Re: [Bug-gnubg] Bug#659494: Resigns more points than necessary when fewer points would still result in opponent win

[Russ Allbery]

(Trying this again from the address from which I'm actually subscribed.
Sorry about the duplicate, Josh.)

Hello, folks,

I received the following bug against the Debian gnubg package.  I vaguely
recall that this implementation decision was discussed on the mailing list
before and that the current behavior is intentional, but I don't recall
whether there was discussion of making it a tuneable option or exactly
what the rationale was.

Josh Triplett <address@hidden> writes:

> Package: gnubg
> Version: 0.90+20120127-1

> When gnubg knows it will lose the current game, and knows that doing so
> will provide its opponent with enough points to win the match, gnubg
> will resign a gammon or backgammon if those still have a non-zero
> possibility of occurring, rather than just resigning a normal game.  For
> example, in a 7-point match in which the doubling cube had reached 8,
> once I had enough checkers borne off to guarantee a win, gnubg offered
> to resign a gammon.  Checking the probabilities afterward, gnubg had a
> 100% chance of losing, but only a 0.4% chance of losing a gammon.

> I assume its offer to resign more points than necessary occurred because
> gnubg considers all match losses equal in value.  With all else equal,
> gnubg should not throw away more points than necessary; that would make
> gnubg's play more natural, and avoid making a loss seem unnecessarily
> overwhelming.

-- 
Russ Allbery (address@hidden)               <http://www.eyrie.org/~eagle/>