UmdTask443_DATA605_Spring2026_HuggingFace_Text_Classification_Model_1

[riyaapuri]

Related to #443

HuggingFace Text Classification - AGNews Dataset

Contains code deliverables for the end-to-end pipeline that fine-tunes DistilBERT on the AG News dataset (120K articles, 4 classes) for news article classification.
The full pipeline - data loading, preprocessing, training, interactive prediction, and model evaluation - is capable of running inside Docker with no local Python setup required.

---

Files Added

• config.py - Central config for all hyperparameters and constants
• utils/dataset_loader.py - Loads AG News from HuggingFace Hub, creates 90/10 train/val split
• utils/preprocessing.py - Cleans text (HTML, URLs) and tokenizes with AutoTokenizer
• utils/metrics.py - Trainer callback and sklearn classification report utility
• scripts/train.py - Fine-tunes DistilBERT; saves best checkpoint by macro-F1
• scripts/evaluate_model.py - Batch inference on test dataset; exports report, confusion matrix, metrics chart, predictions CSV
• scripts/predict.py - Single article, file, and interactive inference with confidence scores
• run.sh - Unified pipeline wrapper; forwards CLI flags to each step
• requirements.txt - Full Python dependency list
• Dockerfile + docker shell scripts - Full standalone Docker integration for all pipeline steps for efficiency and reproducibilty

If needed, please refer to the commit history on this branch for an iterative breakdown of changes and progress across development stages.

---

Authors: @riyaapuri @stupatel17
Reviewers: @protocorn @gpsaggese