Practical Neural Network Recipies in C++

Autoassociation

37

Let us take a common case as our example. Suppose that the

event is well defined, both in shape and in duration. In order to

sample every border condition, as well as the complete event, we must

prepend and append nonpattern data of the same length as the event.

Then, the length of each training sample would be equal to the length

of the event. The first sample is taken from the beginning of the

extended series and is entirely nonpattern. The second sample is one

to the right and includes the leftmost part of the pattern. This is

repeated until the last sample is taken, entirely past the event. For

an image, this would b