Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Endgame tablebases - how much do they really affect playing strength

Author: Robert Allgeuer

Date: 09:24:33 10/03/03

Go up one level in this thread


On October 03, 2003 at 09:45:36, Mark Rawlings wrote:

>On October 02, 2003 at 04:09:22, Robert Allgeuer wrote:
>
>>Hi,
>>I have run two self-play tests on this subject in the past, please refer to
>>these links:
>>
>>http://www.chess-archive.com/ccc.php?art_id=281455
>>http://www.chess-archive.com/ccc.php?art_id=307405
>>
>>The bottom-line of these tests (and at least another one that I am aware of) is
>>that 5 men EGTBs do not give you a statistically measurable increase in playing
>>strength compared to having no EGTBs.
>
>Is this because it slows down the search?  The "no increase in strenght" has
>been reported before, but it's still hard for me to believe...
>
>Mark
>


I do not believe that it is due to a slowdown in search, in these tests the
configurations (of the same engine) with tablebases still searched deeper -
despite slower node speed - than the non-tablebase configurations.

Pure speculation, but I would think:
- there are probably not that many games percentage-wise, where really an
endgame position is reached where the search of an engine without tablebases
would indeed pick an incorrect move that decides the game. Probably today's
engines on today's hardware search sufficiently deep to play very well without
tablebases also in the endgame, most likely a bit, but not much worse than with
tablebases.
- Analysing the games it becomes evident, that long games where 5 men positions
are reached are won by the engines probing tablebases with a high probability.
But engines with 5 men tablebases lose games with 6 and 7 men over proportion. I
also think due to the reasons that Uri explains: the engine sometimes tries to
avoid to play into drawn 5 men positions, as long as it has a positive score,
and then it might lose instead. It appears that this probably small effect
cancels out the small advantages from the first point. In any case in the end
there appears to be no statistically significant difference in overall playing
strength.

Robert



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.