Text this: Discretization and approximation methods for reinforcement learning of highly reconfigurable systems /