Skip to content

Modified versions of the SAC algorithm from spinningup for discrete action spaces and image observations.

License

Notifications You must be signed in to change notification settings

ac-93/soft-actor-critic

Repository files navigation

soft-actor-critic

This repo consists of modifications to the Spinningup implementation of the Soft Actor-Critic algorithm to allow for both image observations and discrete action spaces.

Trained Atari agents (courtesy of https://github.com/yining043):

BeamRider Enduro Breakout SpaceInvaders Qbert

Dependencies:

tensorflow 1.15.0
gym[atari] 0.15.7
cv2
mpi4py
numpy
matplotlib

Implentations of Soft Actor Critic (SAC) algorithms from:

  1. Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor, Haarnoja et al, 2018 https://arxiv.org/abs/1801.01290

  2. Soft Actor-Critic Algorithms and Applications, Haarnoja et al, 2019, https://arxiv.org/abs/1812.05905

  3. Soft Actor Critic for Discrete Action Settings, Petros Christodoulou, 2019, https://arxiv.org/abs/1910.07207 (authors implementation here: https://github.com/p-christ/Deep-Reinforcement-Learning-Algorithms-with-PyTorch)

Based on the implementations given in Spinningup

https://spinningup.openai.com/en/latest/algorithms/sac.html

Different approaches for discrete setting

Two different methods given for using SAC with discrete action spaces.

  • sac_discrete_gb uses the Gumbel Softmax distribtuion to reparameterize the discrete action space. This keeps algorithm similar to the original SAC implementation for continuous action spaces.

  • sac_discrete avoids reparmeterisation and calculate the entropy and KL divergence from the discrete actions given by the policy network. This is based on the method described in [3] and is most accurate to the original SAC papers, I also find best results with this method.

Versions of the algorithms that work with image observations such as the atari gym environments are in the image observation directory.

About

Modified versions of the SAC algorithm from spinningup for discrete action spaces and image observations.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages