Modelling Agent Policies with Interpretable Imitation Learning