Hands-On Q-Learning with Python

Adding an updated epsilon value

The last hyperparameter we want to add in is epsilon, so that our agent has the ability to explore new actions it hasn't taken yet and balance out its exploitation of the high-valued actions it's already taken. (For more detail on this, refer back to the exploration versus exploitation sections in Chapter 1, Brushing Up on Reinforcement Learning Concepts, and Chapter 2, Getting Started with the Q-Learning Algorithm.)

First, we choose an epsilon value between 0 and 1, say 0.1. We'll test and compare different values for epsilon as part of our model-tuning process.

We add the epsilon value to our algorithm and add a condition to the game loop to modify the way we choose actions:

Q = np.zeros([env.observation_space.n, ...