EXE\headless is headless, which means there's no graphic rendering.
EXE\Client has graphic.
Two of Exe run at 10 times speed.
More Document about MLagent is at https://github.com/Unity-Technologies/ml-agents/blob/release_19_docs/docs/Readme.md
Useful Doc:
- API Docs/Python API Documentation
- API Docs/How to use the Python API
- Python Tutorial with Google Colab/Using a UnityEnvironment
- Python Tutorial with Google Colab/Q-Learning with a UnityEnvironment
- create an enviroment with Python 3.6 or 3.7
- Install the pytorch from https://pytorch.org/get-started/locally/
- Install the mlagent with pip
python -m pip install mlagents==0.28.0
- Install importlib-metadata
pip install importlib-metadata==4.4
More Installation Detail at https://github.com/Unity-Technologies/ml-agents/blob/release_19_docs/docs/Installation.md
Run the MLAgent Default Model(PPO/SAC) by Anaconda command prompt under the folder with exe
mlagents-learn <config path> --env=<exe name> --run-id=<run_name>
It should be like
mlagents-learn config\player_config.yaml --env="CRML" --run-id=test
To load a Unity environment from a built binary file, put the file in the same directory as enviroment(exe), run:
from mlagents_envs.environment import UnityEnvironment
# This is a non-blocking call that only loads the environment.
env = UnityEnvironment(file_name="CRML", seed=1, side_channels=[])
# Start interacting with the environment.
env.reset()
behavior_names = env.behavior_specs.keys()
...more Details at https://github.com/Unity-Technologies/ml-agents/blob/release_19_docs/docs/Python-API.md
SPEC of observation and action in under each version's folder.