________________________________ From: Magnus Persson <[email protected]> No this is not doable. Especially not in the endgame-subgame sense, because there are no independent sub-games in 99% of a normal gogame. This is the reason non-MCTS old school programs got stuck. Search must be global. I am not saying that local search cannot be helpful. It just has to be integrated efficiently into global search. Sorry for sounding so negative. But reinventing old square shaped wheels will not bring us forward. Magnus At the level of pro games, it may be more common to leave things unresolved until late in the game, but it seems to me that strong players have a sense that "this group is settled; I'll respond to threats, but otherwise leave it be" - and they'll have a keen sense of when a nearby play changes any group from stable to unstable. If that sort of knowledge can be efficiently integrated into the global search, it should help. When computer go programs fail, it appears that they misunderstand life-and-death, semeai, and especially seki. To a human observer, those positions can be resolved independently, especailly for dan-level players. Put another way, it may be true that the vast number of Go positions - especially for the first 50 moves or so - can not be partitioned into independent sub-games; but as the game progresses, such partitioning becomes not only possible but essential to understand and win the game. By the time a human player reaches shodan, she knows how to create ko and seki positions and use them to exploit a weaker opponent; this is the achilles heel of most Go programs today.
_______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
