Skip to content

Latest commit

 

History

History
28 lines (17 loc) · 1.88 KB

Training-the-Agent.md

File metadata and controls

28 lines (17 loc) · 1.88 KB

Training the agent

  1. Start AI Unity Simulator project in Unity and open the scene AISimulation.

  2. Start mlagents-learn with the following commands in your project's root folder. The <some-id> is a string tag you give to identify this training run.

    source venv/bin/activate     # Activate virtual environment
mlagents-learn config/trainer_config.yaml --run-id=<some-id>

  3. After running the commands go to the Unity window and press the Play button in Unity to start the training process.

  4. After the training is finished, the new trained brain is created to the results-folder's subfolder which is called the same as the <some-id>-tag you used in the --run-id-argument in the training command. The training process will run long, therefore, you might need to stop it by clicking the pause button in the Unity app.

  5. Now, we have a trained model ready. The trained Brain file is called PushBlock.nn. in the /results/<some-id>-folder

Training config file

The config/trainer_config.yaml file has two sections:

  1. behaviors
  2. environment_parameters

behaviors-section has the parameters to configure the neural network model. The environment_parameters-section is explained here

The training uses curriculum learning and the config file's environment_parameters-section has the parameters which change the simulation as the agent gets better. The environment_parameters-section is explained here

For more information about the training process and parameters please follow the documentation of Unity at the following link.