Re: [Bug-gnubg] I'm a bit confused

[Joern Thyssen]

On Wed, Jul 30, 2003 at 11:03:42AM +0200, Jim Segrave wrote
> On Wed 30 Jul 2003 (07:35 +0000), Joern Thyssen wrote:
> > On Wed, Jul 30, 2003 at 12:07:13AM +0200, Jim Segrave wrote
> > > > 
> > > > You may see similar problems for evaluations as well, but as evaluations
> > > > are never truncated then the problem is usually not so large.
> > > 
> > > Now I'm confused - I would think that the wins/gammons/backgammons
> > > figures are accurate, but probably not complete enough. In the
> > > situation you describe where you'll be doubled out, your wins should
> > > be 0.0, lose gammon and lose backgammon will be either 0.0 or 1.0
> > > depending on where your pieces are.
> > 
> > NO! If the rollout is truncated at "double, pass" gnubg uses the
> > cubeless gwcs from the cube decision leading to "double, pass", It does
> > *NOT* use 1 0 0 0 0!!!  "1 0 0 0 0" is a *cubeful* gwc and we can't mix
> > that with the cubeless gwcs from the other trials.
> 
> OK - I haven't looked at BasicCubefulRollout in detail and I naively
> assumed that it would use 1 0 0 0 0 0 when it concluded the player on
> roll would double the opponent out. 

OK


> I've also been assumming that for
> cubeful rollouts, when games are rolled out to completion, the cubeful
> equity was adjusted by crediting the winner with a win/gammon/bg and
> calculating the delta to the accumulating cubeful equity based on the
> current cube value. 

When a "double, pass" is encountered the rollout is truncated.

For a trial ending in "game over" the cubeful equity is equal to the
cubeless one (e.g., +1, -1, +2, -2, etc), so we add, say, +1 to the
cubeful equity. It's the same as normal play: even though the match or
sesion is played with a cube, one player is awarded "cubeless" points
when the match or game is over -- the value of holding the cube is nil
when the game is over.

Of course, for a rollout with variance reduction the cubeful result of,
e.g., +1 will be corrected with the total luck for that trial.

Continuing the rollout has actually suggested by Michael Zehr a long
time ago as extra variance reduction. I think the idea is to branch at
every "double, pass" or "double, take" decision.

> And for games being truncated that the evaluation
> at the truncation point was given a cubeful adjustment and then
> factored in to the accumulated rollout stats.

No. The win/lose stats only counts actual games being lost or won -- not
games truncated (either by "double, pass" or by reaching the user's
truncation setting).

Jørn