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

[Jim Segrave]

On Wed 30 Jul 2003 (13:56 +0000), Joern Thyssen wrote:
> On Wed, Jul 30, 2003 at 01:25:36PM +0000, Joern Thyssen wrote
> > On Wed, Jul 30, 2003 at 12:57:03PM +0200, Jim Segrave wrote
> > > I just googled and as I read it, he's suggesting branching at
> > > double/takes, not continuing after double/pass.
> > > 
> > > http://www.bkgm.com/rgb/rgb.cgi?view+567
> > 
> > Yes, exactly. 
> > 
> > How does the Zehr-algorithm work? You have the results from the  two
> > branches. What is it exactly that you add to the cubeful equity?
> 
> After careful re-read I understand it now. 

I assume that you recusively average the trees at each branch point.
 
> One other question: will this be extendable to match play?

I would have thought so.

But I wonder how well it works. Take a rollout of a position where one
player has an equity well past the doubling point. The initial action
is a double, so you now have two branches, one with an immediate
double, the other branch will probably fork again with a double on the
next play. This will continue until either the equity goes above the
too-good point or drops below the doubling point. But what has it
actually done to improve the variance reduction?


-- 
Jim Segrave           address@hidden