Some chess stuff
- This is the specification of the Polyglot
opening book format. This is the most widely understood non-proprietary
opening book format. It is at least understood by Polyglot, Fruit, Toga,
Glaurung, Hamsters, Mediocre, TwistedLogic, jose-chess and the latest versions of Scid. I am providing this specification
in the hope that other programs will also use this book format. UPDATE: Polyglot books are now also understood by Winboard/Xboard 4.3.15.
- This is a proposal for adding a header to Polyglot books in an extensible and backwards compatible way. Here is a utility
for handling such headers.
- A new version of icsdrone (called icsdroneng). It has a number of new features (such as
feedback on tell commands). It is not quite finished (and perhaps never will be)
but nonetheless rather useful. It is used by a number of bots on FICS.
Update: the latest versions of icsdroneng are now available on sourceforge. Update2: these days almost all bots on FICS use icsdroneng 0.29 (the latest version). So there must be something good in it!
- I made an attempt to transform GNU Chess
5.07 into a modern chess engine, both on Windows, Linux and
Android. It now supports UCI and polyglot books as well as Gaviota
Table Bases. Meanwhile the Free Software Foundation has released
GNU Chess 6
. Ironically the latest descendant of GNU Chess 5.07
GNU Chess 5.60 with its ancient code base is some 180 Elo stronger than
GNU Chess 6! Update: I renamed GNU Chess 5.60 to GnuCheese 1.00 to avoid confusion with GNU Chess 6.
- Here is a little python script that may help you to design truncated sequential likelihood ratio tests. I derived
myself. Hopefully they are correct (I did do a few sanity checks on their validity).
- Here is a simplified version of the previous script. It does a lot less but it does not have any external dependencies.
- Here is a script to experiment with the awesome 2-SPRT.
- This is a practical
introduction to the
with an emphasis on chess engine testing.
- The motivating example in the previous document assumes that we use
the match score as a test statistic for comparing engines. Is this still the best choice when strongly biased opening positions are used? This document says yes, provided we assume the Davidson elo model.
- computeLLR.py is a little Python script
which implements the LLR computation described in GSPRT_approximation.pdf for the 3-nomial and 5-nomial model for engine matches.
- normalized_elo.pdf explains how to compare opening books with different characteristics.
- Posix ports of recent versions of Toga II. They have been reported to work on Linux, MACOSX and BSD.
- An Android version of Toga 1.4.1SE. This is a raw binary which needs a GUI (see below). Source code: Toga141SE-android.tar.gz. The only change is that this version of Toga loads the Scorpio bitbases probing dll egbbso.so from
the current directory, rather than from the bitbases directory (presumably the bitbases will be located on external storage).
- An Android version of RobboLito 0.085g3l. This is a raw binary which needs a GUI (see below). Source code: robbolito-android.tar.gz. This is basically a quick
and dirty modification of the original source. Since I don't know ARM assembler, I simply replaced the assembler instructions by
Ucibench. A small benchmarking utility for uci engines.
- Please consider donating to Project Mahi, a cool project to send an autonomous boat to the Caribbean, run by my son and three of his engineer friends.
michel (dot sign) vandenbergh (at sign) uhasselt (dot sign) be