ML Ops on Google Cloud Platform
summary
Overview of deploying ML on Google Cloud with concrete tools and results: uses Vertex Pipelines (serverless, step caching) and the kfp v2 SDK to build a pipeline that trains an AutoML tabular classifier for spam detection, logs artifacts to a managed metadata server, and conditionally deploys models via a custom eval component. Demo runs successfully in the console, showing lineage, run comparison, and reuse of prebuilt Google Cloud pipeline components; also covers model monitoring for training‑serving skew and the Vertex feature store.
AI-written summary · about
chapters (14)
- 0:00Start
- 1:04Why Productionizing Ml Is Hard
- 3:52Ml Ops
- 4:27Mlops Patterns and Practices
- 5:54Typical Machine Learning Workflow
- 10:06Vertex Pipelines Sdks
- 11:17How Vertex Pipelines Relates to Acute Flow Pipelines
- 18:01Google Cloud Pipeline Components
- 19:11How Can Vertex Pipelines or Cubaflow Pipelines Play a Part in Operationalizing Helpful Ml Ops Design Patterns and Practices
- 24:07Vertex Model Monitoring
- 24:55Prediction Drift
- 25:17Vertex Feature Store
- 26:20Support for Point in Time Look Up
- 27:05Resources for Learning More about Vertex Pipelines
related talks
03mML Ops on Google Cloud
32mMachine Learning with Google Cloud Platform
23mBuilding Scalable, Observable MLOps Systems on Google Cloud
31mProductionizing ML with ML Ops and Cloud AI
28mMastering MLOps: Building Future Ready ML Systems
49mTechnology Mix: Containers + ML + Cloud — MLOps on GCP (...)
42mMachine Learning APIs, from Training [...]
59mMachine Intelligence at Google Scale