forum.alglib.net

Hi, I'm trying to determine the generalization error of different neural networks in the excellent ALGLIB library on a set of training data. Alglib includes handy functions that perform cross-validation and return a mlpcvreport object containing the generalization error or at least a set of numbers. My question relates to what error is represented by each variable contained in the object:

I've included my guesses above. The first two especially are a mystery to me.

Thanks!

relclserror is a relative classification error - number of incorrectly classified cases divided by total number of cases.

avgce is average cross-entropy (measured in bits per sample, see Wikipedia for more information about cross-entropy)

other guesses are correct :)