Learning for the Value of Moves by Iteration of Generation of Decision Tree in Go

スポンサーリンク

概要

• 論文の詳細を見る

• It is necessary in a game "Go" to limit candidates of moves for searching because the total number of possible moves of a position exceeds 200 on an average. We performed the generation of candidate moves by the decision tree in the past. We got the features around an empty point which could be selected (legal move), using not a pattern but the information about the stone which was in the circumference of selected moves or the coordinates of selected moves from a professional Go player's record. In this paper, first, we generate the decision tree, collect data using generated decision trees, then generate new decision trees. A decision tree is improved by repeating. An evaluation of our method is shown correspondence between our system's optimal candidates and actual moves made by professional Go players. As a result, the number of candidate moves until actual moves appeared was an average of 21.96.

• 一般社団法人情報処理学会の論文
• 2002-07-24

著者

• Kotani Yoshiyuki

  Tokyo University Of Agriculture And Technology

• ABE Nobuharu

  Tokyo University of Agriculture and Technology

関連論文

• Learning for the Value of Moves by Iteration of Generation of Decision Tree in Go

スポンサーリンク