https://github.com/denys88/rl_animal

AnimalAI Winner Solution
https://github.com/denys88/rl_animal

Last synced: 29 days ago
JSON representation

AnimalAI Winner Solution

• Host: GitHub
• URL: https://github.com/denys88/rl_animal
• Owner: Denys88
• License: mit
• Created: 2019-07-28T19:39:09.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2022-11-21T21:39:32.000Z (about 2 years ago)
• Last Synced: 2024-11-26T11:49:36.598Z (about 1 month ago)
• Language: Python
• Homepage: https://twitter.com/DenysM88
• Size: 697 MB
• Stars: 6
• Watchers: 2
• Forks: 2
• Open Issues: 6
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Animal AI 1st Place Solution Presented on NIPS 2019 (my nickname is Trrrrr)

## Challenge website and results :http://animalaiolympics.com/results.html

## My NIPS 2019 Presentation [Animal AI Presentation.pdf](/Animal%20AI%20Presentation.pdf)

I used this exact code, and only made slight changes before sharing.

I added linux enviroment binaries by default. I tested everything on Linux only.

You might have some problems to run training on other platforms.

## Steps to Run My Code:

1) pip install -r [requirements.txt](/requirements.txt)

2) Run [download_networks.py](/download_networks.py) to download my networks. 

3) There are [hyperparams.py](/hyperparams.py) with all parameters which I used during training process. Please change BASE_DIR = '/home/trrrrr/Documents/github/ml/rl_animal' in hyperparams.py to your folder before run anything.

We were allowed to submit two networks and currently I don't know which one has won.

You can find different game configurations in [game_configurations.py](../blob/master/games_configurations.py)

You can find networks in [networks.py](/networks.py)

You can use [Player.ipynb](/Player.ipynb) to run my agents

To run my validation please use [Validation.ipynb](/Validation.ipynb)

To run train please use [test_a2c.ipynb](/test_a2c.ipynb)

There is chance that you will not be able to run my code with current parameters. Please try to reduce batch size and number of enviroments in this case.