Playout Policy Adaptation for Games

Monte Carlo Tree Search (MCTS) is the state of the art algorithm for General Game Playing (GGP). We propose to learn a playout policy online so as to improve MCTS for GGP. We test the resulting algorithm named Playout Policy Adaptation (PPA) on Atarigo, Breakthrough, Misere Breakthrough, Domineering, Misere Domineering, Go, Knightthrough, Misere Knightthrough, Nogo and Misere Nogo. For most of these games, PPA is better than UCT with a uniform random playout policy, with the notable exceptions of Go and Nogo.