This paper defines a metric for comparing play traces (sequences of user decisions) in a game-independent way. The properties of this metric are determined by a proposed tool we call the Gamalyzer, an exploratory visualization of arbitrary games which clusters together similar play traces. Our Gamalyzer metric is based on refinements to edit distance and has broad uses outside of visualization and, indeed, outside of games (e.g. player and opponent modeling, goal recognition, player mimicry,

more »

... r testing, and so on). We validate our metric against one synthetic and one real-world data set, finding that Gamalyzer discerns designer-relevant differences between play traces nearly as well as hand-tuned feature selection while remaining game-and genre-agnostic. The main problem with conventional game visualizations is that judging the similarity of two game states is an underspecified problem requiring knowledge of game rules and of the purpose for which the states are being compared. Gamalyzer, in contrast, directly compares sequences of actions: it considers strategies instead of states. Existing game visualizations require significant development effort for individual genres, games, and even specific queries. Our proposed visualization (based on the Gamalyzer metric) could quickly and inexpensively show designers the strategic landscape of their game without requiring specialized knowledge of statistics or machine learning.