First, let's us recall some basic elements of reinforcement learning that we discussed in the first chapter:
- State: The state space defines all the possible states of the environment. In Atari games, a state is a screen image or a set of several consecutive screen images observed by the player at a given time, indicating the game status of that moment.
- Reward function: A reward function defines the goal of a reinforcement learning problem. It maps a state or a state-action pair of the environment to a real number, indicating the desirability of that state. The reward is the score received after taking a certain action in Atari games.
- Policy function: A policy function defines the behavior of the ...