A simple program where the 'Agent' plays the Frozen Lake game of the OpenAI gym environment. The 'Agent' tries to maximize it's rewards using Q Learning.
For continuous 'Actions' we use a Discounted reward system, so that the Agent prioritizes the present reward rather than some future reward.
where γ(gamma) is the discounted value
We use the Bellman equation to find the optimal 'q' value
where α(alpha) is the Learning Rate
We use the ε-Greedy algorithm to determine whether the 'Agent' would Explore the environment or Exploit the past information of the environment, gathered overtime. We then we Exponential Decay to reduce the Exploration rate so that at some later timestep, the 'Agent' would select to Exploit rather than Exploring the Environment.