Select matrix for training,testing and validation on ANN

Hello to all,

completeX = [1:+1:100;2:+2:200]';
completeY = [1:+1:100]';

From my data set i divided in a specific form and I got a Xtrain (72x2), Xcv (8x2) and Xtest (20x2)

i would like to tell the net which matrix is the training, validation and testing, instead of matlab performing the random spliting, is that possible?

net = fitnet(10,'trainlm');
net.divideParam.train = Xtrain;
net.divideParam.val = Xcv;
net.divideParam.test = Xtest;
[net, TR] = train(net,completeX',completeY');

Hope it was clear,

Thanks!

Answer 1

[Raunak Gupta]

Hi,

You may try dividing the whole dataset based on the indices as understandable from the question. Below code may help.

completeX = [1:+1:100;2:+2:200]'; 
completeY = [1:+1:100]'; 
  
net = fitnet(10,'trainlm'); 
net.divideFcn = 'divideind'; 
net.divideParam.trainInd  = 1:72; 
net.divideParam.valInd    = 73:80; 
net.divideParam.testInd   = 81:100; 
[net, TR] = train(net,completeX',completeY'); 

TR.trainInd , TR.valInd , TR.testInd will give the indices of training , validation and test data which can be used to find performance of the network. You may manipulate above indices vector as required.

[Tiago Dias]

Thanks, but i did a workaround.

I forced the validation and testing to be zero. So that i could use the train function only for the training set, then i validate with the testing set, in order to choose the right number of neurons for the hidden layer. Then i tested with the testing set.

What you wrote it was a good idea as well. since I got the 3 matrixes, i could merge them, and in that matter the first rows are training, then validation and testing, didnt think of that thanks!