A collection of A.I. implementations, created by TeamiumPremium, for a competitive variant of Snake. For the course: Artificial Intelligence Through Machine Game Playing, Spring 2021.

Yes, we know the name "TeamiumPremium" is terrible great. It keeps things light.

TeamiumPremium: Directory containing the AI implementations and associated code (helper classes).

AI-Game-Playing: Directory containing the Snake environment, provided by https://github.com/mlepinski/AI-Game-Playing (main branch).

• Neuroph
• Deep Learning 4 Java (beta7)
• JavaFX

Claustrophobium: AI that chooses a random move in an empty or food space (i.e., no snakes). This is the best performing A.I.

NeuroPhremiumPlayer: Deep learning network using Neuroph. It was abandon due to issues getting Neurph to train multiple games.

DeepQiumPlayer: Deep-Q network implemented in Deep Learning 4 Java. Random moves use Claustrophobium to select a "safe" move. It learns to move north.

ConvolutiumPlayer: Same as DeepQium, but uses a few convolution layers. It learns to move north.

A note on encoding... for neural networks, spaces are one-hot encoded as empty, enemy snake body, enemy snake head, self head, self body, or food (6 total).

The Snake environment is handled using git subtree (as opposed to submodule). Subtree was chosen over submodule for usability when the Snake environment is updated. Subtree simply requires the repository to be updated. Submodule requires all contributors to update, making user-error likely.

Subtree also allows the repository to be cloned without extra commands.

The subtree can be manually updated with: git subtree pull --prefix AI-Game-Playing https://github.com/mlepinski/AI-Game-Playing.git main --squash

Time permitting, I hope to replace this manual update with a GitHub Action.