ABSTRACT
Monte Carlo Tree Search (MCTS) is a new and increasingly popular technique that has shown it is capable of dealing with more complex games. This chapter outlines solutions to common pitfalls when using MCTS and how to adapt the algorithm to work with strategy games. The strength of MCTS lies in its ability to cope with large, complex problems in which the long-term ramifications of a move are generally more important than its short-term effects. In other words, in problems where the actual effects only become apparent after executing long sequences of moves, that is, a deep search space. By default, at each expansion step MCTS expands into all possible moves from the current node. The problem with this type of expansion is that there is no transfer of information between moves at the same level. Using default expansion, only complete moves are explored in MCTS.
