bomberman_rl

Setup for a project/competition amongst students to train a winning Reinforcement Learning agent for the classic game Bomberman.

setup conda env

create conda env by using

conda create --name bomberman python=3.8
pip install numpy tensorflow tqdm keras pygame tables matplotlib

play and train

python main.py play --my-agent dqn

starts the game and executes our own trained agent

python main.py play --my-agent dqn --train 1 --no-gui --n-rounds=200

starts the training process (python evaluate_training.py create must be called before training if evaluation is enabled)

evaluate training

1. In parameters.py: choose FILENAME, EVALUATION=True
2. python evaluate_training.py create Attention: Already existing files are overwritten!
3. train model as described above
4. python evaluate_training.py plots the data