Hands-On Feature Store and Model Deployment

on YouTube ↗02h 33m

" summary

Splice Machine runs a single hybrid DB (OLTP + OLAP) on Kubernetes with Spark and an mlflow fork to build a feature store and deploy models as in‑database UDFs. It creates real‑time and history tables linked by triggers so updates move old values automatically, produces point‑in‑time training views via generated ANSI SQL, and logs run IDs plus DB transaction IDs for replay. Deployments take ~15s, supports scikit‑learn/keras/h2o/sparkml, claims sub‑20ms SLAs for serving, and uses background Spark compaction and external S3 tables for scale.

AI-written summary · about

" chapters (41)

1. 0:00Start
2. 0:24What Is Splice Machine
3. 1:13Hybrid Database
4. 7:32Single Database Design
5. 8:47Sign Up for a Splice Machine Free Trial
6. 9:33User Interface
7. 10:05Free Trial
8. 15:45Create a Feature Store
9. 16:13Start a Spark Session
10. 17:00Create a Feature Store from Scratch
11. 33:27Feature Discovery
12. 35:18Using the Feature Store for Future Discovery
13. 37:06Display Feature Search
14. 38:06Feature Distribution
15. 41:23Can You Add Your Own Fields to the Feature Store Attributes
16. 41:47Query Based on Tags
17. 43:36Options To Add Data Types
18. 45:27Model Training
19. 52:04Create Training View
20. 1:01:38Automation of Model Training
21. 1:02:31Train a Model
22. 1:05:44Run Id
23. 1:05:54Database Transaction Id
24. 1:09:33Model Deployment
25. 1:09:57Complexities and Limitations
26. 1:10:17Permissioning
27. 1:10:37Model Governance
28. 1:12:47Database Deployment
29. 1:14:00Key Benefits
30. 1:15:34Steps for Creating a for Deploying a Table into the Database
31. 1:15:52Mlflow Deploydb
32. 1:21:29Create a Machine Learning Prediction
33. 1:26:28Deploying a Model
34. 1:27:52Benchmarks for Real-Time Latency
35. 1:29:35Display Model Drift
36. 1:36:15Model Interpretability
37. 1:41:55Questions
38. 1:43:41Optimizer
39. 1:48:06External Table
40. 1:53:26What Type of Support Do You Have for Media Data Images Sound and Video
41. 1:54:23Are We Hiring

" related talks

• 32mA feature Store for Machine LearningJim Dowling · PyData Fest Amsterdam 2020
• 22mSPOK - Managing ML/Big Data Spark Workloads at scale on KubernetesNagaraj Janardhana, Mikhail Arov · OpML '20
• 27mDeveloping and Deploying Models at Scale — without Spark!Eduardo Ariño de la Rubia · AI By the Bay 2017
• 38mScaling Data ScienceDenys Kovalenko · Data Science fwdays'19
• 10mFast Sub-ML Usecase Development using Feature StoresAchint Thomas · Feature Store Summit 2022
• 15mFeature Stores: Powering Machine Learning Infrastructure with Apache FlinkSimba Khadder · Flink Forward Global Virtual 2020
• 29mIntegrating High Performance Feature Stores with KServe Model ServingTed Chang, Chin Huang · Open AI + Data Forum @ Open Source Summit North America 2022
• 15mKick-offJim Dowling · Feature Store Summit 2022