[Computer-go] MCTS and perfect endgame

Andrés Domínguez andresdju at gmail.com
Sun Jul 3 18:30:05 PDT 2011


2011/7/3 Jean-loup Gailly <jloup at gailly.net>:
> Leon,
>> One of problems (which I tested with gogui, thankyou very much)
>> was losing points in endgame when program is winning.
> This is by design. Pachi maximises the chance of winning, not the number
> of points. But if you want Pachi to win by more points while increasing
> the risk of losing,

 I totally disagree. Playing stupid moves in the endgame doesn't
increase the probability of winning, probably the other way  around.
The MCTS probability is not real, but estimanted. I think the problem
here is that the noise is bigger than the useful data. Maybe adding or
extracting komi when the score is nearly 0 or 1 can help.

0.001 vs 0.000 or 0.999 vs 0.998 probability is not statistically
significant when you only play a few thousand playouts.

Andrés



More information about the Computer-go mailing list