astooke / accel_rl

Accelerated Methods for Deep Reinforcement Learning

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Accelerated Methods for Deep Reinforcement Learning

Code behind "Accelerated Methods for Deep Reinforcement Learning" by Adam Stooke and Pieter Abbeel: https://arxiv.org/abs/1803.02811. Includes single- and multi-GPU versions (both synchronous and asynchronous learners) of A2C, PPO, DQN + variants--all optimized for speed--along with an optimized implementation of the Atari environment (using ALE).

Borrows from the original rllab (including installation instructions), with documentation at: https://rllab.readthedocs.org/en/latest/.

Examples are in accel_rl/scripts/examples, and include bare "train" scripts, which can be called directly, and associated "run" scripts, which invoke multiple train scripts to run experiment sweeps.

About

Accelerated Methods for Deep Reinforcement Learning


Languages

Language:Python 95.4%Language:HTML 1.8%Language:JavaScript 1.5%Language:Ruby 0.6%Language:CSS 0.4%Language:Mako 0.2%Language:Shell 0.1%