GRIFFIN

GRIFFIN is a PyTorch-based neuro-fuzzy inference system designed specifically for classification tasks. It builds upon the UNFIS framework but introduces several key enhancements:

	•	Feature Selection – automatically learns which input features are most relevant.
	•	Relaxation Mechanism – smooths rule activations for improved generalization.
	•	Literal Modulation – adds flexibility in handling fuzzy membership functions.
	•	Correlation-Aware Design – captures dependencies among input features for more expressive rules.

By combining these mechanisms, GRIFFIN provides a robust and interpretable alternative to black-box neural models in classification problems.

📂 Project Structure

GRIFFIN/
├── model/                  # Core model definitions
│   └── griffin.py          
├── train/                  # Simple training / evaluation scripts
│   ├── classification_test.py
│   └── early_stop.py
├── tests/                  # Dataset wrappers + quick tests
│   ├── uci_test.py         # UCI data-set helpers
│   └── class_test.py       # Digits / Segmentation helpers
├── notebooks/              # Jupyter demo (no heavy dependencies beyond the requirements file)
├── requirements.txt
└── .gitignore
                            # Datasets (excluded from repo)

🚀 Installation

1. Clone this repository:

   git clone https://github.com/your-username/GRIFFIN.git
   cd GRIFFIN

2. Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate   # on Linux/Mac
   .venv\Scripts\activate      # on Windows

3. Install dependencies:

   pip install -r requirements.txt

📁 Data

Datasets are expected in a data/ directory at the project root. This folder is not included in the repository due to size. Place your datasets there before running training/testing.

🧪 Tests

Unit tests are located in GRIFFIN/tests/. They include both classification and UCI dataset examples.