Monte-Carlo tree search with Epsilon-Greedy for game of amazons

Abstract

The Game of the Amazons is an abstract strategy board game. It has a high computational complexity similar to the game of Go. Due to its NP-complete nature and large branching factor of game tree, finding the optimal move given a specific game state is infeasible and it is not trivial to design a computer algorithm that is competitive to an expert in the game of amazons. One way to tackle this problem is to leverage the Monte-Carlo Tree Search by using random simulations. In this article, a computationally cheap heuristic function is proposed and use together with Monte-Carlo Tree Search algorithm with Epsilon-Greedy policy aiming to design a competitive AI for the Game of the Amazon. The effectiveness of the -greedy based Monte-Carlo algorithm is compared to the widely used MCTS with Upper Confidence Bound and other classical tree search method such as breadth-first search, depth-first search, minmax search and alpha-beta pruning.