Re: [Bug-gnubg] Some Questions about rollouts.

[Joern Thyssen]

On Sun, Nov 03, 2002 at 04:49:16PM -0800, David Montgomery wrote
> 
> > There is still one thing I don't understand, even though I have read the
> code
> > several times: Which seven rolls does it pick? How does it make the
> choice? It
> > doesn't pick random as I understand. And as I understand this it doesn't
> find
> > the order in respect of how good the rolls are either. Please explain.
> 
> I haven't looked at this code in a long time, but it came from
> my program, so I think I remember how it works.  My recollection
> is that it rotates among three groups of 7 rolls, so that all 21
> rolls are represented.    That is, 1/3 of the 1-ply continuations
> are evaluated at 2-ply with one set of 7 rolls, another 1/3 are
> evaluated with a *different* 7 rolls, and the remaining 1/3 are
> evaluated with the remaining 7 rolls.
> 
> The partitioning of the rolls into three groups was done with some
> effort toward evening up the luck.  For example, I tried to roughly
> balance the number of doublets and the number of individual sixes,
> fives, fours, threes, twos, and ones.
> 
> It does not, however, do any analysis of which rolls are likely to
> be good or bad in the current board situation.  You can think of
> it as a "random" sample of the continuations, with a little effort
> to make it better (on average) than a truly random sample.

Do you have code for other reductions, e.g., 66%, 50% or 25%? Also, it
would be very nice to make this work for 3-ply evaluations as well. 

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!