Greenpeep is a program that plays the game
It uses the Monte Carlo method for Go, which plays out many
randomized games from a position and chooses moves which obtain
better win rates. The Monte Carlo method is integrated with
game tree search using the UCT algorithm
[L. Kocsis & C. Szepesvari, "Bandit based Monte-Carlo planning." ECML 2006]. Greenpeep incorporates
enhancements to UCT that were first developed in
the MoGo project [S. Gelly & D. Silver, "Combining online and offline knowledge in UCT." ICML 2007]. Greenpeep is distinctive in its use
of additional machine learning techniques: pattern weights are
learned from offline self-play and are used to bias UCT/Monte Carlo choices towards
stronger moves, and online learning is used during UCT to generalize
across lines of play. Greenpeep plays a strong game on the small 9x9 Go board.
Greenpeep has played on the Computer Go Server CGOS and in Computer Go tournaments on KGS.
Greenpeep is being developed by Chris Rosin.