10-23-2013, 09:48 PM
(10-23-2013, 09:04 PM)ymko Wrote: i was thinking to write a bot that will basically think like a chess player, 5 steps ahead and adjust accordingly
My goal is to write an AI that does something similar and most of the work is in place for it; just copy the synth object, run the simulation N steps with a particular strat, then score it, keeping the N-step-plan that had the best score.
The trick is coming up with a utility/scoring function that can evaluate how good a particular situation is accurately. Obviously it should get 0 points if completion is no longer possible and points for quality, but it's hard to put actual numbers on how good the fact that some buff is up or your control is increased...