Constructing Pin Endgame Databases for the Backgammon Variant Plakoto

摘要

Palamedes is an ongoing project for building expert playing bots that can play backgammon variants. Until recently the position evaluation relied only on self-trained neural networks. This paper describes the first attempt to augment Palamedes by constructing databases for certain endgame positions for the backgammon variant of Plakoto. The result is 5 databases containing 12,480,720 records in total; they can calculate accurately the best move for roughly 3.4 × 10~(15) positions. To the best of our knowledge, this is the first time that an endgame database is created for this game.