Can calculate AAPE values for my training data, validation data and testing data from my trained network as part of the process, like calculation of performance?

I have used a feedforward backprop to train a network. I realise some authors calculate Absolute Average Percent Error (AAPE) values for test data, validation data and testing data sets respectively.

1. Is there a code to calculate these AAPE values just like there is to calculate performance or i have to extract the values for the mentioned data sets and calculate manually?

2.If I have to calculate manually, how do I extract these data sets?

Thank you

Answer 1

[KAE]

If you train a network as follows,

[net, tr, y, e] = train(net,x,t)

then the tr.trainInd, tr.valInd, and tr.testInd fields allow you to index the datapoints used for training, validation and testing. For example,

tTest = t(:, tr.testInd);

allows you to extract the target values used for testing. You can then calculate any statistic you like.