Training AI to play Tetris using Unity's Reinforcement Learning package ML-Agents. The SAC training algorithm was used for training with other hyperparameters defined in trainer_config.yaml.
This project was created using Unity 2019.4.0 LTS and ML-Agents release 2. Python 3.7 and the ml-agents 0.16.1 Python package is required for training a new model.
Launch mlagents-learn from the command line and run the Train.unity scene.
mlagents-learn config/trainer_config.yaml --run-id TetrisLearning
View the results using tensorboard.
tensorboard --logdir=summaries --port=6006 --bind_all
Test the generated model in the Play.unity scene, simply add the .nn file in the model field of behaviour parameters and check the Behaviour Type is set to Inference.
https://github.com/Unity-Technologies/ml-agents/blob/release_2/docs/Readme.md
https://github.com/Unity-Technologies/ml-agents/blob/release_2_verified/docs/Training-Configuration-File.md
https://codemyroad.wordpress.com/2013/04/14/tetris-ai-the-near-perfect-player/