MLflow in Practice — Live Demos

Demo notebooks for the "MLflow in Practice" talk at Cloud Native Trento.

All demos use the Jena Climate dataset (daily mean temperature) with ARIMA models for time series forecasting.

Prerequisites

• Python 3.10+
• Docker & Docker Compose

Quick Start

# 1. Install dependencies
pip install -r requirements.txt

# 2. Start MLflow server (PostgreSQL backend)
docker compose up -d --build

# 3. Open MLflow UI
open http://localhost:5050

# 4. Launch Jupyter
jupyter notebook notebooks/

Demo Sequence

#	Notebook	Topic
0	00_download_data.ipynb	Download & preprocess Jena Climate data
1	01_experiment_tracking.ipynb	Experiment tracking: params, metrics, artifacts
2	02_optuna_tuning.ipynb	Hyperparameter tuning with Optuna + nested runs
3	docker-compose.yml	Infrastructure: MLflow server + PostgreSQL
4	03_model_registry.ipynb	Model Registry: versioning & aliases
5	04_model_serving.ipynb	Model serving: REST API & Docker
6	05_production_setup.ipynb	Production: Supabase + Azure Blob

Notebook 00 is to download the data. Notebooks 01-04 each work standalone as long as data/jena_daily_temp.csv exists.

Inspecting the local PostgreSQL Database

# Connect to the local PostgreSQL
docker compose exec postgres psql -U mlflow -d mlflow

# Quick queries
SELECT * FROM experiments;
SELECT run_uuid, name, status FROM runs ORDER BY start_time DESC LIMIT 10;
SELECT * FROM registered_models;