Have you ever bought a pet simulator game, just to realize it was boring? What if your virtual pet could be considered alive? Introducing: NEURALNET-GAME (rebrand coming soon!)

Challenging the definition of "sentient" and "alive," NEURALNET-GAME uses reinforcement learning to provide a virtual pet with a virtual brain, allowing your pet to "live," which provides an incentive to keep taking care of it.

HOW TO RUN:

• Clone the repo to a local drive, and then install the requirements in requirements.txt to a conda environment.
• To run the game in game mode, run "python neuralnetgame" in the project directory.
• To run the game in training mode, run the program like above but with the "--train" flag.

CLI PARAMETERS

• --train : Runs the program in training mode
• --train-duration [int] : How many episodes to train for
• --resume [str] : The name of a model file inside the models folder to resume training frome
• --report-interval [int] : How often (in steps) to report the current step and episode of the model
• --save-interval [int] : How often (in episodes) to save the model's current state
• --model-name [str] : The name of a model (in the models folder) to use when in play mode. Defaults to "pretrained-model"

TODO

• Finish Gymnasium environment
• Implement general TensorFlow functionality
• Switch to Pytorch because general tensorflow functionality means no functionality at all because tensorflow lacks an observer library for windows
• Finetune training settings
• Comfort features
  • Full click support for easier training
  • IN PROGRESS - Replayability to watch model progress
  • Game element (player interaction with the model - placing food and water around the map)