Book Learning - a Methodology to Tune an Opening Book Automatically

Article type: Research Article

Authors: Hyatt, R.M.^{a; 1}

Affiliations: [a] Birmingham, USA

Note: [1] Computer and Information Sciences, University of Alabama at Birmingham, 115A Campbell Hall, UAB Station, Birmingham, AL 35294-1170. Email: byatt@cis.uab.edu.

Abstract: The paper describes some experiments in book learning by the program CRAFTY, a computer-chess program developed by the author. CRAFTY plays on various Internet chess servers, and as a result, typically plays over 20,000 games per year. A favourite tactic of many human chess players is to find a weak book line that the program follows repeatedly. They play that opening over and over until they find a way to win the game. Thereafter they replay this game many times. Previously the author would simply edit the book by hand when such cooks were found, to prevent the program from playing these openings time after time. Recently the author has developed an approach that automates this book editing or book tuning by letting CRAFTY determine which openings are good and which are bad, and adjusting its book selections accordingly.

DOI: 10.3233/ICG-1999-22102

Journal: ICGA Journal, vol. 22, no. 1, pp. 3-12, 1999