bug-gnubg
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Bug-gnubg] Experimental rollout feature


From: Jim Segrave
Subject: [Bug-gnubg] Experimental rollout feature
Date: Mon, 23 Jun 2003 23:53:52 +0200
User-agent: Mutt/1.2.5.1i

I've added an option to the Rollout general settings page. It allows
you to tell gnubg to terminate a rollout if, for all the displayed values:

win/win gammon/win backgammon/lose gammon/lose backgammon/cubeless
equity/[cubeful equity]

the standard deviation is less than a user chosen fraction of the
value. You can also set a minimum number of games, to try to ensure
that a fluke run or rollouts doesn't give a temporarily low standard
deviation with misleading results.

Default is for the option to be off, the minimum number of games to be
144 and (if the option were enabled) to stop when the standard
deviation is less than 10% of the value.

It's only been lightly tested, but it does appear to work.

This is part of the work I want to do in general on rollouts - saving
positions, adding a batch rollout option and saving and restoring
enough data to allow extending a rollout with little or no loss in
accuracy.

-- 
Jim Segrave           address@hidden





reply via email to

[Prev in Thread] Current Thread [Next in Thread]