HISTORY AND TERRITORY HEURISTICS FOR MONTE CARLO GO

Abstract

Recently, the Monte Carlo approach has been applied to computer go with promising success. INDIGO uses such an approach which can be enhanced with specific heuristics. This paper assesses two heuristics within the 19 × 19 Monte Carlo go framework of INDIGO: the territory heuristic and the history heuristic, both in their internal and external versions. The external territory heuristic is more effective, leading to a 40-point improvement on 19 × 19 boards. The external history heuristic brings about a 10-point improvement. The internal territory heuristic yields a few points improvement, and the internal history heuristic has already been assessed on 19 × 19 boards in previous publications. Most of these heuristics were used by INDIGO at the 2004 Computer Olympiad.