Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: interference between verified null move R=3 and transposition table

Author: Anthony Cozzie

Date: 08:49:16 06/30/04

Go up one level in this thread


On June 30, 2004 at 11:30:19, Stuart Cracraft wrote:

>Hi -- I threw together a quick chess program in the past 3
>weeks to do some research and recently added a testsuite
>capability to at least get some measure of QA for changes.
>
>I then ran testsuite on the program at about 12-13 seconds
>per move against Win-at-Chess (300 positions) on a slow
>(for these days) PC (1ghz P3 w/ 512mb).
>
>To my surprise, Verified Null Move with R=3 scored better
>without transposition table than with. Transposition table was
>set to 0.5M entries in all cases.
>
>See below where #3 result is greater than all the others.
>It's puzzling to me why transposition table should reduce
>the result for #3 down to the level of #5.
>
>Has anyone else seen this with their program or programs
>in general? It is most puzzling/counter-intuitive to me.
>
>1. ***** base/plain/no nullmove/no transposition/no optimizations *****
>*** Test score: 61% correct (183 out of 300)
>*** Grand total nodes: 440075104 time: 3868.21
>*** Average nodes/time/nps per test position: 1466917 13 113767
>
>2. ****** opt trans *******
>*** Test score: 67% correct (203 out of 300)
>*** Grand total nodes: 620892160 time: 3635.11
>*** Average nodes/time/nps per test position: 2069640 12 170804
>
>3. ****** opt verified null 3 *******
>*** Test score: 69% correct (207 out of 300)
>*** Grand total nodes: 656987264 time: 3591.53
>*** Average nodes/time/nps per test position: 2189958 12 182927
>
>4. ****** opt trans verified null 2 *******
>*** Test score: 64% correct (194 out of 300)
>*** Grand total nodes: 663116096 time: 3521.66
>*** Average nodes/time/nps per test position: 2210387 12 188296
>
>5. ****** opt trans verified null 3 *******
>*** Test score: 66% correct (200 out of 300)
>*** Grand total nodes: 678707776 time: 3521.42
>*** Average nodes/time/nps per test position: 2262359 12 192737
>
>(I also tested against non-verified null move and its result was
>far worse than the above which made sense, but that transposition
>effect sent me for a loop.)
>
>Any comments would be appreciated?
>
>Stuart

You have transposition table bugs.

anthony



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

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