Neural network (NN)

Hi. I have problem to train my Neural Network. This is the coding.

clear
clc

load c.mat
load OUTPUT1.mat

inputrain = c(1:1000,:);
targetrain = OUTPUT1(1:1000);


[pn,mininputrain,maxinputrain,tn,mintargetrain,maxtargetrain] = premnmx(inputrain,targetrain);

net=newff(minmax(pn),[20,10,1],{'logsig','logsig','purelin'},'trainlm');
net=init(net);
net.trainParam.show = 1; 
net.trainParam.lr = 0.9; 
net.trainParam.mc = 0.9; 
net.trainParam.epochs = 1000; 
net.trainParam.goal = 1e-3;

net = train(net,pn,tn);
an = sim(net,pn);
[a] = postmnmx(an,mintargetrain,maxtargetrain);

result_norm = [an' tn'];
result_denorm = [a' targetrain'];
error = mse(tn-an);
mape = mean(abs(error./tn));
rmse = sqrt(mean((error - tn).^2));
[m,b,r] = postreg(a,targetrain);

save net.mat net

The error said " Output data size does not match net.outputs{3}.size." Help me please. Thank you

[KSSV]

Attach your data/ mat files.

[Kerrollenna Martta]

Meaning? 😅

Answer 1

[Walter Roberson]

net=newff(minmax(pn),[20,10,1],{'logsig','logsig','purelin'},'trainlm');

You are missing a parameter. The {'logsig'} and so on vector must be the 4th parameter. The first three parameters must be P, T, S where P is input vectors, T is target vectors, and S is sizes of the input layers. S can be [] if you want it to be calculated based upon the sizes of the entries in T.

[Kerrollenna Martta]

Do you have any idea on how can I change it?

[Walter Roberson]

premnmx was obsoleted in R2006a, so it is difficult to find documentation for it.

I suspect that maybe you should use

net=newff(pn, tn, [20,10,1], {'logsig','logsig','purelin'}, 'trainlm');

but newff() was obsoleted in R2010b . You should probably modernize the code.