Q-Learning Agent - Taxi-v3

This model was trained from scratch using Q-Learning for the Hugging Face Deep Reinforcement Learning Course, Unit 2.

Evaluation

• Environment: Taxi-v3
• Evaluation episodes: 100
• Mean reward: 7.56
• Standard deviation: 2.71
• Certification score: 4.85
• Required score: 4.50

Training

• Algorithm: Q-Learning
• Training episodes: 50,000
• Learning rate: 0.8
• Discount factor (gamma): 0.95
• Initial epsilon: 1.0
• Minimum epsilon: 0.05
• Epsilon decay rate: 0.0001

The trained Q-table is stored in q-learning.pkl.