PyTorch0.4 implementation of: actor critic / proximal policy optimization / acer / ddpg / twin dueling ddpg / soft actor critic / generative adversarial imitation learning / hindsight experience replay
Switch branches/tags
Nothing to show
Clone or download
Latest commit 277966f Jun 21, 2018
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
common Adding init file Jun 8, 2018
1.actor-critic.ipynb Add files via upload May 26, 2018
2.gae.ipynb update learning rate May 30, 2018
3.ppo.ipynb update learning rate May 30, 2018
4.acer.ipynb Add files via upload May 26, 2018
5.ddpg.ipynb Fix plot issue Jun 20, 2018
6.td3.ipynb Add files via upload May 26, 2018
7.soft actor-critic.ipynb Add files via upload May 26, 2018
8.gail.ipynb Add files via upload May 26, 2018
9.her.ipynb Add files via upload May 26, 2018
README.md Correct link to A3C Paper Jun 8, 2018

README.md

RL-Adventure-2: Policy Gradients

PyTorch tutorial of: actor critic / proximal policy optimization / acer / ddpg / twin dueling ddpg / soft actor critic / generative adversarial imitation learning / hindsight experience replay

The deep reinforcement learning community has made several improvements to the policy gradient algorithms. This tutorial presents latest extensions in the following order:

  1. Advantage Actor Critic (A2C)
  1. High-Dimensional Continuous Control Using Generalized Advantage Estimation
  1. Proximal Policy Optimization Algorithms
  1. Sample Efficient Actor-Critic with Experience Replay
  1. Continuous control with deep reinforcement learning
  1. Addressing Function Approximation Error in Actor-Critic Methods
  1. Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor
  1. Generative Adversarial Imitation Learning
  1. Hindsight Experience Replay

If you get stuck…

  • Remember you are not stuck unless you have spent more than a week on a single algorithm. It is perfectly normal if you do not have all the required knowledge of mathematics and CS.
  • Carefully go through the paper. Try to see what is the problem the authors are solving. Understand a high-level idea of the approach, then read the code (skipping the proofs), and after go over the mathematical details and proofs.

RL Algorithms

Deep Q Learning tutorial: DQN Adventure: from Zero to State of the Art N|Solid Awesome RL libs: rlkit @vitchyr, pytorch-a2c-ppo-acktr @ikostrikov, ACER @Kaixhin

Best RL courses

  • Berkeley deep RL link
  • Deep RL Bootcamp link
  • David Silver's course link
  • Practical RL link