Machine Learning Engineering principles with Python and MLFlow
summary
Recreates a Michelangelo/Bighead-style pipeline using Python and MLflow to handle training, deployment, serving and champion/challenger tests, packaging models with dependencies into REST APIs and Docker; logs experiments, artifacts and Git hashes for reproducibility; integrates Delta Lake, Feast, Great Expectations and Amazon DQ for data/version checks; shows an NLP PTSD project with 35 engineers and a toy model (1,000 examples ≈ 20 minutes) and reports the main failure mode as lack of monitoring. MLflow model logging feature store (Feast/Palantir)
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:25Outline
- 2:32Context
- 3:12Problems
- 3:53Data entanglement
- 4:28Tracking experiments
- 5:00Metrics
- 5:13Machine Learning Engineering
- 7:43Machine Learning Engineers
- 8:28Why Machine Learning Engineering
- 9:46Guiding Principles
- 12:26Uber ML Platform
- 23:57Feast
- 26:24Pipeline
- 31:33Predictions monitoring
- 35:26Questions
related talks
57mManaging Machine Learning Lifecycle with MLflow
31mMLflow: Platform for Complete Machine Learning Lifecycle
33mOperationalizing Model Lifecycle with MLFlow
30mMLflow: an Open Source Tool for Managing the ML Project Lifecycle
45mMachine Learning Lifecycle Made Easy with MLflow
37mPlatform for Complete Machine Learning Lifecycle
18mAutomate the boring ML stuff with pipelines
01h 03mOne Flow to Rule Them All: Building the end to end ML project to production