I'm attempting to code a NEAT algorithm from scratch using C#. I'm very early in the process and I'm now going about writing a method for randomly initializing a network. I want to use the mutations detailed in this video and apply a certain amount of these mutations until a good starting neural network is created.
As I thought about it, I realized that if I just do this randomly there's the possibility of circular dependency. Is there a better way of initializing and mutating these neural networks or will I have to implement a dependency graph as well to ensure no circular dependencies occur? That seems like it could end up being pretty computationally slow, which I'd like to avoid as much as possible.
Any material explaining in detail how these neural networks are initialized and mutated properly would be much appreciated.
