Re: [Bug-gnubg] tty version rollouts, equity & RNG seed

[Gary Wong]

On Mon, Jan 13, 2003 at 02:15:28PM +0100, Morten Wang wrote:
> I'm confused by the numbers output when running gnubg with the '-t'
> option.  What exactly are the numbers supposed to be.  The heading is:
> 
>          Win   W(g)  W(bg) L(g)  L(bg)  Equity   Cube E    n
> 
> I thought "Cube E" would be the cubeful equity, like it is on the gtk
> version, but instead the number seems to be something completely
> different:

Yes, Cube E is supposed to be cubeful equity.  

> Then, when I do a hint after the rollout, the shown equity is way off:
> 
> 3. Rollout          8/5 8/2                      Eq.:  +1.605 ( +1.819)
>    0.391 0.067 0.002 - 0.609 0.082 0.002 CL  -0.234 CF  -0.395
>   [0.001 0.001 0.000 - 0.001 0.001 0.000 CL   0.003 CF  +0.006]
>     Full cubeful rollout with var.redn.
>     2592 games, seed 692318375 using quasi-random dice and Mersenne Twister
>     Play and cube: 2-ply cubeful 100% speed [world class]
> 
> The -0.395 value should be correct.  +1.605 is easily +2.000-0.395
> (the cube's turned in the game), but why that value?

No idea, something looks buggy...

> I'm also confused by the reported seed used.  The commands file I used
> had "set seed 1652552448" (which was the seed used when I ran the
> rollout with 1ply in the gtk version).  Is there an overflow here?

Rollouts use their own seed -- "set rollout seed <n>" is the command you
need for that one.

> I'm not sure where to look for version info.  The last entry in the
> Changelog is Sat 2003-01-11 11:29.  The code I have was checked out
> from CVS a half hour later my time (noon CET).

Hmmm, this might explain part of the problem... the new cubeful evaluation
cache gave incorrect cubeful equities in some circumstances.  Hopefully
this has been fixed since 2003-01-12.

Cheers,
Gary.
-- 
   Gary Wong           address@hidden           http://www.cs.arizona.edu/~gary/