Hands-On Q-Learning with Python

Utility functions and optimal decisions

It's helpful to think of reinforcement learning agents, and all optimizing agents, as operating under the constraint of a utility function. In the case of our Q-learning agents, the utility functions they're working with are what we call the policies they're following. We can think of the Q-value of an action as its utility.

A utility in real-life terms is a reward, such as money, that either can or can't be easily quantified. It is the reward an agent is pursuing and gauging its actions and decisions against. For real-world learning agents such as human beings, utility functions are obviously complex and involve many different kinds of motivations and drivers that interact with each other in ways that ...