16 Hyperparameters

16.1 Learning rate $\alpha$

Controls how much each new experience overwrites prior estimates.
Original value: $\alpha = 0.9$. The agent heavily trusts new samples, which works here because the environment is deterministic.

16.2 Discount factor $\gamma$

Determines how much future rewards are worth relative to immediate ones.
Original value: $\gamma = 0.9$. A reward $k$ steps away is worth $0.9^k$ of its face value. This ensures the cumulative sum converges and encodes a preference for reaching the goal sooner rather than later.

16.3 Exploration rate $\epsilon$ for $\epsilon$-greedy action selection

At any given step there’s a $\epsilon$% chance the agent takes a random action instead of the best known one.
This handles the explore-exploit tradeoff — without exploration the agent might never discover the goal at all.

16.4 Episodes

The number of complete run-throughs of the environment the agent gets to learn from.

16.5 Max steps

caps the length of each episode to prevent infinite loops in cases where the agent never reaches a terminal state.