Quick Start Guide

Note: This quick start guide is only applicable for using this RL Subsystem as a standalone. If using in conjunction with the REALM_AI's game plugin, the below steps are already automated for you.

Building the environment

The first step is to build a ML-Agents compatible Unity environment, and build it into an executable. Here is a tutorial of building a Unity ML-Agents compatible environment from the official ML-Agents documentation.

Remember where the path of the executable is, because we will need to use it in the later steps!

Installing REALM_AI's RL-Subsystem

In terminal, enter the following commands to pull and setup the python package (e.g., install the necessary dependencies).

1. git clone git@github.com:realm-ai-project/RL-Subsystem.git

2. cd RL-Subsystem

3. pip install -e .

Training

While what makes realm-tune powerful is its configurability, to get started it is as simple as performing the following command in the terminal:

realm-tune --env-path <env_path>

where <env_path> is replaced with the path to the unity executable built above.

If you see a lot of output to the console, it means the program is working as expected! You may see a tiny game window show up from time to time, and that is to be expected!

If you wait for a while, the program will eventually be completed, and all training outputs will be stored at runs/RealmTune_xx-xx-xxxx_xx-xx-xx, where the x represents a datetime stamp so that your training outputs are always easily recognizable. Congrats, you have tuned your first agent!

Learn More

There are much more that realm-tune can do that will be covered in the subsequent pages, such as

• configuring hyperparameters to tune over

• picking the hyperparameter tuning algorithms

• integrate with Weights and Biases with wandb-mlagents-learn

• automatically initiate the full training run after the hyperparameter tuning stage, using the best set of hyperparameters

• many more