Derivative Evaluation Function Learning Using Genetic Operators
David H. Lorenz, Shaul Markovitch · 1993
This work studies the application of genetic algorithms to the domain of game playing, emphasising on learn-ing a static evaluation function. Learning involves ex-perience generation, hypothesis generation and hypoth-esis evaluation. Most learning systems use preclassified examples to guide the search in the hypothesis space and to evaluate current hypotheses. In game learning, it is very difficult to get classified examples. Genetic Algorithms provide an alternative approach. Compet-ing hypotheses are evaluated by tournaments. New hy-potheses are generated by genetic operators. We intro-duce a new framework for applying genetic algorithms to game evaluation-function learning. The evaluation function is learned by its derivatives rather than learn-ing the function itself. We introduce a new genetic op-erator, called derivative crossover, that accelerates the search for static evaluation function. The operator per-forms cross-over on the derivatives of the chromosomes. We have demonstrated experimentally the advantage of tile derivative crossover for learning an evaluation func-tion.