Skip to content

Latest commit

 

History

History
43 lines (31 loc) · 3.09 KB

README.md

File metadata and controls

43 lines (31 loc) · 3.09 KB

What's in this repository?

This repository contains codes that I have reproduced (while learning RL) for various reinforcement learning algorithms. The codes were tested on Colab.

If Github is not loading the Jupyter notebooks, a known Github issue, click here to view the notebooks on Jupyter's nbviewer.


Implemented Algorithms

Algorithms Discrete Continuous Multithreaded Multiprocessing Tested on
DQN ✔️ CartPole-v0
Double DQN (DDQN) ✔️ CartPole-v0
Dueling DDQN ✔️ CartPole-v0
Dueling DDQN + PER ✔️ CartPole-v0
A3C (1) ✔️ ✔️ ✔️ ✔️(3) CartPole-v0, Pendulum-v0
DPPO (2) ✔️ ✔️(3) Pendulum-v0
RND + PPO ✔️ MountainCarContinuous-v0 (4), Pendulum-v0 (5)

(1): N-step returns used for critic's target.
(2): GAE used for computation of TD lambda return (for critic's target) & policy's advantage.
(3): Distributed Tensorflow & Python's multiprocessing package used.
(4): State featurization (approximates feature map of an RBF kernel) is used.
(5): Fast-slow LSTM with an overly simplified VAE like "variational unit" (VU) is used.


misc folder

The misc folder contains related example codes that I have put together while learning RL. See the README.md in the misc folder for more details.


Blog

Check out my blog for more information on my repositories.