Text this: Exploring how different state representations and configurations affect the learning process and outcome of deep Q-learning algorithms