Python Deep Learning - Second Edition

Q-learning

Q-learning is an off-policy temporal-difference reinforcement learning algorithm. What a mouthful! But fear not, let’s not worry about what all this means, and instead just see how the algorithm works. To do this, we’ll use the game of chess we introduced in the previous section. As a reminder, the board configuration (the locations of the pieces) is the current state of the environment. Here, the agents can take actions, a, by moving pieces, thus changing the state into a new one. We'll represent a game of chess as a graph where the different board configurations are the graph’s vertices, and the possible moves from each configuration are the edges. To make a move, the agent follows the edge from the current state, s, to a new ...