🤖 General AlphaZero

A generalized, high-performance reinforcement learning agent based on DeepMind's AlphaZero algorithm.

---

🌟 Overview

This repository provides a clean, modular, and highly parallelized implementation of the AlphaZero algorithm. Originally introduced by DeepMind in their groundbreaking paper Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm, this implementation represents a generalized game-playing agent capable of mastering symmetric zero-sum games solely through self-play and reinforcement learning.

By combining Monte Carlo Tree Search (MCTS) with a deep neural network that evaluates board states and policy distributions, this agent iteratively improves without any human data or domain-specific heuristics.

✨ Key Features

• 🧠 Generalized Self-Play: The agent plays against itself, dynamically generating high-quality training data.
• ⚡ Parallelized Execution: Leverages multiprocessing for tree search and self-play (Alpha_MCTS_Parallel.py, Alpha_Zero_Parallel.py), massively accelerating data generation.
• 🎮 Multi-Game Support: Pluggable game environment architecture. Easily extensible to various board games (e.g., TicTacToe, ConnectFour).
• 📊 Intuitive Web Interface: Includes a fully-featured Streamlit Web UI for seamless interaction, configuration, and game monitoring.
• 💾 Automated Checkpointing: Built-in utilities (save_games.py, train_from_saved_games.py) for caching game histories and resuming training runs seamlessly.
• 🏟️ Competitive Arena: Pit different neural network models against each other in Arena.py to evaluate ELO and true model performance.

📚 Foundational Research

The implementation is heavily inspired by the following foundational research papers:

• Mastering the game of Go without human knowledge
• Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm
• Human-level control through deep reinforcement learning

🛠️ Installation

Prerequisites

• Python: Version 3.12 or higher is strictly required for optimal performance.

Setup Instructions

1. Clone the repository:

   git clone https://github.com/amanmoon/general_alpha_zero.git
   cd general_alpha_zero

2. Create and activate a virtual environment:

   # Create a virtual environment natively with Python 3.12+
   python3 -m venv venv
   
   # Activate on macOS/Linux
   source venv/bin/activate
   
   # Activate on Windows
   venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

🚀 Usage Guide

🎨 Run the Streamlit Application UI

We provide a beautiful, interactive web interface to interact with the models and environments.

1. Ensure your virtual environment is active and all dependencies are installed.
2. Launch the Streamlit server:

   streamlit run app.py

3. Open the provided localhost URL in your browser to start exploring!

🏋️‍♂️ Train a New Model

1. Open Train.py.
2. Import the desired game class and configure the hyperparameters dictionary (args).
3. Execute the training script:

   python3 Train.py

   Tip: Use train_from_saved_games.py to bootstrap learning from previously serialized MCTS explorations!

🕹️ Play Against the Agent

Want to test your skills against the neural network?

1. Open Play.py.
2. Select your desired model checkpoint and configure the MCTS simulation count for the AI's turns.
3. Run the script:

   python3 Play.py

⚔️ Model Arena (Evaluating Checkpoints)

To definitively prove which iteration of your neural network is superior, let them battle:

1. Open Arena.py and assign the respective paths to the models you wish to evaluate.
2. Run the tournament:

   python3 Arena.py

📁 Repository Architecture

├── Games/                     # Game logic implementations (TicTacToe, Connect4, etc.)
├── Alpha_MCTS.py              # Core Monte Carlo Tree Search logic
├── Alpha_MCTS_Parallel.py     # Asynchronous MCTS implementation
├── Alpha_Zero.py              # Self-play and neural network training loops
├── Alpha_Zero_Parallel.py     # High-performance parallelized self-play
├── Arena.py                   # Model vs Model evaluation environment
├── Play.py                    # Human vs AI interactive script
├── Train.py                   # Entry point for training from scratch
├── app.py                     # Streamlit frontend application
└── requirements.txt           # Python dependency specifications

📫 Contact

If you are a recruiter, researcher, or just an enthusiast who wants to discuss reinforcement learning, AI architecture, or optimal search algorithms, I'd love to connect!

Email: amanmoon099@gmail.com

---

"The only way to learn is by playing."