A collection of code for learning, using, evaluating, and visualizing abstractions in Reinforcement Learning in Python. This code is used to run experiments for our 2018 ICML paper: State Abstractions for Lifelong Reinforcement Learning and the earlier workshop paper Toward Good Abstractions for Lifelong Learning, presented at the NIPS Hierarchical Reinforcement Learning Workshop in 2017.
Experiments require simple_rl, which can be installed with the usual:
pip install simple_rl
For the ICML paper, run run_icml_learning_experiments.py to reproduce plots from Figure 3a/3b and Figure 4. Run run_icml_planning_experiments.py to reproduce plots from Figure 5. Run chain.py to reproduce the plot in Figure 2.
Authors: David Abel and Dilip Arumugam. Let us know if you have issues!