Learning Reinforcement Learning

A collection of Jupyter notebooks implementing core Reinforcement Learning (RL) concepts and algorithms using Gymnasium.

Contents

• 01_Basic_RL.ipynb: Introduction to MDPs and Dynamic Programming.

  • Policy Evaluation & Improvement
  • Policy Iteration (PI)
  • Value Iteration (VI)
  • Environment: FrozenLake-v1

• 02_Multi_Arm_Bandits.ipynb: Exploration vs. Exploitation strategies.

  • Pure Exploitation/Exploration
  • Optimistic Initial Values
  • $\epsilon$-greedy strategy
  • Upper Confidence Bound (UCB)
  • Thompson Sampling
  • Environment: BanditTenArmedGaussian-v0

Environment Setup

The project uses conda environment. Key dependencies include:

• numpy
• gymnasium
• matplotlib
• tqdm
• gym-bandits
• gym-walk