Re: [Bug-gnubg] non-random seeding

[Jim Segrave]

On Sun 18 Jul 2004 (18:57 +0000), Joern Thyssen wrote:
> On Sun, Jul 18, 2004 at 10:26:49AM -0500, Chuck Bower wrote
> > Hello, all interested parties,
> > 
> >      It is sometimes advantageous to roll out the same position on
> > multiple computers to speed up the attainment of results.  I've
> > noticed (unfortunately after the fact) that when I set up the position
> > on different computers I end up with the *same* 10 digit seed.  In
> > the case that occurred, the seed was:  2032818304 on three computers.
> > I'm currently running the build of 040712 but I'm pretty sure the
> > rollouts were begun on builds dated 040405.
> > 
> >      I start GNU-bg, begin a new game, then edit the
> > position, setting score, cube-location, checker positions.  I then
> > use the <hint> button to evaluate @World Class, adjust the rollout 
> > parameters, and start the rollout from the <hint> window.
> > 
> >      If this isn't a bug, then can you add a 'random seed' box which
> > can be checkmarked in the rollout settings window?
> 
> I'm not sure if it's a bug or feature. Anyway, if you always want a
> random seed, you can make a file .gnubgrc in the same directory as
> .gnubgautorc and add the following command: 
> 
> set rollout seed

As I posted on GOL I think it's a bug. The call to InitRNG to set up
the rollout RNG in gnubg.c specicially tells it not to initialise the
seed. It then xor's the seed with a magic number.

Possible solutions:

1) change the fSet paramter in the call to TRUE

2) xor the rollout rng seed with the dice rng seed since that one has
   been set. There will always be a relationship of sorts between the
   two RNGs, but it's my belief it probably isn't going to seriously
   bias things.

-- 
Jim Segrave           address@hidden