This is my implementation of the MultiStep MNIST Game from Learning to Communicate with Deep Multi-Agent Reinforcement Learning using Python and Torch.

In order to obtain the results showed in the report execute the following commands:

• Set up the environment:

python -m venv games-env
source games-env/bin/activate
pip install -r requirements.txt

• To replicate the experiments varying the sigma:

python main.py -conf conf.json -parameter sigma

• To replicate the experiments varying the gamma:

python main.py -conf conf.json -parameter gamma

Original Paper

Heavily inspired by