AI Reinforcement Learning with CartPole

This project implements a Deep Q-Network (DQN) agent that learns to balance a pole on a cart using reinforcement learning techniques. The agent is trained in the classic CartPole environment from OpenAI's Gym library.

Project Overview

This project demonstrates:

• Fundamentals of reinforcement learning
• Building and training neural networks with Keras
• Creating a DQN agent using Keras-RL2
• Training an agent to solve the CartPole balancing problem
• Saving and loading trained models

Prerequisites

• Python 3.8
• Basic understanding of neural networks and reinforcement learning concepts

Dependencies

The following libraries are required with specific versions:

gym==0.25.2
keras==2.10.0
keras-rl2==1.0.5
tensorflow==2.3.0
numpy==1.18.5
protobuf==3.20.0

Installation

1. Clone this repository:

git clone https://github.com/yourusername/AI-Reinforcement-Learning.git
cd AI-Reinforcement-Learning

Running the Project

Recommended: Using VSCode with Jupyter Extension

1. Install Visual Studio Code from https://code.visualstudio.com/
2. Install the Jupyter extension for VSCode
3. Open the project folder in VSCode
4. Open the Reinforcement Learning.ipynb notebook file
5. Run the cells sequentially to see the agent learn and perform

Project Structure

• Reinforcement Learning.ipynb: Main Jupyter notebook containing all code and explanations
• dqn_weights.h5f: Saved model weights for the trained agent
• checkpoint: TensorFlow checkpoint file

What You'll Learn

• Fundamental concepts of AI and reinforcement learning
• How to set up and interact with environments in OpenAI Gym
• Building neural network architectures with Keras
• Creating and training DQN agents
• Testing and visualizing agent performance
• Saving and loading model weights

Key Components

1. Environment Setup: Initializing the CartPole environment
2. Model Building: Creating a neural network with Keras
3. Agent Creation: Implementing a DQN agent with Keras-RL2
4. Training: Teaching the agent to balance the pole
5. Testing: Visualizing the agent's performance
6. Model Persistence: Saving and loading trained weights

Credits

This project is a remaster of Nicholas Renotte's workshop on RL which this repo serves to maintain for future learners. The original workshop can be found at YouTube - Nicholas Renotte.