Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: A BETTER WAY TO TEST CHESS PROGRAMS

Author: José Carlos

Date: 19:43:43 01/14/00

Go up one level in this thread


On January 14, 2000 at 15:46:32, James A. Tackett wrote:

>      The standard way to test chess programs, i.e., have them play a "regualr
>match" against each other does provide interesting information about their
>relative strength and playing style.  However, the opening books of the programs
>and the relatively short matches tend to add noise to the final result and may
>distort the truth.  How many times have we seen the opening book of a major
>program be set up to take advantage of the weaknesses of a competitor?  This may
>be good for marketing the new program, but it is bad for us consumers. I want to
>know who has the strongest playing program, not the slickest, deal from the
>bottom of the deck, opening book!
>
>     I do my testing by picking 20 early middlegame positions from grandmaster
>play (e.g., Dragon, Ruy, King's Indian, Queen's Gambit etc.) and have the
>programs play two games (one White, one Black) from these set positions.  A
>total of 40 games with opening books turned off!  You cannot run a big
>tournament this way due to the time requirement, but to decide if program A is
>better than program B, this system is very good.

  Programmers spend a lot of time improving their books (it took me some days to
make it work in Averno). Why should we forget that work?
  And if you think that the book is not part of the program, what about
tablebases? and what about hash tables, and what about extensions? and what
about null move?
  What's the difference?
  The book is part of the program, as it is for humans too. Don't professional
chess players study openings to beat their opponents?
  Think about it...

  Regards,

  José C.



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.