Re: [Bug-gnubg] some problems

[Jim Segrave]

On Tue 20 Jul 2004 (11:00 +0200), max-d wrote:
> 
> ----- Original Message ----- 
> From: "Jim Segrave" <address@hidden>
> To: "max-d" <address@hidden>
> Cc: <address@hidden>
> Sent: Tuesday, July 20, 2004 1:36 AM
> Subject: Re: [Bug-gnubg] some problems
> 
> 
> >
> > > - an incredible but probably well known mistake played by Parlorbot on
> FIBS
> > > this evening ...
> > > please see move 14 ,2 plies analysis agrees
> > > Thanks for your work !!
> >
> > I must be missing something
> 
> No it's me !
> i did realize the situation just after having posted this message .
> Match length is 11 , sorry to have forgot to mention it .
> About MWC , rollouts gave 24/18 9/4 -1.40% vs 24-13 ,612 trials but
> I believe this was 0ply rollouts then i tried analysing just the game 2 with
> 2 plies and roll this position out again .
> As you say the process is very slow and after some trials i had
> 24/18 9/4 ahead too!
> i will try to understand  the rollouts settings well and roll it again since
> i  think lot of players should miss the right move here !
> 
> thanks for your answer

A bit more playing with this position is interesting:

With ParlorBot owning the cube, 24/18 9/4 is the better play and 24/13 is
a fair sized mistake (0.060). With a centred cube, 1296 trials of 0
ply rollout make the two plays equal. With you owning the cube, 24/18
9/4 is a serious blunder (0.117). I think in the position shown, I
would have made the wrong play rather than face a double shot on the
18 point. Maybe that's why I'm not a better player.

-- 
Jim Segrave           address@hidden