aphrodite91/imdb-tfidf-neural-net

IMDB Sentiment Analysis with TF-IDF + Neural Network

This project trains a simple neural network for binary sentiment analysis on an IMDB review dataset. It is designed for a GitHub CI/CD workflow that automatically trains the model and uploads the artifacts to Hugging Face Hub.

Project idea

The model uses:

1. TF-IDF text vectorization for movie reviews.
2. A feedforward neural network implemented with PyTorch.
3. Binary Cross Entropy with logits for positive/negative classification.
4. GitHub Actions for automatic training and deployment.

Repository structure

.
├── data/
│   └── imdb_balanced_10k.csv or imdb_top_500.csv
├── model/
│   ├── model.pt
│   ├── vectorizer.pkl
│   ├── config.json
│   └── metrics.json
├── train.py
├── predict.py
├── requirements.txt
└── .github/workflows/train-and-upload.yml

Dataset

Put one of these files inside the data/ folder:

• imdb_balanced_10k.csv
• imdb_top_500.csv

The code automatically detects common text columns such as review or text, and label columns such as sentiment or label. Labels can be positive/negative or 1/0.

Train locally

pip install -r requirements.txt
python train.py

The training script saves:

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

Predict locally

python predict.py "This movie is wonderful and exciting."

CI/CD deployment

On every push to main or master, GitHub Actions will:

1. Install dependencies.
2. Run python train.py.
3. Run a small prediction test.
4. Upload the trained model artifacts to Hugging Face Hub.

Before pushing, edit this line in .github/workflows/train-and-upload.yml:

repo_id = "YOUR_HF_USERNAME/imdb-tfidf-neural-net"

Replace YOUR_HF_USERNAME with your Hugging Face username.

Also add a GitHub Actions secret:

HF_TOKEN = your Hugging Face write token

Final submission

Submit:

• GitHub repository URL
• Hugging Face model repository URL