[computer-go] Monte Carlo combined with minimax search

[Rémi Coulom]

Peter Drake wrote:
> On Jul 22, 2006, at 11:19 PM, Rémi Coulom wrote:
> 
>> Peter Drake wrote:
>>> One thing I'm going to try next is counting the score not as 
>>> sum(own(p)) but as sum(signum(own(p))).  This way, it would be more 
>>> interested in
>>
>> Do you mean signum(sum(own(p))) ?
> 
> That wasn't what I had in mind.  Wouldn't this mark every node as +1, 0, 
> or -1?

Ah yes, I think I understand what you meant now.

> 
>>> scoring points than in strengthening its hold on points it already 
>>> has.  Of course, ideally there would be a continuum between these 
>>> two, where the program would become more conservative when it was ahead.
>>
>> I am not sure that a continuum would be better than using the 
>> probability of winning all the time. At least, I am certain that using 
>> the probability of winning all the time is much better than using 
>> expected territory all the time.
> 
> Can you say more on this?

When I switched from using territory to probability of winning, Crazy 
Stone changed from scoring 36% against GNU Go 3.6 at level 10 to scoring 
more than 60%, at 16 minutes per game, single CPU.

> 
> Peter Drake
> Assistant Professor of Computer Science
> Lewis & Clark College
> http://www.lclark.edu/~drake/
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/