Training ANN and gaining RMSE values

Hi all,

I am trying to create and train an ANN to predict power output of a wind turbine based on some inputs. I have followed a helpful video on youtube on how to create the ANN and return RMSE values for the training, validation and test cycles of the ANN. The problem I am having is that when I run the code, one of my RMSE values comes back as a NaN. The data used is a numeric Matrix obtained from a simulation and contains no NaN's at all. What is more problematic is that if I run the code 3 times consecutivly, each time a differant RMSE value returns the NaN. Please see code below:

%data

x=winddata(:,[2 5 17 20]);

y=winddata(:,22);

m=length(y);

%Train ANN

xt=x';

yt=y';

hiddenlayersize=10;

net=fitnet(hiddenlayersize);

net.divideParam.trainRatio=30/100;

net.divideParam.ValRatio=30/100;

net.divideParam.testRatio=40/100;

[net,tr]=train(net, xt, yt);

yTrain=net(xt(:,tr.trainInd));

yTrainTrue=yt(tr.trainInd);

rmse_train=sqrt(mean((yTrain-yTrainTrue).^2))

yVal=net(xt(:,tr.valInd));

yValTrue=yt(tr.valInd);

rmse_val=sqrt(mean((yVal-yValTrue).^2))

yTest=net(xt(:,tr.testInd));

yTestTrue=yt(tr.testInd);

rmse_test=sqrt(mean((yTest-yTestTrue).^2))

The results obtained from running the ANN 3 times consecutivly can be seen below:

>> ANNcheck1

rmse_train = NaN

rmse_val = 292.1152

rmse_test = 290.5616

>> ANNcheck2

rmse_train = 290.6516

rmse_val = 290.7914

rmse_test = NaN

>> ANNcheck3

rmse_train = NaN

rmse_val = 289.9657

rmse_test = 293.0484

Any help would be greatly appreciated!

Answer 1

[Matt Egan]

I have found the error in the code! It was to do with the transposition from of the matricies, this is what I had:

xt=x';

yt=y';

And this is what it should be:

xt=x.';

yt=y.';