Wine Classifier - MLOps Demo

Scikit-learn wine classification with a modern MLOps pipeline featuring MLflow tracking, Ray for distributed training and serving, hyperparameter optimization, and production-ready deployment patterns.
Explore OpenCloudHub »

---

📑 Table of Contents

1. About
2. Features
3. Getting Started
4. Usage
5. Project Structure
6. MLOps Pipeline
7. Contributing
8. License
9. Contact

---

🍷 About

This repository demonstrates a complete MLOps pipeline for wine classification using scikit-learn and the UCI Wine dataset. It showcases production-ready machine learning practices including experiment tracking, hyperparameter optimization, model registration, and containerized deployment.
Ray is used for distributed training and scalable model serving.

Key Technologies:

• ML Framework: Scikit-learn (Logistic Regression)
• Distributed Training & Serving: Ray
• Experiment Tracking: MLflow
• Hyperparameter Optimization: Optuna
• Containerization: Docker
• Dependency Management: UV
• Development: DevContainers for consistent environments

---

✨ Features

• 🔬 Experiment Tracking: MLflow integration with model registry
• 🎯 Hyperparameter Tuning: Automated optimization using Optuna
• 🐳 Containerized Training: Docker-based training environment
• ⚡ Distributed Training & Serving: Ray for scalable workflows
• 📊 Model Evaluation: Comprehensive metrics and visualization
• 🚀 CI/CD Ready: GitHub Actions workflows for automated training
• 📁 MLflow Projects: Standardized, reproducible ML workflows
• 🔄 Model Registration: Threshold-based automatic model promotion
• 🧪 Development Environment: VS Code DevContainer setup

---

🚀 Getting Started

Prerequisites

• Docker and Docker Compose
• VS Code with DevContainers extension (recommended)
• MLflow tracking server (for remote tracking)
• Ray (for distributed training/serving)

Local Development

1. Clone the repository

   git clone https://github.com/opencloudhub/ai-ml-sklearn.git
   cd ai-ml-sklearn

2. Open in DevContainer (Recommended)

   code .
   # VS Code will prompt to reopen in container

3. Or setup locally

   # Install UV
   curl -LsSf https://astral.sh/uv/install.sh | sh
   
   # Install dependencies
   uv sync --dev

---

MLflow Tracking Server

Start MLflow locally (accessible from Docker containers):

mlflow server --host 0.0.0.0 --port 8081
export MLFLOW_TRACKING_URI=http://0.0.0.0:8081
export MLFLOW_EXPERIMENT_NAME=wine-quality
export MLFLOW_TRACKING_INSECURE_TLS=true

---

Ray Development Workflow

1. Start a Local Ray Cluster

ray start --head

2. Training Workflows

Submit Ray jobs for training and hyperparameter optimization:

RAY_ADDRESS='http://127.0.0.1:8265' ray job submit --working-dir . -- python src/training/train.py
RAY_ADDRESS='http://127.0.0.1:8265' ray job submit --working-dir . -- python src/training/optimize_hyperparameters.py

3. Model Serving with Ray Serve

Make sure you have promoted a model to prod.wine-classifier with @champion alias, as service is looking for that To run the model serving application locally:

serve run --working-dir /workspace/project src.serving.wine_classifier:deployment

---

💻 Usage

Training

python src/training/train.py --C 1.0 --max_iter 100 --solver lbfgs

Hyperparameter Optimization

python src/training/optimize_hyperparameters.py --n_trials 50 --test_size 0.2

Local Model Serving

serve run --working-dir /workspace/project src.serving.wine_classifier:deployment

To test the model, run:

python tests/test_wine_classifier.py

You can also visit the Swagger documentetion of the Application at http://localhost:8000/docs

---

📁 Project Structure

ai-ml-sklearn/
├── src/
│   ├── training/                       # Training and optimization scripts
│   │   ├── train.py
│   │   ├── optimize_hyperparameters.py
│   │   └── evaluate.py
│   ├── serving/                        # Model serving (Ray Serve/FastAPI)
│   │   └── wine_classifier.py
│   └── _utils/                         # Shared utilities
│       ├── get_or_create_experiment.py
│       ├── logging_callback.py
│       └── logging_config.py
├── tests/                              # Unit tests
├── .devcontainer/                      # VS Code DevContainer config
├── .github/workflows/                  # CI/CD workflows
├── Dockerfile                          # Multi-stage container build
├── MLproject                           # MLflow project definition
├── pyproject.toml                      # Project dependencies and config
└── uv.lock                             # Dependency lock file

---

🔄 MLOps Pipeline

1. Development & Experimentation

   • Local development in DevContainers
   • Jupyter notebooks for data exploration
   • MLflow experiment tracking

2. Training & Optimization

   • Distributed training and hyperparameter tuning with Ray and Optuna
   • Model evaluation and metrics logging
   • Threshold-based model registration

3. Model Registry

   • Automatic promotion to staging registry
   • Model versioning and lineage tracking
   • Performance comparison and rollback capability

4. Deployment

   • Ray Serve for scalable, production-ready model serving
   • (Planned) KServe integration and GitOps-based deployment automation

---

👥 Contributing

Contributions are welcome! This project follows OpenCloudHub's contribution standards.

Please see our Contributing Guidelines and Code of Conduct for more details.

---

📄 License

Distributed under the Apache 2.0 License. See LICENSE for more information.

---

📬 Contact

Organization Link: https://github.com/OpenCloudHub

Project Link: https://github.com/opencloudhub/ai-ml-sklearn

---

🙏 Acknowledgements

• UCI Wine Dataset - The dataset used for classification
• MLflow - ML lifecycle management
• Optuna - Hyperparameter optimization framework
• Ray - Distributed computing and serving
• UV - Fast Python package manager

(back to top)