Automated Machine Learning: Building a Pipeline
summary
Describes an automated ML production pipeline that collects online features to serve low-latency predictions, stores feature snapshots in an online feature store, and routinely retrains to handle churn and spam. Uses Random Forest and gradient boosting with grid search over tree ensembles, runs on a small cluster (2–3 servers, ~24 cores), and exports lightweight JSON models for deployment; an email model took ~10–15 person-hours and ~5 machine-hours and reduced sent emails by 60%, saving tens of thousands of dollars.
AI-written summary · about
chapters (7)
- 0:00AutoML introduction
- 6:24Click prediction data
- 15:41Building AutoML pipeline
- 24:33Model training and evaluation
- 30:12Business optimization
- 32:39Deploying to production
- 34:28Model monitoring
related talks
32mAutomating Production Production Level Machine Learning on AWS
18mDesigning a scalable ML Ops Pipeline: Insights and best practices
25mContinuous ML Applications in Production
22mAccelerating ML impact
23mAutomation: Build A Training Pipeline
41mInfrastructure for ML Applications
38mReal Time ML Pipelines in Multi-Tenant Environments
39mMachine Learning in Financial Credit Risk Assessment