Re: [Bug-gnubg] Re: Cubeless Rollout Bug In New Build (Now fixed! )

[Jim Segrave]

On Sat 11 Jan 2003 (07:08 +1300), Joseph Heled wrote:
> 
> 
> Moore, Dave wrote:
> >This is a pretty promising result!  Just to play Devil's advocate, 
> >though, isn't it possible that this shows that GnuBG is perhaps worse at 
> >defending against a backgame while being equally good at playing a 
> >backgame?
> > 
> Neil choose this position because he thinks it represents a skill test 
> for the backgame player only. I think this is mostly true. Some of us 
> have looked at games played and I think no one saw big errors on the 
> trapped side.

It depends on the move filters though. When I was syslogging
yesterday, looking for odd play during rollouts, I captured this
game. Rollout was at 2 ply, with the default filter setting:

2 moves at 0plky  +  3 at 0.1
no one ply pruning

Re-analysing with a filter set

0 moves at 0 ply + 10 @ 0.5
no one ply

The backgame shows an overall error for the game of 0.076, with the
very first move at 0.43

This is not a bug, but simply an indication that in these positions,
the zero ply pruning needs to be very conservative (or liberal in the
sense of accepting lots of moves).



-- 
Jim Segrave           address@hidden

g1-vr.sgf
Description: Text document