Q-Learning Algorithm Implementation Example

I would like to share a practical implementation of the Q-learning algorithm to provide valuable insights for developers and researchers. Q-learning is a reinforcement learning algorithm designed to determine optimal policies within given environments by observing agent-environment interactions. This example demonstrates how to implement Q-learning for a pinball game scenario. The implementation covers environment setup using state-space representation, reward function design with positive/negative reinforcement mechanisms, and Q-table initialization with exploration-exploitation balance through epsilon-greedy policy. The code includes key functions for state transitions, Q-value updates using the Bellman equation (Q(s,a) = Q(s,a) + α[r + γmaxQ(s',a') - Q(s,a)]), and policy extraction. Complete annotated code is provided with detailed comments explaining the learning rate (α), discount factor (γ), and convergence criteria to help readers understand the algorithm's mechanics and implementation approaches. This example serves as a practical foundation for understanding Q-learning and applying it to future projects involving decision-making systems.