mincheolseong / multiagent-DQN

Multiagent DQN for collective learning

Note

This is in Python 3.5. Requirements: in requirements.txt + OpenCV

Setting up

Set up virtualenv

http://python-guide-pt-br.readthedocs.io/en/latest/dev/virtualenvs/

Download required packages

Activate your virtual environment, then issue this command:

pip install -r requirements.txt

Change euclid.py file

Euclid has some syntax that doesn't work in Python 3, so you need to change the file in the installed packages. Go to your virtualenv folder > lib > python 3.5 > site-packages. Then replace the euclid.py file there with the one in this repo.

Set up IPython kernel to work with the virtualenv

Make sure your virtualenv is activated. Then issue the following commands:

(env)$ pip install jupyter
(env)$ pip install ipykernel
(env)$ python -m ipykernel install --user --name testenv --display-name "Python3 (dqn-virtualenv)"

Files

Jupyter notebooks

training

testing_5preds

testing_independents

analysis

Python files

tf_rl/simulate.py

tf_rl/simulation/karpathy_game.py

Issues

1. Inactive predators cannot eat
2. There can only be one physical cue (but several cue types)