WinePy is a Python implementation of the Wine Gomoku (Five in a Row) AI engine. This project enhances the original C++ implementation with improved AI algorithms, a user-friendly interface, and a data generation system for neural network training.

• 💪 Powerful Gomoku AI based on alpha-beta search with pattern recognition
• 🎮 Interactive console-based user interface
• 🤖 Computer vs. computer demonstration mode
• 📊 Training data generator for machine learning models
• 🧠 Pre-generated AI training dataset available on Hugging Face
• 🏆 Supports Gomocup protocol for competition

• Python 3.6 or higher
• NumPy (for data generation and storage)

1. Clone the repository:

   git clone https://github.com/Karesis/Winepy.git

2. Install the required dependencies:

   pip install numpy

Run the main program to access the menu:

python main.py

Select option 1 to play against the computer. Input moves using coordinates like h8 (column + row).

• quit or exit: Exit the game
• restart or new: Start a new game
• help: Show help information

Select option 2 from the main menu to watch two AI players compete against each other.

Select option 3 from the main menu to generate self-play games for AI training:

1. Specify the number of games to generate
2. Choose the number of parallel processes
3. Set the output directory
4. Select the output data format

The data generator creates a dataset suitable for training neural networks to predict optimal Gomoku moves.

Select option 4 to enter Gomocup protocol mode, which allows the AI to interact with other programs using the standard Gomocup protocol.

• ai.py: Implementation of the AI engine and search algorithms
• board.py: Game board representation and pattern evaluation
• main.py: User interface and main program
• data_generator.py: Training data generation for machine learning

WinePy uses an enhanced alpha-beta search algorithm with the following features:

• Iterative deepening search
• Transposition tables for move caching
• Pattern-based evaluation function
• Threat detection and response
• Improved move ordering for better pruning

The AI recognizes and evaluates various patterns on the board:

• Connected five (win condition)
• Open four (one move away from winning)
• Blocked four (potential four-in-a-row with one end blocked)
• Open three (two moves away from winning)
• Blocked three (potential three-in-a-row with one end blocked)
• Open two (early stage pattern)
• Blocked two (early stage pattern with one end blocked)

The data generator creates training examples in two formats:

1. Full board representation (15×15 grid, ML-friendly NumPy format)
2. Sparse representation (original format with move history)

Each training example consists of a board state and the corresponding optimal move as determined by the AI.

Pre-generated Dataset: A large dataset of 26,378 examples from 875 games is available on Hugging Face. This dataset is split into training (80%) and test (20%) sets and is ready for machine learning experiments.

Contributions are welcome! Here are some ways you can contribute:

• Improve the AI evaluation function
• Enhance the user interface
• Optimize the search algorithm
• Add support for different board sizes
• Implement a graphical user interface

• Original Wine Gomoku AI: Wine by jinjiebang
• Python implementation and enhancements: [Karesis]

This project is licensed under the MIT License - see the LICENSE file for details.