[Computer-go] 7.0 Komi and weird deep search result

Brian Sheppard sheppardco at aol.com
Sun Apr 3 13:21:10 PDT 2011


>AIUI, RAVE without special modifications (like those done in Mogo later)
> does not have any convergence guarantees either.

MCTS using RAVE prioritization *does* converge to game theoretic values in a
binary-valued space.

But not in a ternary-valued space. In a ternary valued space, it is possible
for both players to become discouraged while trying to find a win, and then
settle for a draw without exploring exhaustively.

-----Original Message-----
From: computer-go-bounces at dvandva.org
[mailto:computer-go-bounces at dvandva.org] On Behalf Of Petr Baudis
Sent: Sunday, April 03, 2011 3:04 PM
To: computer-go at dvandva.org
Subject: Re: [Computer-go] 7.0 Komi and weird deep search result

On Fri, Apr 01, 2011 at 02:00:24PM -0400, Brian Sheppard wrote:
> That's all fine, but UCT with exploration term == 0 does not 
> asymptotically converge in a ternary-valued space.

Does this matter in real life? AIUI, RAVE without special modifications
(like those done in Mogo later) does not have any convergence guarantees
either. I suppose Valkyria uses RAVE, so RAVE acts as its exploration term,
like in Pachi.

By the way, in Pachi I have floating point values in nodes, so I just count
jigo as value of 0.5 (as opposed to 0 or 1). I have no further special
handling (as far as I can remember) and this works nicely as it is. What
Magnus describes sounds most like some implementation artifact to me, but
maybe Pachi just handles something else differently, avoiding this as well
then.

-- 
				Petr "Pasky" Baudis
UNIX is user friendly, it's just picky about who its friends are.
_______________________________________________
Computer-go mailing list
Computer-go at dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go




More information about the Computer-go mailing list