Rabbocop is Arimaa bot. For more information about Arimaa see http://arimaa.com
The goal of this project is to compare AlphaBeta search with Monte Carlo Tree search. Disclaimer: I am not supposed to develop strong bots with highly tuned up static evaluation function, I use one not highly optimized.
Requirements (i am sorry but I do not know all minimal needed versions):
python with WxPython bzr git ghc (>= 7.0.0) gcc make
Just type:
git clone git@github.com:JackeLee/rabbocop.git cd rabbocop
and then make playMCTS or make playAB. The playMCTS option is for
playing against bot using Monte Carlo Tree Search and the playAB option against
bot using AlphaBeta search.
Before compiling with additional options it is necessary to run make
clean. You can change compilation options by adding some of the listed
options to make:
- VERBOSE=n
- For IterativeAB program, this will print actual best move and score anytime a depth is fully explored. For MCTS it will print actual best move and score after each n iterations of UCT algorithm.
- EVAL=fairy
- Builds program using Fairy's evaluation function.
- NULL_MOVE=1
- Enables Null moves for AlphaBeta search.
- canPass=1
- Adds possibility to generate less than 4 moves.
- CORES=X
- Compiles program with the number of available CPU's set to X. (Default is 1) When running program compiled with CORES=X by yourself it is necessary to add +RTS -NX as additional arguments.
- WINDOW=X
- It switches on aspiration window option for AlphaBeta algorithm with Window set to X.
- noHH=1
- Disables history heuristics optimisations for MCTS.
- abHH=1
- Enables history heuristics optimisations for AlphaBeta.
- noHeavyPlayout=1
- This disables heuristic in playouts.
- PROF=1
- Enables profiling for bots.
- playMatch
- Starts new game AlphaBeta search versus MCTS. Use it instead of playMCTS or playAB.