Crafty and Recent Developments in Computer Chess
Robert M. Hyatt,
Associate Professor
Department of
Computer and Information Sciences
The University of Alabama at Birmingham
hyatt@redmont1.cis.uab.edu
Abstract
Professor Hyatt will describe his highly acclaimed program Crafty
(ICS rating 2491), which incorporates many recent advances in
computer chess technology. These advances include:
- using parallel machines to search deeper into the game tree,
- improving the chess
knowledge contained in the program so that it plays better positional
and goal-oriented chess,
- improving search strategies so that the
program analyzes deeper in selected positions that require additional
search, and
- new fast algorithms for move generation.
Computer chess expert Professor Hyatt is the author of Cray Blitz,
World Computer Champion from 1983 to 1989, developed jointly with
Lawrence Livermore Laboratory and Cray Research, Inc.
Schedule
Wednesday, December 13, 1995
11am - Talk to general computer science audience
2pm - Specialized technical talk to computer chess seminar
Location
Both talks in ECS Room 210i (CSEE conference room),
UMBC
Everyone is welcome to both events.
Directions to UMBC.
Hosts
Alan T. Sherman
and James Mayfield
CMSC-791 Graduate Seminar in Computer Chess
Department of Computer Science and Electrical Engineering
UMBC
This event is held in cooperation with the
UMBC Chess Club.
Kasparov vs. Deep Blue
In February 1996, World Chess Champion
Gary Kasparov will play
IBM's Deep Blue in Philadelphia.
Notes
In reflecting on the success of Crafty, Professor Hyatt stressed the
importance of parallel search, selective-deepening search, null-move
heuristic, transposition tables, move-ordering, bit-board data
structures, and the significant feedback he received from playing
thousands of games on the ICC. Paradoxically, he also noted that the
null-move heuristic, and transposition tables in parallel search, are
also major sources of headaches. He described a new move-generation
algorithm based on bit-board structures in which collections of
moves are described by a single bitboard.
With his flexible Crafty source code, Professor Hyatt has been
able to experiment with variations in this algorithms. He noted that
what works for suites of chess problems does not always work well
against ICC players. He advises against playing a program against
itself because his experience shows that doing so inappropriately
accentuates minor differences in program versions. Professor Hyatt
believes strongly in testing ideas experimentally and gave examples of
theoretical results from the literature that he was unable to verify
experimentally. Also, he cautioned against the uncareful use of
certain quantitative benchmarks: for example, a high number of nodes
searched per second often is the result of a poor move-ordering
heuristic (and not the sign of a strong progam).