https://github.com/amir22010/ppo-dash

PPO Dash: Improving Generalization in Deep Reinforcement Learning

https://github.com/amir22010/ppo-dash

Science Score: 10.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • CITATION.cff file
  • codemeta.json file
  • .zenodo.json file
  • DOI references
  • Academic publication links
    Links to: arxiv.org
  • Academic email domains
  • Institutional organization owner
  • JOSS paper metadata
  • Scientific vocabulary similarity
    Low similarity (5.4%) to scientific vocabulary
Last synced: 10 months ago · JSON representation

Repository

PPO Dash: Improving Generalization in Deep Reinforcement Learning

Basic Info
  • Host: GitHub
  • Owner: Amir22010
  • Language: Python
  • Default Branch: master
  • Size: 315 MB
Statistics
  • Stars: 0
  • Watchers: 1
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Fork of Sohojoe/ppo-dash
Created almost 7 years ago · Last pushed almost 7 years ago

https://github.com/Amir22010/ppo-dash/blob/master/

# PPO-Dash
Code for reproducing the results found in [PPO Dash: Improving Generalization in Deep Reinforcement Learning](https://arxiv.org/abs/1907.06704)

# About
PPO-Dash is a modified version of the PPO algrothem that utalises the following optimizations and best practices:
* Action Space Reduction
* Frame Stack Reduction
* Large Scale Hyperparameters
* Vector Observations
* Normalized Observations
* Reward Hacking
* Recurrent Memory

PPO-Dash was able to solve the first 10 levels of the Obsticle Tower Enviroment without the need for demonstrations or curosity based algorthemic enhancements.

The version of PPO-Dash in the technical paper, [placed 2nd](https://blogs.unity3d.com/2019/05/15/obstacle-tower-challenge-round-2-begins-today/) in Round One of the [Obsticle Tower Challenge](https://www.aicrowd.com/challenges/unity-obstacle-tower-challenge) with an average score of 10. We were able to reproduce this score in Round Two of the challenge, with a minor modifiaction (randomizing the themes during in training). We [placed 4th overall](https://www.aicrowd.com/challenges/unity-obstacle-tower-challenge/leaderboards), with a score of 10.8 with the addition of demonstrations.

# Reproducing Results
To reproduce the results listed in the paper and for round one of the competition, see [ReproduceRound1](ReproduceRound1.md)

# Acknowlegements
This codebase derives from [pytorch-a2c-ppo-acktr](https://github.com/ikostrikov/pytorch-a2c-ppo-acktr-gail) - [#8258f95](https://github.com/ikostrikov/pytorch-a2c-ppo-acktr-gail/commit/8258f95d6c1959d02c6a412415138b95c32837a0)

# Citation
If you use PPO-Dash in your research, we ask that you cite the [technical report](https://arxiv.org/abs/1907.06704) as a reference.

Owner

  • Name: Amir Khan
  • Login: Amir22010
  • Kind: user
  • Location: India

working on developing a state of art AI solutions mainly in computer vision, chat bots and nlp domain. building an awesome AI as a professional developer 😍.

GitHub Events

Total
Last Year