Reinforcement-Learning-on-FrozenLake

Reinforcement Learning on FrozenLake is a collection of jupyter files that you can learn and try basic reinforcement learning algorithms.

This repo is written for people who want to quickly learn basic concepts of Reinforcement Learning with code.

💡 Features

• Easy explanation of RL concepts
  • This book contains key part of the book "Reinforcement Learning: An Introduction"(pdf) by Richard S. Sutton and Andrew G. Barto.
  • The lecture "Introduction to Reinforcement Learning with David Silver"(link) is also referred.
• Interactive RL algorithm
  • You can also run RL algorithms in FrozenLake-v1, a OpenAI Gymnasium environment, with hyperparmeter customization.
  • Wrapper for the environment can render not only the environment, but also state-action value or model of the algorithms.

  

☝️ Requirements

• python >= 3.6
• gymnasium >= 0.26.1
• pygame >= 2.3.1
• tensorflow >= 2.8 (for Chapter 7)
• tensorflow-probability >= 0.22.1 (for Chapter 7)
• ipykernel
• ipython
• numpy
• matplotlib

You can install the requirements by using Poetry.

git clone https://github.com/moripiri/Reinforcement-Learning-on-FrozenLake.git
cd Reinforcement-Learning-on-FrozenLake

poetry install #--with ch7 #(If you want to run chapter7.ipynb, optional tensorflow dependency have to be installed.
poetry run python -m ipykernel install --user --name [virtualEnv] --display-name "[displayKernelName]"

where [virtualEnv] is the name of the python environment(ex. rl-introduction-py3.9) and "[displayKernelName]" is the jupyter kernel name you want (ex. frozenlake).

Then run

poetry run jupyter notebook

to run jupyter files.

☝️ Running in Google Colab

If you run any jupyter file in google colab, run the following commands first.

!git clone https://github.com/moripiri/Reinforcement-Learning-on-FrozenLake.git
%cd Reinforcement-Learning-on-FrozenLake/
!pip install gymnasium[classic_control]==0.26.3

📖 Contents

Chapter1: Introduction to Reinforcement Learning

Chapter2: Markov Decision Processes

Chapter3: Dynamic Programming

• Policy iteration
• Value iteration

Chapter4: Model-Free Prediction

• Monte-Carlo Prediction
• TD(0)

Chapter5: Model-Free Control

• On-policy Monte-Carlo Control
• SARSA
• Q-learning

Chapter6: Eligibility Traces

• SARSA(λ)

Chapter7: Policy Gradient Methods

• REINFORCE
• Actor-Critic

Chapter8: Integrating Learning and Planning

• Dyna-Q