Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Book and EGTB self play test (long post)

Author: Dann Corbit

Date: 14:46:19 02/03/03

Go up one level in this thread


On February 03, 2003 at 17:08:47, Robert Allgeuer wrote:
>Others certainly have already done similar tests: while I was recently
>away for some time, I let my computer at home run a self-play experiment
>with the aim to investigate what the impact of opening books and
>endgame tablebases is on the playing strength of a modern chess program.
>
>Method:
>=======
>Crafty 19.1DC was configured with seven different book and tablebase
>combinations. These seven Crafty versions played a round robin tournament
>with 48 subrounds, altogether 1008 games. All duplicate games were
>removed (which left 972 unique games) and these remaining games were
>used to calculate the final rating list, start ELO was 2500.
>Book learning was enabled in order to avoid too many duplicate games,
>preparatory tests had indicated that disabling book learning resulted in
>too many game repetitions, in particular with those Crafty versions
>playing with small books or against an opponent that had no book at all.
>The book´s learning weights were reset before the tournament start.
>In addition to six different "pure" Crafty configurations, one Crafty
>version played with BookThinker, in order to investigate the impact of
>using a much simpler book move selection mechanism than Crafty´s own
>techniques.
>Time controls were deliberately set not too fast in an attempt to obtain
>stable results (i.e. minimal influence of eventual background processes
>etc.).
>
>Participants and Settings:
>==========================
>
>Crafty_lgbk_egtb: Crafty 19.1DC with all 5 piece tablebases and large
>    book (12.8MB) as downloaded from the Crafty ftp site.
>Crafty_lgbk_notb: Crafty 19.1DC with no tablebases and large book (12.8MB)
>    as downloaded from the Crafty ftp site.
>Crafty_aobk_egtb: Crafty 19.1DC with all 5 piece tablebases and a medium
>    sized book (4.25MB) built from Arturo Ochoa´s 2600_updated.pgn file
>    (last updated 2.2.2001) with building parameters 22, 1 and 50%.
>Crafty_aobk_notb: Crafty 19.1DC with no tablebases and a medium sized book
>    (4.25MB) built from Arturo Ochoa´s 2600_updated.pgn file (last updated
>    2.2.2001) with building parameters 22, 1 and 50%.
>Crafty_aobt_egtb: Crafty 19.1DC and BookThinker 4.1c combination with all
>    5 piece tablebases and a book (4.4MB) built from Arturo Ochoa´s
>    2600_updated.pgn file (last updated 2.2.2001) with BookMaker 1.0b.
>Crafty_smbk_notb: Crafty 19.1DC with no tablebases and a small book (1.0MB)
>    built from medium.pgn as downloaded from the Crafty ftp site with
>    building parameters 20, 2 and 50%.
>Crafty_nobk_notb: Crafty 19.1DC with no tablesbases and no book.
>
>All versions with Crafty books played with the standard books.bin file
>    built from the start.pgn file as downloaded from the Crafty ftp site.
>
>crafty.rc (example for Crafty_lgbk_egtb):
>hash 96M
>hashp 6M
>cache 32M
>drawscore=0
>log off
>book on
>book random 1
>book width 5
>learn 5
>resign 6
>tbpath=f:\tb
>egtb
>
>Main hash:  96MB
>Pawn hash:  6MB
>Tablebases: all 5, 4 and 3 piece EGTBs (when used), 32MB tablebase cache
>Time Control: 20min+5sec
>Ponder off
>Book leaning on
>
>
>Platform and Tools:
>===================
>
>Athlon Thunderbird 1.1MHz
>512 MB RAM
>Windows 2000
>
>Crafty 19.1DC
>BookThinker 4.1c
>BookMaker 1.0b
>Elostat 1.1b
>PGN-Extract 15.0
>Winboard 4.2.3
>WB Tourney Manager 0.60 (Jori Ostrovskij)
>PGN-Sammler
>
>
>Results:
>========
>
>    Program           Elo    +   -   Games   Score   Av.Op.  Draws
>
>  1 Crafty_lgbk_egtb: 2517   40  23   281    52.8 %   2497   53.0 %
>  2 Crafty_aobt_egtb: 2510   40  23   287    51.7 %   2498   51.9 %
>  3 Crafty_smbk_notb: 2507   42  24   270    51.1 %   2499   51.1 %
>  4 Crafty_lgbk_notb: 2507   41  22   277    51.1 %   2499   58.1 %
>  5 Crafty_aobk_egtb: 2490   22  40   283    48.4 %   2501   55.1 %
>  6 Crafty_aobk_notb: 2486   24  41   271    47.6 %   2502   51.7 %
>  7 Crafty_nobk_notb: 2482   23  40   275    47.1 %   2503   53.5 %
>
>From this the following can be concluded:
>
>The increase of playing strength due to a complete set of 5 piece EGTBs is
>with 95% confidence equal or less than 67 ELO points.

I don't see why you would bother with the number 67.  You may as well pick any
number.  With NO EGTB, Crafty_smbk_notb and Crafty_lgbk_notb were within 10 ELO
of the best finisher: Crafty_lgbk_egtb.  You are missing Crafty_smbk_egtb and so
we cannot conjecture about the difference between having/not-having egtb for the
small book.

> The results suggest
>that the increase is even significantly smaller than that, if not almost
>negligable.

This would be my best guess.  However, the uncertainty in the data is enough
that I think any sort of conclusion is questionable.

>The increase of playing strength due to an untuned opening book is with
>95% confidence equal or less than 105 ELO points. The results suggest that
>the increase is even significantly smaller than that.

consider the difference betweeb Crafty_aobk_notb and Crafty_nobk_notb.  It seems
to be 4 ELO.  If we repeat the experiment, will it be 4 elo again?  Or 140?  Or
something else?

I don't think we can draw any conclusions.

>There is no measurable difference in playing strength between the Crafty
>version using BookThinker and the comparable versions using native Crafty
>books.
>
>These results are certainly only valid for chess programs that have good
>endgame knowledge coded into their evaluation and whose evaluation function
>allows them to  make sensible moves during the opening. If this is not the
>case differences will be bigger. Likewise, tuning a book specifically for
>the playing style of an engine will very probably yield a bigger increase
>in playing strength as perceived here.
>
>Of course - as always - more games would have helped to get more exact
>results. This test also only looks at the influence on playing strength
>against computer opponents.
>
>I would be very interested to have comments, interpretations etc. from
>people who know more about chess programs than I do. I would also be
>interested, whether these results are in line with experience or other
>tests.

Because you had learning enabled, I think the results are very hard to fathom.
The reason I say that is that how a book is interpreted will change over time.
Also, bad lines will be considered in the learning.  A version without any
opening book may use this learning data and in essense has not only an opening
book, but also a corrected one.

It looks to me like we might say the following:

It appears unlikely that EGTB files or computer generated opening books will add
several hundred ELO to the crafty chess program when run under an autoplayer at
20min+5sec on {unknown hardware configuration} with 96 MB hash and 6 MB pawn
hash.

It seems likely that an opening book is more valuable than a set of EGTB files
for playing strength, but that is uncertain.

Your conclusions (since they are ceilings) may be fairly safe.  But learning
adds another monkey-wrench.




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.