Re: [Bug-gnubg] Recube efficiency in short bearoffs

[Joseph Heled]

Øystein Johansen wrote:
> 
> >===== Original Message From Joseph Heled <address@hidden> =====
> >Øystein O Johansen wrote:
> >
> >> Joseph -- did you make some work on estimating a recube efficiency based on
> >> the length of a race?
> >
> >Look again at this thread
> >http://mail.gnu.org/pipermail/bug-gnubg/2002-July/000437.html
> 
> Oh, thanks!
> 
> in the example position from previous mail, the average number of rolls for
> the player is 2.951 and for the opponent it's 2.675.
> 
> a = M * P * I = 0.0606 * 2.675 * 1 = 0.162
> 
> This sound to me like a very small value, though I have not calculated the
> cubeful equities with this values. Is this model valid for this position?
> 
> (or maybe I misunderstand something again...?)
> 
> -Øystein
> 
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/bug-gnubg

This number is just the ratio to average the 'live' vs 'dead' values in my
model. I think I concentrated on longer races when I did this, and it might be
that the cube vigorish starts to climb again for really short races, or that for
"balanced" positions the model does not work very well. I think it need more
looking into.

On your position (25,25 away however) I got (fibs2html cubefull rollout)

1296-Games  Equities: Money  0.239     
 
            0.0%   0.0%  62.0%    38.0%   0.0%   0.0%  
  Match win probabilities:
        No Double   Double/Take   Double/Drop
          52.46%     51.84%        56.00%

             Cube action: No double, take

The 2ply said ND/take too (but barely), 0ply thought Double/Take, so I am
inclined to agree the value of the live cube is underestimated here.

-Joseph