Implementing MLOps practices with Amazon SageMaker
summary
Automates a repeatable SageMaker pipeline that preprocesses US census data, trains a Scikit-learn logistic model on ML M5.xlarge, and evaluates holdout 20% data with F1/accuracy/precision metrics; artifacts and code live in S3 and CodeCommit, deployment uses CodePipeline/CloudFormation to update SageMaker endpoints after a manual approval, and model metadata goes into a registry (Aurora serverless). Demonstrates AWS Step Functions Data Science SDK driving SageMaker processing (Scikit) and a train-once-deploy-many flow.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 3:47Machine Learning Workflows
- 6:07Demo
- 6:56AWS Step Functions
- 8:19Step Function Workflow
- 9:46SageMaker Processing Job
- 11:02SageMaker Workflow
- 12:06SageMaker Training Job
- 12:57SageMaker Evaluation Job
- 14:19Creating a Workflow
- 15:58Automation
- 23:28CI Pipeline
- 28:06Recap
related talks
34mImplement MLOps practices with Amazon SageMaker
34mImplement MLOps practices with Amazon SageMaker
36mImplementing MLOps practices on AWS
43mMachine Learning Ops: DevOps for Data Science
37mStreamlining Machine Learning Operations
45mAutomate your ML workflows with pipelines
33mBuild, train, and deploy ML models using Amazon SageMaker
33mBuilding & Managing MLOps Pipelines in the Cloud