rock-pebble-sand-distilbert-al

RPS DistilBERT Active Learning Model

This repository contains a DistilBERT-based text classification model trained using Active Learning with a Stabilizing Predictions stopping criterion. The pipeline incrementally labels the most uncertain samples from an unlabeled pool to maximize performance while minimizing labeled data requirements.

The project includes:

• Data loading and preprocessing
• Tokenization and encoding for Hugging Face transformers
• Active learning loop with uncertainty sampling
• Stabilizing Predictions stopping criterion
• Evaluation metrics (Accuracy, Macro F1, Precision, Recall)
• Learning curve visualization

---

Table of Contents

• Installation
• Dataset
• Active Learning Pipeline
• Usage
• Evaluation
• Saving and Loading the Model
• Learning Curve
• Contributing
• License

---

Installation

This project requires Python 3.9+ and the following packages:

pip install torch torchvision torchaudio
pip install transformers datasets evaluate scikit-learn matplotlib python-dotenv