[Bug-gnubg] RE: Bug-gnubg Digest, Vol 56, Issue 13

[bob koca]

   It was the starting checker setup for a one point match and I put the advanced player on roll. The pos ID is 
4HPwATDgc/ABMA  and the match id is cAkgAAAAAAAA.  Let me try to geive a more complete list of settings.
The settings used for both rollouts were the following:
 
i) to truncate cubeless at exact bearoff database
ii) variance reduction
iii) use quasi random dice
iv) rollout as initial position
v) cube decisions use same settings as checker play
vi) seed = 1001
vii) 10800 games
viii) player 0 set to play as expert
ix) player 1 set to play as advanced
x) g11 match equity table
 
 
  The only difference in the settings for the two rollouts were that one had cubeful checked and the other did not.
The one with cubefull checked gave .472404 wins and the one with it not checked gave .458272.
 
  I wasn't expecting it to be .50 due to both the fact that one player was set to advanced and the other expert and also as you point out
that one player is set to be on roll. The surprising thing to me is that one setting gave a significatly higher win %age. Since it is a one point
match the cubeful evaluations should be exactly the same as non-cubeful.
 
Bob Koca

 

> 
> Message: 4
> Date: Wed, 11 Jul 2007 22:50:03 +0200
> From: "Christian Anthon" <address@hidden>
> Subject: Re: [Bug-gnubg] checking cubeful or not makes a difference
> when it shouldn't
> To: "bob koca" <address@hidden>
> Cc: address@hidden
> Message-ID:
> <address@hidden>
> Content-Type: text/plain; charset=UTF-8; format=flowed
> 
> It sounds strange, but please provide a __minimal__ example of the
> problem. Including position and match id's and the settings you
> change.
> 
> Note that even though it is an initial position a specific player is
> on roll. So even if you set expert to play against expert you would
> expect to see a resulting equity different from 0.
> 
> Christian.
> 
> On 7/11/07, bob koca <address@hidden> wrote:
> >
> >
> > I rolled out a one point match using player 0 as advanced and player 1 as
> > expert.
> > I had the advanced player on turn and used the rollout as initial position
> > option.
> > In one rollout I checked cubeful under general settings and the other one I
> > didn't. I
> > was expecting the rollouts to be exactly the same since there are no cube
> > considerations to be made in a one point match. However they were not. Even
> > more perplexing is that the difference was definitely statistically
> > significant.
> > With the cubeful box checked off the advanced player had .472404 wins with
> > std error of .001281.
> > With the cubeful box checked on the advanced player won .458272 with SE of
> > .001268
> >
> > Other settings that I had checked were
> > i) to truncate cubeless at exact bearoff database
> > ii) variance reduction
> > iii) use quasi random dice
> > iv) rollout as initial position
> > v) cube decisions use same settings as checker play
> >
> > Is it a bug that the results are not the same? What should I use to see
> > correctly
> > how often advanced would beat expert in one point match?
> >
> > thanks for assistance, Bob Koca
> >
> >

---

Local listings, incredible imagery, and driving directions - all in one place! Find it!