Competitive training methods for reinforcement learning using RLLIB
- Python 3.6
-
Optionally create a Python3 virtualenv called
venv
(separate project dependencies)virtualenv -p python3 venv
-
Activate the virtualenv (if you created one)
source venv/bin/activate
-
Install dependencies
pip install -r requirements.txt
Activate the virtualenv (if used)
source venv/bin/activate
Run using main.py and configure via optional command line arguments
python main.py {options}
Flag | Parameters | Description | Required | Default Value |
---|---|---|---|---|
run | str | An optional name of this experiment run | N | '' |
epochs | int | The number of experiment epochs to run | N | 1 |
env | str | The game environment to use | N | Connect4 |
agent | str | The agent types to use | N | Deep Q-Network (DQN) |
trainer | str | The agent training method to use | N | DefaultTrainer |
save-agent | bool | Whether to save all agents weights/memory/progress. | N | False |
human-player | bool | Allows you to compete against trained agents. | N | False |
visualise | bool | Whether to show a visualisation of the game. | N | False |