Anticipatory Classifier System with Episode-based Experience Replay

** anonymized for Triple-blind submission needs **

The paper describes an attempt to design and evaluate modification to the Experience Replay (ER) extension of the anticipatory ACS2 classifier system. The modification is called Episode-based Experience Replay (ERR) and its main premise is to replay entire episodes instead of single experience samples.

Promising results verified by Bayes estimations are obtained on multi-step problems, albeit limited to deterministic and discrete tasks. The experimental results show that Episode-based Experience Replay extension improves significantly the learning capabilities of the ACS2.

2022_acs2_with_episode-based_experience_replay.zip -> contains jupyter notebooks and python scripts for experiments evaluation and analysis.

pyalcs.zip -> the content of pyalcs library containing the LCS and ACS models' definitions
https://github.com/ParrotPrediction/pyalcs/

openai-envs.zip -> definition of additional openai envs
https://github.com/ParrotPrediction/openai-envs