Productionizing Data Science Workloads
summary
Maps the lifecycle from ideation through automated training to production and shows concrete tooling and patterns used: d-lab for self-service Jupyter sandboxes and isolated data buckets, odaho for ML-focused CI/CD that spawns Spark/TensorFlow clusters, integrates with AWS/GCP, versions models and training reports, and tracks data shape drift. Explains using pickle binaries for deployments, automated provisioning to scale compute up/down, continuous monitoring of prediction accuracy, and A/B traffic splits for rollout.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:24Development to Production
- 5:15Frameworks
- 7:42Dlab
- 9:31Odoo
- 11:20In Continuous Integration
- 14:07Conclusion
related talks
34mData Products, From 0 to data science pipeline
37mA Day in the Life of a Data Scientist. Conquer ML Lifecycle on Kubernetes
39mPyData in Production: Lesson Learned from Various PyData Deployment Strategies
31mProductizing Machine Learning over Big Data with AWS tools
18mFrom DevOps to DataOps - Cloud based Software Development
33mDealing with the versioning of production-ready models
35mGetting Machine Learning Models Ready For Production Using Python
28mDevOps for Data Science With Kubernetes