Pentagod

A pentago AI that implements several algorithms including MCTS (Monte-Carlo Tree Search), PNS (Proof Number Search) and AB (Alpha-Beta). Its primary interface is GTP (Go Text Protocol), but a python webserver that interfaces pentagoo with the gtp protocol is included.

Features

Simple GTP framework
Very fast bitboard implementation
Easy to create new players
Players Includes:
- MCTS
- PNS
- AB
Very strong player, likely the strongest in the world
Web server integrates very well with the Pentagoo front end (a modified version is included in this project)

Requirements

Linux (probably works on any unix)
C++ tool chain
Python for the webserver

Usage

Check out the code from github
Run "make" to compile the code
To use the GTP interface:
- run ./pentagod
To use the webserver:
- run ./web.py
- point your browswer at http://localhost:8080/

TODO

Better scoring function
- Based on patterns
- Meaningful range such as win percentage
Win condition pruning (certain patterns make certain win conditions impossible)
Negamax
- Randomization when two moves are very similar
- Better time cutoffs
- Use the previous iterations for move ordering, might need better scoring function first
- Improved transposition table
MCTS
- Better rollouts: patterns, forced moves
- rotate and move as separate moves
- Prove wins/losses to a greater depth
PNS
- Use scoring function
- Use a small depth of AB?
Opening book
Web/Pentagoo
- Choose opponent AI and time cutoff
- Better move format
- Undo when playing against computers

pentagod

Commits

Indentation fix

Let params be set for the three agents

Add a configurable amount of randomness to the scoring function so alpha-beta is less deterministic

Limit the depth of full symmetry handling

Slightly faster move rotation

Improve the webserver

README