Implementing Feedforward - GA

Hello!
The book "Introduction To Neural Networks for C#" was a great help for me. I want to write a Bot for a dynamic game. The AI will be controlled by a Feedfoward Network which is trained by the Genetic Algorithm. However, I have some difficulties using this classes. I know which inputs to take, and which outputs to retrieve, but I dont know how I can implement the GA and how I can train the network with it.

Every chromosome should be tested in a game untill it gets killed, or until it manages to kill the player (After some generations I will let two Bots fighting each other). Depending on the length of the game, their fitness will be tested to eventually mate them afterwards - but how can i do that? I have no ideal trainingset, and my error calculation function is the whole game, in which the AI tries to survive. I would be grateful if you could help me.