How to get started with MLOps in Azure Databricks
summary
Introduces a concrete Azure Databricks workflow: ingest Airbnb sample (90 features), use Delta Lake time travel for data versioning and audit, register reusable columns in Databricks Feature Store, track experiments with MLflow experiment tracking and register versions in the Model Registry, and create live endpoints (REST API) via Databricks model serving that scale to zero. Demonstrates reverting a table version, transitioning model stages, three deployment patterns (code-first / model-first / hybrid), CI/CD across separate Databricks workspaces, and monitoring for data and model drift.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 5:50What is MLOps and why it matters
- 8:23ML lifecycle overview
- 12:01Data versioning with Delta Lake
- 14:51Feature store for ML features
- 19:26Model tracking with MLflow
- 25:02Model registry for deployment
- 27:32Deployment strategies in Azure Databricks
- 35:26CI/CD pipelines for models
- 35:47Model serving with REST APIs
- 39:02Monitoring data and model drift
- 42:17Key takeaways and Q&A session
related talks
45mMLOps and ML-Flops: going from dev to prod
27mBeginners guide to Azure MLOps
48mMachine Learning in Azure Databricks
44mPractical deep dive into machine learning techniques and MLOps
29mML in Production on Databricks
48mStreamlining your Machine Learning projects with Azure Machine Learning
33mYour best Bet: Effortless MLOps with Python Models in
52mDevOps for AI in the Microsoft World