I am working on a project which uses a control algorithm to complete a task. The project already has a simulator created for it. I would like to experiment and see if a NEAT AI could do a better job and replace the current control algorithm. I've read through the NEAT publication Competitive Coevolution through Evolutionary Complexification. I have also looked at the original c++ implementation.
I was hoping it would be possible to use and existing implementation of NEAT and combine it with the existing simulator in order to train an awesome control algorithm. There is not much documentation on how i could modify the original NEAT C++ for my needs, are there any tutorials or other implementation which will suit my needs?
Thanks
Answer is little late, but maybe it will help someone else.
I do not know if you read initial Evolving Neural Networks through Augmenting Topologies, but there is explained. About code, only documentation is that one in package that you referenced.
I also found that book "AI Techniques for Game Programming" by Mat Buckland had chapter dedicated to NEAT algorithm with NEAT code (different from original referenced), but maybe it could give you some insight into code.
UPDATE: Most easiest way is to check out experiments included in code. From there you can see that you just call NEAT algorithm and define function to calculate fitness.