Counting 3,834 Big Data & Machine Learning Frameworks, Toolsets, and Examples...
Suggestion? Feedback? Tweet @stkim1

Last Commit
May. 21, 2019
Mar. 5, 2019

Caution: Code is under active development. Breaking changes are probable.

Documentation Status


A pipeline for fast prototyping Deep RL problems using PyTorch and OpenAI's Gym / Deepmind's dm_control.

Digideep is written to be developer-friendly with self-descriptive codes and extensive documentation. It also provides some debugging tools and guidelines for implementing new methods. Digideep has the following methods implemented:

  • DDPG - Deep Deterministic Policy Gradient
  • SAC - Soft Actor Critic
  • PPO - Proximal Policy Optimization

Please use the following BibTeX entry to cite this repository in your publications:

  author = {Sharif, Mohammadreza},
  title = {Digideep: A pipeline for implementing reinforcement learning problems},
  year = {2019},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{}},


Please visit for documentation.


The features of Digideep can be listed as following:

  • Developer-friendly code:
    • The code is highly readable and fairly easy to understand and modify.
    • Extensive documentation to support the above.
    • Written for modularity and easy code reuse.
    • Provides debugging tools as an assistance for implementation new methods.
  • Has a single-node multi-cpu multi-gpu architecture implemented to utilize CPU and GPU on a single node.
  • Connects to dm_control and uses dm_control's native viewer for rendering.
  • Provides batch-environment for dm_control through OpenAI baseline's VecEnv wrapper.
  • Controls all parameters from one single parameter file for easier control.
  • Supports (de-)serialization structurally.


  • 2019-03-04: Digideep was launched.


This code is under BSD 2-clause (FreeBSD/Simplified) license. See LICENSE.


I would like to appreciate authors of OpenAI baselines, pytorch-a2c-ppo-acktr, RL-Adventure-2, and RLkit projects.