Depend on Docker for Self-Managed Predictive Modeling
summary
An open-source AWS DO PM template is built and run with Docker Desktop and PyTorch, using Docker Compose locally and ECR + EKS for cloud deployment; code is debugged via VS Code remote attach and the pm CLI is used to configure and run tasks. The demo pushes images, scales an EKS cluster to 100 nodes, deploys 1,000 electric vehicle models as gRPC services, updates models when error exceeds thresholds, and completes the full fleet run in 29 minutes using a graph database for provenance.
AI-written summary · about
chapters (17)
- 0:00Start
- 4:13Clone the repository
- 4:48Build container images /build.sh all
- 5:26Deploy aws-do-pm locally ./run.sh
- 7:49Run ev-demo ev-demo
- 8:57Explore results - data,models, and graph When using Cloud9 or EC2, expose Uls via SSH tunnels
- 9:20SSH TUNNELS TO PLATFORM UI AND GRAPHDB
- 10:29Push images to ECR ./login.sh && ./ecr-setup.sh && ./push.sh all
- 10:55Scale up EKS cluster Update auto-scaling group
- 11:50Configure project for Kubernetes pm config set
- 12:05Deploy aws-do-pm ./run.sh
- 12:30SETUP MONITORING USING WATCH "KUBECTL..."
- 12:42Configure aws-do-pm platform pods pm config set
- 12:59Configure ev-fleet-demo vi src/python/example/ev/ev_fleet_demo.json
- 13:22Run EV fleet demo ev-fleet-demo
- 15:57Scale down EKS cluster Edit auto-scaling group
- 16:24Explore results Browse data, models, and graph
related talks
58mThe path between developing and serving machine learning models.
34mData Science at PMI - The Tools of The Trade
32mDeploying Python models to production
35mAccelerate deep learning in the cloud with custom ML environments
35mGetting Machine Learning Models Ready For Production Using Python
27mDeveloping and Deploying Models at Scale — without Spark!
16mReducing Operational Barriers to Model Training
01h 02mDocker for Data Science: Reproducibility and Deployment