Well, I am not good in analysing other peoples code,
but the GA algorythm got me thinking of
a Testprogram to link this approach with simple Neural Nets,
without using (the unnatural) Backpropagation.

The Neural Nets themselve dont learn. They are Preprogrammed and wired,
much like in simple Insects. The leaning takes place by the genetic mutation
and selection in the genes that build up, wire and weight the neuron.

Seems much more natural to me than some "Big Brother"
that used Sigma Functions, and mathematical propagation Methods.