Self-serve Feature Engineering Platform Using Flyte and Feast
summary
Shows an end-to-end integration that uses Flyte workflows to run feature transforms and Feast as the feature registry and serving layer, demonstrating training a classifier on the UCI horse colic dataset (Gaussian Naive Bayes) and returning online predictions; the pipeline supports declarative feature definitions and automatic offline and online materialization, uses Flyte’s memoization for fast re-runs and full reproducibility, and exposes a practical limitation: Feast’s single-file registry blocks concurrent registry writes today, though reads scale and the demo executes successfully.
AI-written summary · about
chapters (6)
- 0:00Start
- 0:36Feature Engineering
- 1:13Building feature pipelines is hard
- 4:42Duplication of effort
- 10:54Solution #2: Consistent data access
- 15:04Build training dataset
related talks
27mFeast: feature store for Machine Learning
32mBuilding a Cloud Native Feature Store with Feast on Kubeflow
31mFeast as a Cloud Native Data System: Bridging Batch Processing and Online Serving...
46mScaling ML Pipelines with Feast, Ray and Kubeflow
01h 24mRobust and End-to-End Cloud Native Machine Learning & Data Processing Platform
39mFlyte: Production-Grade Data and Machine Learning Orchestration
38mFlyte: Cloud Native Machine Learning & Data Processing Platform
34mSolving the ML Pain You Forgot to Mention