/transfer_rl_icml_2018

A codebase for experimenting with various approaches to action priors.

Primary LanguagePython

Policy and Value Transfer for Lifelong RL

Code for experimenting with transfer approaches to lifelong RL, attached to our 2018 ICML paper Policy and Value Transfer for Lifelong Reinforcement Learning

Experiments require simple_rl, which can be installed with the usual:

pip install simple_rl

To reproduce all of our plots, run run_all_policy_experiments.py (Figure 2), run_all_learning_experiments.py (Figure 3 -- you must choose your learning algorithm in the file to generate Figures 3 (a,d), (b,e), or (c,f)).

Authors: David Abel, Yuu Jinnai, Yue Guo, George Konidaris, Michael L. Littman.

Contact Dave or Yuu with any questions.