Reconceptualizing Machine Learning for the Real-time World
summary
Claim: 87% of data science projects never reach production, so the talk demonstrates moving from batch to continuous learning with concrete tools. Uses River models and a PyEnsign streaming pipeline: a Yelp publisher, a subscriber that does predict then fit per event, a metrics topic and an alerting subscriber. Demo shows per-record precision/recall metrics rising as the model learns and alerts firing when thresholds drop, proving online learning reduces deployment lag and adapts in real time.
AI-written summary · about
chapters (5)
- 0:00Event-driven data science
- 4:43ML projects never reach production
- 12:38Batch to async ML
- 20:22Real-time model validation
- 28:12Q&A
related talks
30mMoving from Offline to Online Machine Learning with River
31mMoving from Offline to Online Machine Learning with River
23mMoving from Offline to Online Machine Learning with River
40mReal-time Learning with Spark
37mBig Data Analytics, Machine Learning, Real Time Processing
39mChanging Realities - The Final Frontier for Machine Learning?
22mReal-world applications of machine learning: Beyond the hype
04mInterview