Skip to content
Code for Policy Consolidation for Continual Reinforcement Learning
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
README.txt
multi_vec_normalize.py
multiagent_monitor.py
play_history.py
policies.py
run_selfplay.py
run_single.py
tc.py

README.txt

Code for Policy Consolidation for Continual Reinforcement Learning
Accepted at ICML 2019
arXiv link: https://arxiv.org/abs/1902.00255

Python version used: anaconda3-5.1.0 
Requirements: numpy, tensorflow 1.7.0, gym, baselines, gym_extensions, robosumo, gym-compete

- tc.py contains the main classes and functions for the PC model
- run_single.py and run_selfplay.py are the run files used for training agents in single agent and selfplay environments respectively
- play_history.py is the run file for playing a self-play agent against its history or against other agents
You can’t perform that action at this time.