For training, what did you use for determining reward?

[NullVoxPopuli]

I'm working on a similar project in my free time, and am curious on much info should go into the reward function, how heavy to weight certain actions or failures, etc