ML CI Django Application

This project is a Django application for continuous integration of a machine learning workflow. It uses the wine quality dataset as the sample model target, Celery for asynchronous training jobs, and GitHub Actions to test the Django app and retrain the model whenever application code, training code, dependencies, or datasets change.

Project Structure

• mysite/ - Django project settings, URLs, WSGI, and Celery application setup.
• blog/ - Existing educational UI pages for the CI/ML introduction.
• ml_pipeline/ - ML workflow app with training run models, Celery tasks, training services, tests, views, and templates.
• train.py - Command-line entry point used by CI to train the model.
• wine_quality.csv - Sample dataset used by the model training workflow.
• .github/workflows/ci.yml - GitHub Actions pipeline for Django and model tests.

Local Development

Install dependencies:

pip install -r requirements.txt

Run migrations:

python manage.py migrate

Start Django:

python manage.py runserver

Open the training UI at:

http://127.0.0.1:8000/ml/

Celery Task Queue

Celery uses RabbitMQ by default:

celery -A mysite worker --loglevel=info

The broker can be changed with:

CELERY_BROKER_URL=amqp://guest:guest@localhost:5672//

For local development without manual setup, run the full stack:

docker compose up --build

This starts Django, a Celery worker, and RabbitMQ.

Model Training

Run the model training workflow directly:

python train.py --dataset wine_quality.csv --output-dir artifacts/local

The command writes:

• metrics.txt
• feature_importance.png
• residuals.png

The Django app stores queued training runs in ml_pipeline.TrainingRun, including status, Git SHA, metrics, errors, and artifact paths.

Continuous Integration

GitHub Actions runs on pushes and pull requests to main when relevant files change:

• Python source files
• requirements.txt
• wine_quality.csv
• files under data/ or datasets/
• the CI workflow itself

The CI pipeline runs:

python manage.py check
python manage.py test
python train.py --dataset wine_quality.csv --output-dir artifacts/ci

Model metrics and generated plots are uploaded as workflow artifacts.