> For anyone who is interested in this, that is reported in this paper
> (see section 3):
>  http://hal.archives-ouvertes.fr/docs/00/36/97/83/PDF/ouvertures9x9.pdf
> 
> (I think these results should be taken with a grain of salt, as it is
> just self-play, but it does fit the other evidence.)
> 
> Darren


Hi all,

When reading this paper, which describes the "coevolution" of a 9x9 opening 
book using mcts on top of mcts, I was wondering whether this would make for a 
strong go player under regular time-settings, too.
In other words, that player would be a standard MCTS player, but for playouts, 
move generation is done by MCTS, too. So you get "inner" and "outer" playouts.

I think this would require that the inner playouts are rather light to get any 
benefit at all. Also, strong move ordering / pruning should be used to get 
values more quickly, including RAVE. Maybe this is feasible with current 
hardware with the amount of inner playouts being around 100-1000, and still get 
100-1000 "outer playouts"/second (9x9)?

Regards,
Isaac
-- 
Neu: GMX De-Mail - Einfach wie E-Mail, sicher wie ein Brief!  
Jetzt De-Mail-Adresse reservieren: http://portal.gmx.net/de/go/demail
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to