

        w = odp(model,wzor(i-sp+1:i-1),1,wzor(i-1)); % predykcja wartosci szereg(i)
        predc = w; % nowa wyliczona wartosc szeregu czasowego
        wzor = [wzor , predc];
        wylicz = [wylicz, predc];  
    end

end

% liczenie bledu

srednia = mean(szereg(1:kon)); for i=pocz:kon bladg =bladg + abs(wylicz(i) - szereg(i)); bladd = bladd + abs(szereg(i)- srednia); end

blad = bladg/bladd;

plot(wylicz,'r'); hold on; plot(szereg);

result = blad;

Answer 1

[Greg Heath]

Kohonen is not appropriate for time series prediction.

If you have input and target series

 help narxnet
 doc narxnet

If you only have one series

 help narnet
 doc narnet

You can practice on MATLAB data

 help nndatasets
 doc nndatasets

You can search the NEWSGROUP and ANSWERS for some of my examples

 greg narxnet

or

 greg narnet

Hope this helps.

Thank you for formally accepting my answer

Greg