Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: I'm being too harsh, but still

Author: Uri Blass

Date: 13:37:06 12/18/02

Go up one level in this thread


On December 18, 2002 at 16:17:31, Bruce Moreland wrote:

>On December 18, 2002 at 15:55:58, Omid David Tabibi wrote:
>
>>Std R=2 is shown to be stronger than std R=3 (see Heinz '99). I don't have to
>>prove it again. This is how scientific research is conducted. If you disagree
>>with Heinz' paper, then publish your counterclaims; but as long as you haven't,
>>his work can be considered accurate, and a source for future work to be based
>>upon.
>>
>>Taking your line of thought, one can continue the criticism, saying why Gamma,
>>Marginal Forward Pruning, or Razoring were not compared to Null-Move Pruning.
>>
>>This is the basic of scientific research, you don't have to re-prove something
>>others have already proved. Anyone who has ever engaged in scientific research,
>>should know that.
>
>All of this "science" relies on similarities between chess programs whose
>sources in many cases are not even made public.  You can't assume that anything
>will produce the same results in your program that it does in other programs.
>
>If we were all working on the same program, in order to find the *one* change
>that makes it better, of course what you are saying is true.
>
>We have a bunch of blind men sharing tips with each other about how to avoid
>tripping over furniture in their own houses.  Each house has a similar
>floorplan, but is not identical.  Sometimes one blind man will say to the
>others, "If I take three steps into my bedroom, I will trip over the bed."
>
>The other blind men have to derive what they can from this, but they can't
>assume that if they take two steps into their bedroom that they won't trip over
>their bed, or if they take three steps that they will.  The bed might be in a
>different spot.
>
>The only way to do "real" science here is to figure out *why* R=3 cannot beat
>R=2, *ever*, and write a paper on *that*.  But this kind of root cause thing is
>probably beyond all of us.

There are programs when R=3 beats R=2
I think that it may be interesting to test it also for gerbil at longer time
control(more than 20 seconds per position).

It may be also interesting to know if the gap between R=3 and R=2 is reduced
because if it is reduced then it suggests that with more time R=3 has chances to
beat R=2.

I believe that for movei R=3 is better than R=2 but I did not do a test that
tested only R=2 against R=3 so I have no proof.

I only tested that R=3+checks in the qsearch made movei better in the GCP test
suite and it also got good results in the games that I tested.

I strongly believe that R=3 is better than R=2 for movei based on  testing few
positions but it is not a proof.

Note that my R=2 and R=3 does not prune when the remaining depth is 1 and I have
different pruning rules there.
I have also other forward pruning rules that are not null move pruning.

Uri



This page took 0.03 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.