Re: [Bug-gnubg] Possible evaluation bug

[Joseph Heled]

Since backgammon pieces don't come back from the dead, a system with plentiful memory will have a set of nets for one-side-has-1-checker, one-side-has-2-checkers, etc/ That will help a lots of those boundary (but relatively rare) situations.

Another possible approach in that direction is to have a per-remaining-checkers setting for move filters.

-Joseph

On 18 February 2018 at 11:37, Philippe Michel <address@hidden> wrote:

On Sat, Feb 17, 2018 at 02:09:33PM +0100, Øystein Schønning-Johansen wrote:

> So.... the conclusion must be that there is something funny with the
> movefilters. Don't know what.

0-ply evaluates the resulting positions quite haphazardly and there is
only one move with 24/19 if the first eight choices. The wider filter
gets three more and these four get the top spots at 2-ply. This is
better but there are still a few reasonable 24/19 plays missing.

The 0-ply evaluations of the next roll position are probably rather poor
as well since the 2-ply equities of these moves are much more dispersed
than they should. 3-ply is better and 4-ply seems right, with all the
plays breaking the 24 point sensibly in a 0.02 interval.

_______________________________________________
Bug-gnubg mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/bug-gnubg