Aya Bot

    Keywords: Online Go

Aya uses the difference black first and white first for all points in 19x19. But in 9x9, Aya uses selective full board search.

Aya usually searches 6-7 plies in 30 seconds.

Nodes per second is about 500-800 nps, and ladder search is about 600,000 nps in 9x9. (Opteron248 2.2GHz). Evaluation function includes strings capture and some connections search. Life and death, eye-shape recognition are static.

Moves are generated from simple stone shape pattern. (Aya does not have pattern database. This pattern is written many "if" sentence.) Life and death, eye-shape and connection points from evaluation function are also generated.

The moves around weaker and bigger stones get higher temporary value.

Moves are selected by its value on each depth. For example, 10 moves are selected in depth 4.

At first Aya tries pass and collects opponent best 5 moves. In first depth(root) Aya generates many moves, but in second depth, Aya generates only nearby last moves and best 5 moves after root's pass. Over 3 depth, Aya generates only nearby last moves and around getting weak stones by last opponent move.

These idea are mainly from Haruka, Japanese strong Go program's explanation. There is its PDF (in Japanese) [ext] http://www.cs.ualberta.ca/~games/go/seminar/2003/031103/vol5-4.pdf And you can see Mr. Kishimoto's translation. [ext] http://www.cs.ualberta.ca/~games/go/seminar/2003/031103/haruka.ppt

My current target is to beat Gnugo in 19x19. But current win rate is only 13 percents... Road is faraway, but it's challenging :-)

Aya Bot last edited by tapir on May 31, 2009 - 15:58
RecentChanges · StartingPoints · About
Edit page ·Search · Related · Page info · Latest diff
[Welcome to Sensei's Library!]
Search position
Page history
Latest page diff
Partner sites:
Go Teaching Ladder
Login / Prefs
Sensei's Library