Re: [Bug-gnubg] gammon rate in bearoff situations

[Joern Thyssen]

On Fri, Nov 08, 2002 at 06:43:42AM +1300, Joseph Heled wrote
> 
> 
> No magic involved here. getBearoffGammonProbs takes the position of the 
> side wanting to bear one checker. (Since it is a bearoff, all other 19 
> points are know to be empty, so only 6 are needed). It return the 
> probability of doing so in 1,2,3 or 4 moves, just like the one sided 
> database does for complete bearoff.
> 
> Then you do exactly the same thing as in the bearoff computation - I save 
> the gammon if I bear one checker in the next move, and if I don't and op 
> can't bear all of his in his next move, I save by bearing one checker in 2 
> moves etc.

I do exactly the same for the bearoff-databases. Besided storing the
probabilities for bearing off all chequers  the probabilities for
bearing at least one chequer off are stored. Extremely good gammon probs
can be calculated by combining the former with the latter. 

The problem is that the probabilities for bearing at least one chequer
off was not stored in gnubg.bd, so gnubg used some heuristic scheme to
calculate gammon probs, but apparently the scheme was often wrong by a
"large" margin ruining the rollouts. 

I had also noticed some large discrepancies in gammon changes. That was
actually one of the reasons why I suggested changes to gnubg.bd in an
earlier posting.

Jørn

-- 
Joern Thyssen, PhD
Vendsysselgade 3, 3., DK-9000 Aalborg, Denmark
+45 9813 2791 (private) / +45 2818 0183 (mobile) / +45 9633 7036 (work)
Note: new mobile number!