Reinforcement learning refers to a paradigm where an agent learns from environment feedback by virtue of receiving observations and rewards in return for actions it takes. The following diagram captures the feedback-based learning loop of reinforcement learning:
Although mostly applied to learn how to play games, reinforcement learning has also been successfully applied in digital advertising, stock trading, self-driving cars, and industrial robots.
In this chapter, we will use reinforcement learning to create a PacMan game and learn about reinforcement learning in the process. We will cover the following topics:
- Reinforcement learning
- Reinforcement learning versus supervised and unsupervised learning
- Components of reinforcement learning
- OpenAI Gym
- A PacMan game in OpenAI Gym
- DQN for deep reinforcement learning:
- Q Learning
- Deep Q Network
- Applying DQN to a PacMan game
Let's get started!