DataEngBytes 2023 - MEL-T2-01 - Mikiko Bazeley
summary
Claims that 80% of a data scientist’s time is spent on data and that 1 in 5 companies build a feature store drive the talk, backed by concrete patterns: Mailchimp runs batch predictions for 80% of models, Shopify reports 48–72 hours to prototype-and-deploy, and feature stores aim to version transformations, features and metadata across a training store / inference store / transformation workflow. The pitch: feature stores reduce iteration time, enforce lineage and governance, but only pay off after certain platform and team maturity.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:01Feature stores aren’t always answer
- 2:47Analytics versus ML
- 4:54ML lifecycle
- 7:25MLOps pipeline challenges
- 9:53Feature engineering
- 11:19MLOps versus DevOps
- 14:27Core MLOps problems
- 17:21Cultural hurdles in MLOps
- 20:40Tooling paradox in MLOps
- 22:05Feature stores case study
- 25:20Feature stores in stack
- 27:53Tooling won’t solve MLOps
- 28:56Final thoughts on MLOps
related talks
30mHow Feature Stores Enable Operational ML
39mDataEngBytes 2023 - BRI-T1-02 - Mikiko Bazeley
08mFeature Store Usage Patterns - From a Single Data Scientist to an Enterprise
35mPanel 03- How to make the Feature Store become a hidden part of the productionalization of ML models
28mData VS Engineers: a long lasting story
25mBuilding A Feature Store For Hyper Growth
20mNavigating the Skies of Data: A Flight Through Feature Stores
27mDataEngBytes 2023 - SYD-T1-10 - Kanishka Mohaia