Hands-On Intelligent Agents with OpenAI Gym

Reinforcement learning

Reinforcement learning is kind of a hybrid way of learning compared to supervised and unsupervised learning. As we learned at the start of this section, reinforcement learning is driven by a reward signal. In the case of the kid with their homework problem, the reward signal was the chocolate from their parents. In the machine learning world, a chocolate may not be enticing for a computer (well, we could program a computer to want chocolates, but why would we? Aren't kids enough?!), but a mere scalar value (a number) will do the trick! The reward signals are still human-specified in some way, signifying the intended goal of the task. For example, to train an agent to play Atari games using reinforcement learning, the ...