Bayesian pattern ranking for move prediction in the game of Go
ICML '06 Proceedings of the 23rd international conference on Machine learning
Combining online and offline knowledge in UCT
Proceedings of the 24th international conference on Machine learning
Reinforcement learning of local shape in the game of go
IJCAI'07 Proceedings of the 20th international joint conference on Artifical intelligence
Bandit based monte-carlo planning
ECML'06 Proceedings of the 17th European conference on Machine Learning
Hi-index | 0.00 |
3×3 patterns are widely used in Monte-Carlo (MC) Go programs to improve the performance. In this paper, we propose a direct indexing approach to build and use a complete 3×3 pattern library. The contents of the immediate 8 neighboring positions of a board point are coded into a 16-bit string, called surrounding index. The surrounding indices of all board points can be updated incrementally in an efficient way. We propose an effective method to learn the pattern weights from forty thousand professional games. The method converges faster and performs equally well or better than the method of computing "Elo ratings" [4]. The knowledge contained in the pattern library can be efficiently applied to the MC simulations and to the growth of MC search tree. Testing results showed that our method increased the winning rates of Go Intellectagainst GNU Goon 9×9 games by over 7% taking the tax on the program speed into consideration.