Skip to content

A Go A.I. based on MCTS WITHOUT Neural Networks

License

Notifications You must be signed in to change notification settings

Zadranzadran123/FoolGo

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

97 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

FoolGo is a Go A.I. based on MCTS (Monte Carlo Tree Search).

It shows the beginner level on the 9x9 board currently, for it does not use neural networks, but only the simple UCT (UCB for Tree) algorithm and Monte Carlo simulation.

The underlying data structure and algorithms are quite efficient. Specifically, it runs 40 thousand Monte Carlo games per second on my MacBook Air 2014.

The key ideas of the Go game's implementation are as follows:

  • We implement a string of connected pieces using the disjoint set to update piece strings during the game process efficiently.
  • We use bit operations to update the liberty point number of piece strings efficiently. For example, we use or when merging several piece strings into a single one.
  • We do not implement a tree structure for tree search, but use a hash table to keep game states. Especially, we use Zobrist Hashing to compute the hash value of a game state incrementally.

For MCTS, we use multi-thread search on CPU. In detail, we simply prohibit a thread from exploring nodes that are being explored by peer threads.

This project is highly readable and extensible due to its well-designed OOP design. Some important classes are listed as follows:

  • FullBoard A game board class providing efficient methods such as PlayMove to update the game state.
  • Player An interface providing inheritable methods such as NextMove. For example, RandomPlayer, a Player subclass, will return a random legal position of the next move when called its NextMove, while UCTPlayer performs the UCT algorithm to find the next move.
  • Game An interface containing two Player objects. For example, MonteCarloGame contains two RandomPlayer objects.

Make

  • boost is required
  • mkdir build && cd build
  • cmake ..
  • make

Contact

About

A Go A.I. based on MCTS WITHOUT Neural Networks

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C++ 98.5%
  • CMake 1.3%
  • C 0.2%