weird plotregression plots for a 10% of my fitnet neural networks

Dear comunity,

i trained a fitnet with a inputmatrix (7x2118 double) and a targetvector (1x2118 double) like this:

hiddenLayerSize = [105 15];   
trainFcn = 'trainlm';        
net = fitnet(hiddenLayerSize,trainFcn);
for netI=1:100
        [TrainedNet(netI).net,trainRecords(netI).tr] = train(net,inputmatrix,targetvector);
end

for 9 of my 100 ANNs i got a really strange result in the training.

Have you ever seen this kind of strips within a network output? Why dos it happen? What could be an explanation?

Kind regards, MLP

Answer 1

[Greg Heath]

Sometimes training gets into a parameter space rut. That is why it is wise to train multiple models.

Hope this helps.

Greg

THANK YOU FOR FORMALLY ACCEPTING MY ANSWER

Greg