Continuous Delivery Principles for Machine Learning
summary
Describes applying Continuous Delivery practices to machine learning deployments at Indix using open-source tools, focusing on the ‘last mile’ of getting models into production. Covers concrete components: a Models Repository for metadata, lineage and reproducibility; ML pipelines to orchestrate and visualize multi-stage workflows; model QA and regression testing strategies; serving and scaling heterogeneous models (Scala, Python, etc.); and production maintenance and monitoring.
AI-written summary · about
chapters (29)
- 0:00Intro
- 1:02About Me
- 2:38Six Business Critical Indexes
- 3:48Data Pipeline @ Indix
- 5:19E-Tailers & Marketplaces
- 6:00Ad Display & Exchange Platforms
- 7:06Data Scale @ Indix
- 7:55ML @ Indix - Classification
- 8:49ML @ Indix - Attribute Extraction
- 9:26Machine Learning Workflow
- 10:15Machine Learning Sandwich?
- 11:25Different Skillsets
- 12:55Pain Points
- 15:13Reality
- 15:47Dejavu?
- 16:38What is Continuous Delivery?
- 17:10Principle #1
- 17:31Training Pipelines
- 19:46Principle #2
- 20:07Model Repository
- 22:12Model Promotion
- 23:32Principle #3
- 23:49Model Container
- 25:57Model Deployment
- 27:34Principle #4
- 28:08Model A/B Testing
- 30:12Spot Checking Example 1
- 30:45Future Work
- 31:37We're Hiring !!!
related talks
18mContinuous deployment of Machine Learning models
39mContinuous Delivery for Machine Learning — The Future of MLOps
17mContinuous deployment for ML: The new software development lifecycle
56mDevOps for Machine Learning
57mDevOps for Machine Learning
01mMLOps: Applying DevOps to ML
31mDevOps for Science: using continuous integration
59mDevOps for Machine Learning Projects