Practical deep dive into machine learning techniques and MLOps
summary
Demo shows Azure Machine Learning wiring a secure workspace with managed network isolation, importing Snowflake data via Azure ML Data Import, editing in VS Code Web, running serverless training jobs, and using experiment tracking to pick the best run ("joyful bean"). Registry shares the model, deployment uses a wizard with mirror traffic (20% mirrored, then ramp to 100%), and monitoring (data drift, prediction drift, data quality) flags LOCALHOUR nulls up to 3% and transaction_amount drift 0.32>0.3. Also demos managed feature store with portable YAML feature retrieval and open-source model catalog; platform claims 300 billion inferencing requests per month.
AI-written summary · about
chapters (12)
- 0:00Welcome
- 1:11Azure Machine Learning
- 3:16Secure ML workspace
- 3:57Credit card fraud model
- 8:10Model evaluation and registration
- 11:42Deploying to production
- 15:55Monitoring production models
- 23:42Managed feature store
- 29:23Feature store success
- 30:47Open-source models in Azure ML
- 36:59Production scale stories
- 41:38Q&A
related talks
48mStreamlining your Machine Learning projects with Azure Machine Learning
45mMLOps and ML-Flops: going from dev to prod
43mA lap around Azure Machine Learning services
27mBeginners guide to Azure MLOps
51mHow to get started with MLOps in Azure Databricks
21mMLOps for R with Azure Machine Learning
58mDeep Dive into Azure Automated Machine Learning
41mTHE STATE OF MLOPS - MACHINE LEARNING IN PRODUCTION AT ENTERPRISE SCALE