This repository utilizes the architecture posed by the paper "CDT: Cascading Decision Trees for Explainable Reinforcement Learning" by Zihan Ding et al, and encapsulates it within a boosted forest architecture for a light-weight methodology that can create powerful models for accurate feature importances. This repository also implements type (a) of the hierarchical structures posed in section 3.2 under the heading "Hierarchical CDT".
This architecture is applied to a stock trading environment which utilizes an easily swappable strategy to distill the state space down to strong indicators for price movements which benefit or hurt the strategy.