Why not train neural network newff?

Question

Veronika 2015-3-14

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/183178-why-not-train-neural-network-newff

评论： Veronika 2015-3-16

Dear all,

I have this code for training neural network (type newff) and MATLAB is almost all day busy, but nothing happens (no training newff). Does anyone know, where is the mistake?

load ('trenovaci_modely2_stredni')
D = [stredni_tvar{1,:}];
F = [stredni_tvar{2,:}];
net_stredni_MLP = newff(D,F,20);
net.trainparam.epochs = 100;
net = train(net_stredni_MLP,D,F);
D = stredni_tvar{1,7};
J = sim(net_stredni_MLP,D);
save net_stredni_MLP
disp('Neuronová síť byla uložena')

Thank you for your answers.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Greg Heath 2015-3-15

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/183178-why-not-train-neural-network-newff#answer_171383

编辑：Greg Heath 2015-3-15

在 MATLAB Online 中打开

 load ('trenovaci_modely2_stredni')
 D = [stredni_tvar{1,:}];
 F = [stredni_tvar{2,:}];
 [ I N ] = size(D)   % = ?
 [ O N ] = size(F)   % = ?
 net_stredni_MLP = newff(D,F,20);
 net.trainparam.epochs = 100;
 net = train(net_stredni_MLP,D,F);

% The trained model is net, NOT net_stredni_MLP!

% D = stredni_tvar{1,7}; DELETE

D7 = D(1,7) % WHY JUST ONE POINT??

% J = sim(net_stredni_MLP,D); DELETE

% save net_stredni_MLP DELETE

 J7 = sim( net, D7); % Only 1 point
 J  = sim( net, D );    % N points
 save net
 disp('Neuronová síť byla uložena')

Hope this helps.

Thank you for formally accepting my answer

Greg

3 个评论
显示 1更早的评论隐藏 1更早的评论

Greg Heath 2015-3-16

在 MATLAB Online 中打开

Your workspace dimensions do not make sense. If you have one O-dimensional target vector for each I-dimensional input vector, then at least one of the dimensions of the input matrix and target matrix has to be the same. Then, transposing if you have to

 [ I N ] = size(input)   
 [ O N ] = size(target)

What is

size(stredni_tvar)?

Veronika 2015-3-16

在 MATLAB Online 中打开

I know, but I don´t know what´s wrong. Because when I used different images in EIDORS (doesn´t matter) in workspace:

input D <208x301 double> and target F <64x301 double>.

Command Window returns:

[I N] = size(D)
I =
     208
N =
     301
[O N] = size(F)
O =
      64
N =
     301
 So the neural network trains good. I think, that I did it exactly the same way. So I don´t know, where is mistake. Regarding size of stredni_tvar:
size (stredni_tvar)
ans =
       2   301