[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnubg] Multi-bot rollouts of backgame position
From: |
Joern Thyssen |
Subject: |
Re: [Bug-gnubg] Multi-bot rollouts of backgame position |
Date: |
Sat, 8 Mar 2003 19:48:44 +0000 |
User-agent: |
Mutt/1.4i |
On Sun, Mar 09, 2003 at 07:53:06AM +1300, Joseph Heled wrote
>
> I don't buy it.
>
> If 2ply came out worse than 0ply I say:
> Either
> He is using the wrong filters for 2ply.
> He is using the version before the rollout bug fix.
> The dice was different for the rolls, with particularly bad dice
> for the 2 ply.
Variance reduction should help eliminating that, but of course
variance reduction may be worthless if the 2-ply evaluation is very wrong,
I guess it depends on how good relative equities 2-ply give for this
position.
I wouldn't have chosen 32000 trials for the 0-ply, but rather something
proportional to 1296, say, 31104, in order to get the rotation of the
rolls. But I'm not sure how much it really matters.
Jørn