🧠 Strategic Board Game AI - Deep Learning Edition

📖 Overview

This project implements an intelligent AI agent for a 7x7 strategic board game. Instead of relying on traditional runtime search algorithms like Minimax or Alpha-Beta Pruning, this AI is powered by Supervised Deep Learning (Behavioral Cloning). The model instantly evaluates board states using a learned Value Function, making optimal decisions in milliseconds.

🚀 Technical Highlights

• Methodology: Behavioral Cloning via Supervised Learning.
• Data Generation: Synthetic, high-quality training data generated through Self-Play using a custom "Pro Teacher" algorithm.
• Teacher Logic: Employs a 2-Step Lookahead and a Weighted Mobility (Killer) Heuristic (heavily penalizing opponent mobility while securing central control).
• Model Architecture: A deep Multi-Layer Perceptron (MLP) built with PyTorch. Configured with 4 hidden layers (256 -> 128 -> 64 -> 32) to capture complex positional patterns.
• Inference Engine: Generates all valid (Move + Remove) combinations, evaluates the resulting board states through the Neural Network, and selects the optimal move instantly—without any recursive tree search at runtime.
• Architecture Design: Fully decoupled logic, modular configuration via YAML, and highly optimized for rapid inference.

📁 Project Structure

• collect_data.py: Generates the dataset (game_data.csv) by running the Teacher bot against itself.
• Train_Model.py: Trains the PyTorch model using the generated dataset (supports CUDA for GPU acceleration and early stopping).
• NeuralNet.py: Contains the BoardEvaluator MLP architecture.
• MainSBG.py: The playable GUI game loop utilizing the trained ai_brain.pth model.
• config.yaml / ConfigLoader.py: Centralized configuration for hyperparameters, network size, and file paths.

🛠️ How to Run

1. Install Dependencies:

   pip install torch pandas numpy pyyaml tqdm