Computer Chess Club Archives




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

Author: Martin Giepmans

Date: 15:39:13 11/20/02

Go up one level in this thread

On November 20, 2002 at 17:45:02, Omid David Tabibi wrote:

>On November 20, 2002 at 17:39:26, Martin Giepmans wrote:
>>On November 20, 2002 at 16:19:29, Omid David Tabibi wrote:
>>>On November 20, 2002 at 16:04:50, Martin Giepmans wrote:
>>>>On November 20, 2002 at 11:43:10, Omid David Tabibi wrote:
>>>>>            ICGA Journal, Vol. 25, No. 3, pp. 153-161, September 2003
>>>>>                          Verified Null-Move Pruning
>>>>>                    Omid David Tabibi and Nathan S. Netanyahu
>>>>>                                   Abstract
>>>>>In this article we review standard null-move pruning and introduce our extended
>>>>>version of it, which we call verified null-move pruning. In verified null-move
>>>>>pruning, whenever the shallow null-move search indicates a fail-high, instead of
>>>>>cutting off the search from the current node, the search is continued with
>>>>>reduced depth.
>>>>>Our experiments with verified null-move pruning show that on average, it
>>>>>constructs a smaller search tree with greater tactical strength in comparison to
>>>>>standard null-move pruning. Moreover, unlike standard null-move pruning, which
>>>>>fails badly in zugzwang positions, verified null-move pruning manages to detect
>>>>>most zugzwangs and in such cases conducts a re-search to obtain the correct
>>>>>result. In addition, verified null-move pruning is very easy to implement, and
>>>>>any standard null-move pruning program can use verified null-move pruning by
>>>>>modifying only a few lines of code.
>>>>>zipped pdf:
>>>>>gzipped postscript:
>>>>If I'm not mistaken this is the well known "verification search" with
>>>>one modification: no verification in the verification search.
>>>>Am I right?
>>>The classical verification search as introduced by Plenkner comes to detect
>>>zugzwangs. Verifeid null-move pruning as presented in the paper, constructs a
>>>smaller search tree with greater tactical strength in middle games (in addition
>>>to detecting zugzwangs).
>>>>Another question:
>>>>your results in table 5 seem convincing, but what about table 4?
>>>>Are these results statistically significant? (my guess is no ..)
>>>For a good estimate of the growth of the search tree as we go deeper, see Table
>>>3 and Figure 4 (which present ECM test positions searched to a depth of 11
>>>The WCS test positions were mainly used for testing the tactical strength
>>>(results in Table 5). Table 4 was provided just for the sake of completeness.
>>I see that I reduced the numbers of the tables (R=1 ;))
>>What I wanted to write is that table 6 is convincing while table 5 is IMO not.
>>Combining table 4 and 5 my impression is that - from a time perspective -
>>R=3 might be better than verified R=2.
>(you mean verified R = 3, don't you?!)
>Even though standard R = 3 constructs a smaller search tree, the problem with
>it, is that it is too risky. Except DIEP which uses a fixed R = 3, I don't know
>of any program that uses that value due to its high risk.
>>Compared to R=3 verified R=2 solves about 3% more positions but is about 40%

Yes, of course I mean verified R=3 (I did it again ;)).

What about my last remark (the percentages)?
From a time perspective your results may indicate that vrfd R=3 is actually
_worse_ than R=3.
OK, R=3 is risky, but for the prize of an occasional oversight (3%) you get
a speedup of about 40% (according to your tables).
The prize for 40% speedup is 1 or 2 extra plies in 3% of the positions ...
I think if you do the math you will see that that is very cheap.

In a tournament game with clocks R=3 is indeed risky. One oversight is often
enough to lose a game. The question is how a (less risky) combination of R=2 and
R=3 compares to your method.


This page took 0.04 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.