Iterative Oblique Decision Trees Deliver Explainable RL Models

The demand for explainable and transparent models increases with the continued success of reinforcement learning. In this article, we explore potential generating shallow decision trees (DTs) as simple surrogate opaque deep learning (DRL) agents. We investigate three algorithms training data axis-parallel oblique DTs help DRL agents (“oracles”) evaluate these methods on classic control problems from OpenAI Gym. results show that one our newly developed algorithms, iterative training, outperforms traditional sampling resulting in well-performing often even surpass oracle which they were trained. Even higher dimensional can be solved surprisingly DTs. discuss advantages disadvantages different insights into decision-making process made possible by nature Our work contributes to development not only powerful but also RL highlights a effective alternative complex models.