Extracting Learning Curves From Puzzle Games

Abstract

Game design is a time-consuming endeavor and, in contrast to other work in game development, has no tools aiding the designer in speeding up the process. In this paper, the initial work towards developing such a tool is presented. It focuses on the extraction of learning curves from single-player, deterministic, turn-based (puzzle) games. Learning curves in such games can be directly associated with their gameplay quality. The extraction is done by training a number of varying size, feed-forward neural networks on the game transition model. The main contribution of the work is the introduction of learning curve based puzzle game design concepts and their practical examples along with a quantitative method for learning curve estimation.