Windows Crashing during NN training

Hello folks,

I am writing here because I have a problem with the Neural Network toolbox: during training I have had several crashes of Windows (BSOD). It also just not crash, it freezes on the "sad-smile blue screen" and I have to force the restart manually.

Apparently, the problem regards the for loop that I am doing to test the performance of the network depending on its hidden layer. I am doing something like this...

for h = H

    for k = 1:K

        % K-Fold Cross-Validation (K=10)
        % ANN definition
        net = fitnet(h,'trainbr');
        net = configure(net,inputs',targets');
        net.layers{1}.transferFcn = 'tansig';
        net.divideParam.trainRatio =1;
        net.divideParam.valRatio = 0;
        net.divideParam.testRatio = 0;
        net.performFcn = 'mse';
        disp('     Training...')
        % ANN training
        [net,tr] = train(net,inputs',targets');       

        % Performance evaluation, MSE.
        % ...

    end

end

What's wrong? TIA

[Greg Heath]

Size of inputs, targets, H and K?

Does the code work on the MATLAB EXAMPLE datasets?

 help nndatasets

 doc nndatasets

[mp1994]

H = 5:12

K = 10

I have checked the code and it works... I have also tried to reduce H (going up to 11), but it crashed anyways... It works quite well instead if I remove the for loop, but I haven't tested for all the values of H... I want the code to do that, as it is supposed to do... The input and target datasets are randomly extracted from a vector of 38420 rows. I am working on a desktop PC, Win10, i7-2600k (at 4.5 GHz), 16 GB RAM (not even at 30%)... CPU temperature always below 60°C...

Answer 1

[Petorr]

Try closing the figure. I am getting a matlab crash (win7, r2017b) on the next train() call if I don't close the previous training progress figure. For some reason the figure is hidden (see here) so it doesn't turn up with gcf(), but you can get the handle with:

findall(0,'Type','Figure','Tag','NNET_CNN_TRAININGPLOT_FIGURE');

- Peter

[mp1994]

Thank you for your answer, but I fixed the issue back in the days improving the stability of my overclock... apparently, the problem was that (low Vcore, if I remember correctly).