xiyou521/snakes_maddpg

Multi-agent learning on an OpenAI snake game implementation.

Snakes MAL

This is the codebase for using the "Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments" https://arxiv.org/pdf/1706.02275.pdf code for the classic "snakes" game. This was created as part of my CISC889-010: Multi-Agent Learning class in the Spring of 2019 semester at the University of Delware. I don't support any of the code, but please take a look and hopefully you can gain some insight into what needs to be done to apply maddpg to a different application area. The run_project.py script is the main script that runs the learning algorithm on the simulated enviroment and visualizes it to the user. Please take a look at the test scripts for simpler versions of the code.

Original Codebase Credit

• Both the original maddpg and mgym are taken from these repositories:
• https://github.com/openai/maddpg
• https://github.com/cjm715/mgym
• I recomend you check those repositories out for details

Install Instructions

1. Create your virtual enviroment

   • pip install virtualenv
   • virtualenv venv
   • source venv/bin/activate

2. Install dependencies

   • pip3 install pyqtgraph (for fast plotting)
   • pip3 install pyqt5 (for fast plotting)
   • pip3 install numpy
   • pip3 install gym (open ai gym)
   • Ensure you have cuda 9: https://www.tensorflow.org/install/gpu
   • pip3 install tensorflow-gpu==1.9.0 (cuda version 9.0)

3. Install mgym

   • A local copy has been included in this repo
   • https://github.com/cjm715/mgym
   • cd mgym
   • pip install -e .
   • cd ..
   • It should be installed

4. Install maddpg

   • A local copy has been included in this repo
   • https://github.com/openai/maddpg
   • cd maddpg
   • pip install -e .
   • cd ..
   • It should be installed

5. Test our installs:

   • I have included a few test scripts that have minimal example.
   • Run each test script and see if you get any loading errors
   • python3 test_mgym.py
   • python3 test_maddpg.py
   • python3 test_tensorflow.py