Machine Learning Ops: DevOps for Data Science
summary
End-to-end pipeline using Amazon SageMaker with CodePipeline/CodeBuild, CloudFormation, CodeDeploy and Lambda automates training, deployment and monitoring for bank-transaction classification; demo builds a Ground Truth + A2I labeling loop, captures inputs/outputs to S3, generates a baseline and a scheduled Model Monitor that flags drift and triggers retrain. Deployment uses Blue-green CodeDeploy rollback and shows 86% accuracy on a 100-sample test while CloudWatch metrics and monitoring schedules detect class-distribution drift.
AI-written summary · about
chapters (19)
- 0:00Intro
- 1:07Dev Ops for traditional software development
- 4:05Challenge: Different teams might own part of process
- 5:47Challenge: Detecting model drift
- 9:19Amazon SageMaker Debugger
- 10:56Amazon SageMaker Operators for Apache Airflow
- 11:13Amazon SageMaker Operators for Kubernetes
- 12:22Continuous deployment goals
- 13:41AWS Step Functions Data Science SDK
- 14:28Third party open source integrations
- 15:25Use case: Transaction categorisation
- 15:55Transaction categorisation architecture
- 18:16Serverless Continuous Integration and Deployment
- 19:16AWS CodeBuild build specification
- 19:55AWS CloudFormation Prod Deployment
- 20:56AWS CodeDeploy Blue/Green Lambda deployment
- 21:29AWS CloudFormation custom resource Lambda
- 38:47Example AWS Step Functions Workflow
- 41:10Sidebar. Data integration options
related talks
33mBuild, train, and deploy ML models using Amazon SageMaker
37mStreamlining Machine Learning Operations
29mImplementing MLOps practices with Amazon SageMaker
34mImplement MLOps practices with Amazon SageMaker
34mImplement MLOps practices with Amazon SageMaker
18mFrom DevOps to DataOps - Cloud based Software Development
19mDrive transformation through machine learning
32mBuild, Train, and Deploy Machine Learning Models on AWS