MLflow Wine Quality Lab

Using MLflow for experimenting, tracking, and logging models.

Setup (Local)

1. Create and Activate Virtual Environment

python3.11 -m venv .mlflow-lab-3.11
source .mlflow-lab-3.11/bin/activate

2. Install Dependencies

pip install -r requirements.txt
pip install -e .

3. Run the Pipeline

./run.sh

This single command:

• Cleans old MLflow data (mlruns, mlflow.db, mlartifacts)
• Starts the MLflow UI in the background
• Loads and prepares wine quality data
• Trains a baseline Random Forest model
• Runs an XGBoost hyperparameter sweep (10 trials via Hyperopt)
• Registers and promotes the best model to Production
• Serves the production model and runs real-time inference

4. View Results

• MLflow UI — http://127.0.0.1:5000
• Runs — untuned_random_forest, xgboost_models (expand for nested trials), best_xgboost
• Model Registry — Click "Model registry" in sidebar for version history

Tip: Filter runs with MLflow syntax: tags.mlflow.runName = "xgboost_models"

5. Stop MLflow UI

kill $(lsof -t -i:5000)

---

Setup (Docker)

See DOCKER_SETUP.md for full instructions on running with Docker Compose.

---

Project Structure

MLflow/
├── src/
│   ├── __init__.py          # Package init
│   ├── data_prep.py         # Data loading and train/val/test splits
│   ├── train_model.py       # RF and XGBoost training functions
│   ├── run_pipeline.py      # Pipeline orchestration
│   ├── inference.py         # CLI inference client
│   └── fastapi-app.py       # FastAPI inference service (Docker)
├── data/                    # Wine quality CSV files
├── logs/                    # Pipeline, MLflow UI, and model server logs
├── run.sh                   # Local entry point
├── docker-compose.yml       # Docker Compose orchestration
├── Dockerfile               # Container image definition
├── .env                     # Environment variables (not tracked in git)
├── pyproject.toml           # Setuptools config for editable install
├── requirements.txt         # Python dependencies
└── starter.ipynb            # Original reference notebook

---

Manual Operation

Serve the Model

source .mlflow-lab-3.11/bin/activate
export MLFLOW_TRACKING_URI=http://127.0.0.1:5000
mlflow models serve -m "models:/wine_quality/production" -p 5001 --env-manager=local

Run Inference

python3 src/inference.py

Test with curl

curl -X POST http://127.0.0.1:5001/invocations \
  -H "Content-Type: application/json" \
  -d '{"dataframe_split": {"columns": ["fixed_acidity","volatile_acidity","citric_acid","residual_sugar","chlorides","free_sulfur_dioxide","total_sulfur_dioxide","density","pH","sulphates","alcohol","is_red"], "data": [[7.0, 0.25, 0.36, 1.6, 0.034, 30.0, 110.0, 0.9906, 3.24, 0.50, 12.8, 0]]}}'