|
The GNU Backgammon download currently includes 15 match equity tables.
Based on the efforts of many peopIe, reported primarily in the
rec.games.backgammon and gammonline.com forums, I would currently
recommend using any of three of these fifteen with GNU Backgammon:
1. Woolsey-Heinrich Table. This is probably still the most widely used
table. I used to use this table with GNU Backgammon because it was
the table I was using over the board. It was created by Kit Woolsey
about 14 years ago by analyzing the thousand-plus tournament matches
in Hal Heinrich's database. However, it is not as accurate as the
newest computer generated tables, mainly because (1) insufficient
data in the database for all match scores necessitated a "mishmash of
empirical data, a program which was based on some assumptions which
may not be sound, a lot of judgment, and some fudging" by Woolsey
[http://groups.google.com/groups?selm=kwoolseyF7E4rs.Ds1@netcom.com]
to come up with reasonable numbers for all scores, and (2)
championship level tournament players play better now. Sharper play
and cube action result, particularly, in a higher gammon rate. When
the Woolsey table was made, a gammon rate of 20-21% was commonly
accepted. The latest bot-generated data find a gammon rate of 26-27%.
As a minor point, this table causes GNU Backgammon to make at least
one obvious post-Crawford cube error, as happened in one game in GNU
Backgammon's last confrontation with BG Blitz in a computer games
olympiad.
2. Snowie version 2.1 table. This table was introduced with the v2.1
patch in July 1999 but is still used today by Snowie version 4. I
started using this table with GNU even though I was still using the
Woolsey table over the board, because (1) it was an improvement over
the Woolsey table (I believe it assumes a gammon rate of 26%) and (2)
using this table with GNU aided comparison of GNU and Snowie
evaluations and rollouts by eliminating the difference in tables as a
source of difference in results.
3. G11 Table (Zorba Table). This is the table I use now. It was
generated by Robert-Jan Velduizen in 2003 and is solely based on
recursive GNUBG full rollouts of the opening position (0-ply play, 2-
ply 100%/25%) up to 9-away scores. It has a theoretical advantage
over the Snowie table because it was generated by (a) cubeful play
(b) according to score, with whatever effect that has at GG (Gammon
Go) and GS (Gammon Save) scores (-2,1 and -1-2 Crawford and similar
scores) particularly and consequent trickle-down effects on the
entire match table.
Among other tables:
Zadeh: The Zadeh table, first published in 1977, is the default when you
download GNU Backgammon, but users should change this immediately:
Settings/Options/Match/Load, followed by Settings/Save Settings
This default should be changed (to Woolsey or G11, probably); nobody
uses Zadeh, and any of the latest good tables is a better choice.
Ortega: The Ortega table was generated by Antonio Ortega and Danny
Kleinman by interpolating between Danny Kleinman's old match table and
Woolsey's table in order to generate three-decimal match equities,
which is useful for computer calculations, but there is no empirical
reason to believe the table is accurate, and it has been superseded by
newer tables.
Dunstan: The Dunstan table was generated with Gnubg 0.14 (2-Ply 50%
Move, 2-Ply 33% Cube) rollouts but only through 5-away scores so far.
This post shows how it differs from the G11 table.
Kazaross-Shaw: I believe that Neil Kazaross and Ian Shaw have stopped
work on the Kazaross-Shaw table in favor of the G11 table, which
Kazaross has endorsed.
Friedman: Roy Friedman's table from his 1989 book World Class Backgammon
Move by Move (still recommended reading) is not included in the GNUBG
download. The table initially enjoyed some popularity but was widely
criticized for assuming a gammon rate of 36%. The bots have
subsequently shown that this rate is much too high.
Jacobs-Trice and MEC26: The Jacobs-Trice and MEC26 tables are both good
and do not differ much from the G11 table. In fact there is not much
difference between any of the best tables, and the latest available
data generated by having GNU Backgammon play itself using different
tables found little difference in results between any of the best
match tables.
Bkgm.com archives much discussion of match tables.
[http://www.bkgm.com/rgb/rgb.cgi?menu+matchequities]
Woolsey's table may still be the best choice for most players over the
board because (1) it is easy to remember, (2) works with Neil's Numbers
[http://www.bkgm.com/rgb/rgb.cgi?view+171] and (3) a gammon rate of
20-21%% may be closer to what most players attain in actual play than
26-27%. Players interested in comparing match tables should pay
particular attention to 3-away and 4-away scores.
No claims for originality here, but browsing through r.g.bg I thought we
might be due for a summary.
|
![[GammOnLine forum]](pics/GammOnLine.gif)
![[Long message]](pics/Long.gif)
Long message
![[Recommended reading]](pics/Star2.gif)
Recommended reading
![[Recent addition]](pics/New.gif)
Recent addition