Description
Drop7 is a puzzle game originally developed by Area/Code Entertainment. I recreated it with C# as a console game with the purpose of implementing machine learning. As a final project in the "AI in Gameplay" course at RIT in Fall 2022, I added a reinforcement learning agent to play the game itself. This was carried out with a method called Q-Learning. While the implementation was successful, the agent did not perform much better than a random agent due to the game's enormus state space. That said, I have plans to add a neural network to the Q-Learning for the sake of space and performance.
As a Developer
This was uncharted territory for me. Reinforcement learning was not something I had done before but had always wanted to. With the recent publishing of Deep Mind's Agent57 demolishing 57 Atari 2600 games, I was inspired to find simple game as a case study. This was a great oportunity to dig into machine learning and put my programming skills to the test. During this project my eyes were opened to the use case of value vs reference types. Saving data points during runtime and exporting it to an excel document needed static data or else the data would become corrupt.
Tools Used
C#, .Net 6.0