Pommerlution

A place to experiment with various reinforcement learning approaches in the Pommerman environment.

The main goals of this project are leveraging self-play for training, evolutionary model selection of the best agents and exploring the usability of object embeddings for the game state.

Prerequisites

• Python 3.9

Development Setup

First clone this repository to a directory of your choice and move into it:

git clone https://github.com/tgru/pommerlution
cd pommerlution

Then it can be installed with development dependencies enabled and in editable mode, so code changes will take effect immediately:

pip install -e .[dev]

Usage

Currently only checkpointless training of a single DQN agent in the PommeFFACompetition-v0 environment is supported. All scripts can be found in the bin folder, while the library code resides in the pommerlution folder.

Training is done with the train.py script. Hyperparameters can be set there with the respective arguments. For example:

python ./bin/train.py --episodes 1000 --opponent simple_agent

To get a list of all possible parameters and their description use the --help switch:

python ./bin/train.py --help