Introspection-based Explainable Reinforcement Learning in Episodic and Non-episodic Scenarios