Modelling the dynamics of multi-agent learning has long been