A Multi-modal Searching Algorithm in Computer Go Based on Test

摘要

Pattern matching algorithms based on domain knowledge have a weak global sense. Monte Carlo algorithms have a weak tactical ability. This article proposed a multi-modal algorithm on the basis of tests. GNU Go software was used to make experiments. Through the analysis of the experiment data, we set the move thresholds of start, middle, and the end of a Go play game. In the start stage, pattern matching algorithm was used. In the middle stage, Monte Carlo tree search algorithm with pruning was used to search the optimal moves. In the end stage, pattern matching and crazy model were used. The multi-modal algorithm was applied to develop software. Experiments demonstrate that multi-modal algorithm can improve the performance of Go in 13 × 13 board and 19 × 19 board.