/rl_animal

AnimalAI Winner Solution

Primary LanguagePythonMIT LicenseMIT

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

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
  2. Run download_networks.py to download my networks.
  3. There are 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 You can find networks in networks.py You can use Player.ipynb to run my agents To run my validation please use Validation.ipynb To run train please use 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.