Moody Learners -- Explaining Competitive Behaviour of Reinforcement Learning Agents

Designing the decision-making processes of artificial agents that are\ninvolved in competitive interactions is a challenging task. In a competitive\nscenario, the agent does not only have a dynamic environment but also is\ndirectly affected by the opponents' actions. Observing the Q-values of the\nagent is usually a way of explaining its behavior, however, do not show the\ntemporal-relation between the selected actions. We address this problem by\nproposing the \\emph{Moody framework}. We evaluate our model by performing a\nseries of experiments using the competitive multiplayer Chef's Hat card game\nand discuss how our model allows the agents' to obtain a holistic\nrepresentation of the competitive dynamics within the game.\n