Online Machine Learning with Creme
summary
Claims that online learning removes retrain schedules and lets models learn per datum using an actual library called creme 0.01, showing an API with fit_one / predict_one and metrics that update online. The codebase implements about 20 models including online decision trees and bagging via a Poisson trick, exact streaming stats (Welford), and a working Django demo that predicts League of Legends game duration built over a weekend and deployed with pickle model storage and live polling.
AI-written summary · about
chapters (15)
- 0:00Online machine learning intro
- 3:35Batch learning problems
- 4:43What is online learning
- 6:30Creme library intro
- 9:22Creme models core API
- 11:53Online statistics & scalers
- 12:58Linear regression in Creme
- 13:40Time series rolling stats
- 14:25Online feature extraction
- 14:45Bagging & boosting online
- 15:33LoL game duration case study
- 17:36Live prediction with Creme
- 19:04Online learning in production
- 20:08Creme future & mindset
- 21:12Conclusion & Q&A
related talks
31mMoving from Offline to Online Machine Learning with River
30mMoving from Offline to Online Machine Learning with River
23mMoving from Offline to Online Machine Learning with River
30mReconceptualizing Machine Learning for the Real-time World
22mReal-world applications of machine learning: Beyond the hype
37mAutomated Machine Learning: Building a Pipeline
54mMachine Learning for Developers
28mMachine Learning Live