Imitating Agent Game Strategies Using a Scalable Markov Model

Humans exhibit regularities in almost everything they do. We describe a Markov model derived from the behavior patterns of an agent, which is used to determine strategies by predicting which action a user is likely to execute next. We evaluate the predictive accuracy of this approach on a large dataset collected from sample Wumpus World games. We demonstrate from this approach that, the model can correctly predict the user’s next action with minimal computation and memory resources. Such predictions can then be used to imitate player strategies in a variety of games and other strategic domains.