session3-practice-huggingface-cicd

IMDB Sentiment Neural Network with GitHub CI/CD

This project trains a neural network sentiment classifier using TF-IDF features and a PyTorch feedforward neural network.

Project Structure

session3-practice-huggingface-cicd/
├── data/
│   ├── README.md
│   └── imdb_balanced_10k.csv
├── model/
│   └── .gitkeep
├── .github/
│   └── workflows/
│       └── train-and-upload.yml
├── train.py
├── predict.py
├── upload_to_hf.py
├── requirements.txt
├── README.md
└── .gitignore

Dataset

The committed training dataset is:

• data/imdb_balanced_10k.csv

It contains 10,000 balanced IMDB movie reviews sampled from the Stanford Large Movie Review Dataset:

https://ai.stanford.edu/~amaas/data/sentiment/

The training script also supports this fallback file name:

• data/imdb_top_500.csv

The training script automatically detects common text columns (review, text, sentence, comment) and label columns (sentiment, label, target).

Model

• TF-IDF vectorizer
• Feedforward neural network
• Binary sentiment classification

The model uses TF-IDF features as input to a PyTorch multilayer perceptron for positive vs. negative sentiment prediction.

Training

Run training locally with:

python train.py

Training saves these artifacts in model/:

• model.pt
• vectorizer.pkl
• config.json
• metrics.json

Prediction

Run inference from the command line with:

python predict.py "This movie is amazing!"

CI/CD

Every push to main or master triggers GitHub Actions to train the model and upload artifacts to Hugging Face Hub.

The workflow:

1. Installs dependencies
2. Runs python train.py
3. Saves model artifacts
4. Uploads artifacts to Hugging Face using HF_TOKEN

Hugging Face Hub

https://huggingface.co/fyangmie/session3-practice-huggingface-cicd

Final Submission

Submit the following:

• GitHub Repository: https://github.com/fyangmie/session3-practice-huggingface-cicd
• Hugging Face Repository: https://huggingface.co/fyangmie/session3-practice-huggingface-cicd