HillBoost-Ensemble

A GPU-powered ensemble learning algorithm using hill climbing optimization to combine models for optimal predictive performance.

Built for speed. Tuned for performance. Inspired by chaos.

---

🚀 What is Hill Boost?

Hill Boost Ensemble finds the best-performing model on a dataset, and then iteratively adds other models only if they improve the overall performance. Think of it as smart ensembling with a ruthless edge.

Currently supports:

• ✅ XGBoost
• ✅ LightGBM
• ✅ CatBoost
• ✅ Keras-based MLP

Note: GPU is strictly required. This repo is born and raised on the Kaggle GPU runtime.

---

🧠 How It Works

1. Train all candidate models on the same dataset
2. Evaluate performance using a custom metric (e.g., AUC, F1, accuracy)
3. Start with the best single model
4. Iteratively test all other models — include only if ensemble improves
5. Final prediction is the weighted sum of selected models

---

📦 Setup

Clone the repo:

git clone https://github.com/rwtarpit/HillBoost-Ensemble.git
cd HillBoost-Ensemble

🤝 Contributing

We welcome contributions!

Check out CONTRIBUTING.md for setup steps, ideas, and good first issues to work on.