Adaptive Genetic Algorithm to Optimize the Parameters of Evaluation Function of Dots-and-Boxes

Abstract

Designed an evaluation function with parameters, and used genetic algorithm to optimize the parameters. This paper considers the objective function’s variation trends in searching point and the information is added to the fitness function to guide the searching. Simultaneously adaptive genetic algorithm enables crossover probability and mutation probability automatically resized according to the individual’s fitness. These measures have greatly improved the convergence rate of the algorithm. Sparring algorithm is introduced to guide the training, using gradient training programs to save training time. Experiments show skills in playing Dots-and-Boxes are greatly improved after its evaluation function parameters are optimized.