Just In Time Features Of Machine Learning Models in Clojure
summary
Centralized Clojure microservice called Bulgogi implements just-in-time feature computation as plain functions (each takes an input map), runs a six-lines preprocess that looks up feature symbols and p-map parallelization to compute features, returns a feature map and asynchronously persists to a feature store; implemented as a single Babashka script so near-zero boilerplate and language-agnostic reuse, but requires backfilling features before retraining models.
AI-written summary · about
chapters (9)
- 0:00Start
- 2:08Feature Engineering
- 4:11Just in Time Features
- 5:03Real World Context
- 9:17Spark
- 11:35Goals
- 12:34Architecture
- 18:27Preprocess Function
- 23:47Disadvantages
related talks
41mDeploying ML models in a Clojure environment
17mCreating & operating ML models from event-based data using feature stor...
45mBuilding a Feature Platform to Scale Machine Learning
28mMachine Learning Live
40mBuilding Machine Learning Models with Clojure and Cortex
30mHow Feature Stores Enable Operational ML
27mFeast: feature store for Machine Learning
40mCreating and operating ML models from event based data using feature stores