Practical Neural Network Recipies in C++

64

Chapter 4

Measuring Prediction Error

Suppose that we have trained a neural network to do time-series

prediction. We have a test series with which to evaluate the perfor-

mance of the trained network. Using groups of known points from the

test series, we make predictions. For each prediction, we know the

value of the series as predicted by the network, and we know the true

value of the series. The problem now is to express the performance of

the network in a useful manner.

Most neural network training algorithms work by minimizing

the mean square error of the output(s). Naturally, during training this

is done in terms of the scaled data tha ...