Predicting opponent resource allocations when qualitative and contextual information is not available

How one predicts another's behavior depends on the type of behavior being predicted and the context of the prediction. In this paper we describe an agent based on ELPH [1] for a two player, zero-sum game where success depends on predicting the opponent's resource allocation in a domain lacking qualitative and contextual information. This problem is made difficult in that many of the traits necessary for many opponent modeling algorithms do not exist (there is no meaningful context, all options are of equal value, there are no meaningful sequences, no signaling of intention, etc.), the agent's behavior changes significantly and frequently and the agent is actively trying to be unpredictable.