Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Verified Null-Move Pruning, ICGA 25(3)

Author: Vincent Diepeveen

Date: 20:46:36 11/21/02

Go up one level in this thread


On November 20, 2002 at 20:15:47, Bas Hamstra wrote:

>On November 20, 2002 at 14:53:45, Vincent Diepeveen wrote:
>
>>On November 20, 2002 at 11:52:49, Gian-Carlo Pascutto wrote:
>>
>>>My very first thought after looking through this was:
>>>
>>>'You note that Heinz R=2/3 appears to be superior to
>>>R=2 and R=3, but you don't include it in the comparison.'
>>>
>>>--
>>>GCP
>>
>>I would disagree with such a conclusion anyway. R=3 always
>>works better for me in the year 2002 :)
>>
>>In 1996 just R=2 was ok. In 1997 i used a combination of R=3 and R=2,
>>and in 2002 such a combination is not possible. R=3 gives me a ply
>>extra compared to a combination of R=3 + R=2. *on average*.
>
>For me R=2/3 is superior to R=2 and R=3. The biggest problem with R=3 for me, is
>that it makes my search unstable. Some positions it even solves slower. Others
>it searches deeper but misses the right move.
>Bas.

R=2 is keeping your odd and even searches more 'stable' indeed. It is
an evaluation issue too. Consider next search depths versus reduced nullmove
search depths:

depth R=2 R=3
4     1   0
5     2   1
6     3   2
7     4   3
8     5   4
9     6   5

So consider next. White has the move. 5 plies to go. You verify with
2 ply. So not only black may do a move. White too.
With R=3 however only BLACK may do a move.

So with an evaluation function like DIEP, where on average there is
a pretty big difference in middlegame between white and black to move,
as a move can trigger *many* bonuses and penalties, there of course
R=3 is less likely to cutoff than R=2.

In short some of the danger of the bigger reduction in depth is taken
away by the evaluation function!

Now it is trivial for all odd plies, but the opposite is true
for even depths.

However, that odd ply is very important.

On the other hand what i *lose* is that i with 4 plies left already
prune based upon my evaluation function.

The only real risk is at that ply. If my evaluation function has a
wrong score there and there is no tactics in that position to disproof
it, then i've got a problem obviously in DIEP.

A *big* problem.

But the same is true for Tao at depthleft == 3 too :)

So if my evaluation function is 25% better that 25% extra risk i
dare to take :)

Best regards,
Vincent




This page took 0.02 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.