Learnings integrating a machine learning model to existing software
summary
Built an anomaly-alerting module for Trinity smart alerts that learns daily and weekday patterns from production metrics and replaces brittle thresholds. A simple baseline forecasting model proves as good as complex alternatives so baseline model wins, CI/CD pipelines and automated tests ship containerized services, favoring multi-stage builds and slim base images to keep images small and secure, and the team settles on Docker for building, Podman for runtime for better security.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:25Background
- 7:13Finding the right use case
- 10:38A picture is worth a thousand words
- 19:06Using multi-stage builds
- 20:08Picking the container engine
- 22:46Summary
related talks
31mMonitoring machine learning models in production
21mMastering the DevOps Machine Learning Pipeline
39mMastering real-time anomaly detection with open source tools
42mTesting Machine Learning Modules in Production
30mProductionizing an unsupervised machine learning model to...
25mSymptom-based Alerting for Machine Learning - What I Learned...
30mCultivating a company-wide machine learning culture
18mBeyond Inference: Bringing ML into Production