Q Learning vs PPO Machine Learning

Play demo here: https://rainyelephant.itch.io/ai-demo-q-learning-vs-ppo-machine-learning

Features:

• Q Learning fully implemented from scratch
  • Ability to customize these settings:
    • number of episodes
    • number of episode steps
    • exploration probability
    • minimum exploration probability
    • exploration decay
    • alpha
    • gamma
    • number of states
    • number of actions
    • seconds between turns (should be 0 in practice, but useful to slow down for debugging)
  • All user must do is inherit from QAgent and define these two functions:
    • GetNextState(int currentState, int action)
      • Within this function user must execute the action and return the resulting state
    • GetReward(Vector2 oldPosition, Vector2 newPosition, int nextState)
      • Return a reward based on state or based on position
      • If more is required in your implementation, edit the header to pass in more data
  • User can also overload certain important functions:
    • Start()
    • Update()
    • Iterate()
    • OnEpisodeBegin()
    • OnEpisodeEnd()
• 2 sample Q learning implementations
  • 4 direction move to goal
  • Tank/Spaceship controls move to goal
• Live Q Learning training screen that displays Q Table and training data in real time
• Ability to graph values and track performance by editing values under "Data Science"
• Ability to change timescale
• PPO ML Agent using Unity ML Agents Package for same scenarios Q Learning uses
• This project is open source, feel free to use it in your projects!