強化學習（十一）：mcts

Introduction

• Game Tree: use tree to represent the game
• Minimax: a strategy to make decision based on game tree
• Negamax: equivalent to Minimax
• Alpha–beta pruning: a search algorithm that minimize the search space
• Monte Carlo tree search: a heuristic search algorithm that converges to Minimax

Project

mcts is based on the project Reversi ( @林小囧 ).

• reversi.py: the code of the Reversi game
• search1.py: mcts with random policy as defaul policy
• search2.py: mcts with custom policy as default policy
• demo.py: search1.py vs search2.py

Summary

This article introduces mcts. The code is here.

Reference

Paper

• A Survey of Monte Carlo Tree Search Methods
• Monte-Carlo Tree Search
• Mastering the Game of Go with Deep Neural Networks and Tree Search
• Mastering the game of Go without human knowledge

Article

• 28 天自制你的 AlphaGo (6) : 蒙特卡洛樹搜索（MCTS）基礎
• AlphaGo論文的譯文，用深度神經網路和樹搜索征服圍棋：Mastering the game of Go with deep neural networks and tree search
• DeepMind新一代圍棋程序AlphaGo Zero再次登上Nature
• DeepMind 研發的圍棋 AI AlphaGo 是如何下棋的？
• 深度解讀 AlphaGo 演算法原理

Code

• Python Code
• Introduction to Monte Carlo Tree Search

推薦閱讀：