Hands-On Artificial Intelligence for Beginners

Choosing action

Thus far, we've told our network to follow random weights that we've initialized for it, without giving it direction on how to decide what actions to take to update those weights. In the case of policy methods such as the ones we used in Chapter 8, Reinforcement Learning, and preceding with Tic Tac Toe, Q-learning methods work toward approximating the value of the Q-function instead of learning a policy directly. So, what do we do?

Deep Q-networks use a tactic called exploration to determine what actions to take. If we didn't use a tactic, our network would probably be limited to the most basic levels of the game because it wouldn't have any idea what moves would allow it to improve!

To remedy this, we will utilize a strategy ...