Sample Match-3 RL Environment

This repository contains a sample match-3 game RL environment. The environment is developed using Unity game engine and Unity ML-Agents.

The environment can either be used as a Unity ML-Agents environment or as a Gym environment using wrappers.

The environment reward can be configured using a configurating file. The objective of this environment is to maximize the number of matches in limited number of actions which in this case is 15 actions.

Requirements

Windows machine with python 3.7 or newer installed. The environment is provided as a windows binary so it can not run on other operating systems.

Configuration

Inside Build folder, there is a file titled config.json. Only one field, RewardMode exists which can take following values:

• 0 : Immediate match reward. Does not containg reward from gravity and cascaded matches.
• 1 : Cascade mode. Does not contain immediate reward. Contains reward gained due to gravity and cascaded matches.
• 2 : Total mode. Total reward gained

How to use?

1. Create a python virtual environment. (Optional but recommended)
2. Install required packages using requirements.txt file.
3. Investigate random_agent.py file as an example of how to use this environment.

There is a custom Unity to Gym environment wrapper inside lib folder, which only differs from Unity's official wrapper in reset method. Custom wrapper's reset returns two values : (state,action_mask)

Match-3 board size is 8x8 with 5 tile types(colors). The state is 6 binary planes. Layers 1-5 marks each color type in binary. Layer 6 marks tiles that can be part of a valid move.

Good Luck!