Hands-On Reinforcement Learning with Python

Understanding the algorithm

Now, we will see how DQN works overall. The steps involved in DQN are as follows:

1. First, we preprocess and feed the game screen (state s) to our DQN, which will return the Q values of all possible actions in the state.
2. Now we select an action using the epsilon-greedy policy: with the probability epsilon, we select a random action a and with probability 1-epsilon, we select an action that has a maximum Q value, such as .
3. After selecting the action a, we perform this action in a state s and move to a new state s' and receive a reward. The next state, s', is the preprocessed image of the next game screen.
4. We store ...