Paper Info
Title
Optimizing the performance of GNU-chess with a genetic algorithm
Abstract
We apply an artificial intelligence method based upon a distributed simple genetic algorithm which optimizes by "learning from a mentor" to enhance the performance of the open-source, free gnu-chess program. The genetic algorithm manipulates and optimizes the collection of numerical constants within the gnu-chess program which are used to assign values to chess pieces and their positions. These are built into the code of the evaluation-function for game positions. The above collection of numerical constants within the gnu-chess program constitutes, in principle, the genotype of creatures (candidate solutions) while the compiled program using these constants constitutes the phenotype underlying the genetic algorithm used in our approach. In every generation of the genetic algorithm, the so-generated phenotypes in the current population play a fixed number of games against the original gnu-chess program in order to determine their so-defined fitness-value. After up to 17,000 hours of distributed computation-time for a single optimization on a network of 17 linux workstations, the genetic algorithm finds a chess program that shows a moderate performance improvement compared with the original gnu-chess program. What appears to be new in the approach presented here are: (a) a brute-force optimization using a mentor rather than a co-evolutionary approach is actually carried out with contemporary PC hardware, and (b) optimizations for playing white and black are carried out separately which seemingly has not been attempted by other means before.
Year
Venue
Keywords
2010
Humans and Computers
co-evolutionary approach,free gnu-chess program,simple genetic algorithm,chess program,genetic algorithm,numerical constant,gnu-chess program,original gnu-chess program,chess piece,brute-force optimization,optimization
Field
DocType
Citations 
Creatures,Population,Computer science,Meta-optimization,Workstation,Artificial intelligence,Population-based incremental learning,Genetic algorithm,Performance improvement
Conference
0
PageRank 
References 
Authors
0.34
10
2
Name
Order
Citations
PageRank
Tomohiko Mitsuta100.34
Lothar M. Schmitt211612.00