MLflow: Platform for Complete Machine Learning Lifecycle
summary
MLflow demonstrates a concrete lifecycle: mlflow tracking captures parameters, metrics and artifacts; mlflow projects package runs; mlflow models standardize serialization; model registry versions and stages models. Demo runs on Databricks with Delta Lake, logs a scikit-learn random forest, uses Hyperopt to run parallel hyperparameter searches across the Spark cluster, reaches ~0.81 F1, registers versioned models, swaps in a Keras model, and deploys via a Docker image or Spark UDF from S3 with no serving-code changes.
AI-written summary · about
chapters (18)
- 0:00MLflow introduction
- 1:03ML development challenges
- 5:16MLflow overview
- 6:45MLflow Tracking deep dive
- 8:10MLflow Projects
- 9:34MLflow Models
- 10:40MLflow Model Registry
- 11:58Demo setup
- 13:25Product classification model
- 15:53Logging experiments
- 18:36Hyperparameter tuning
- 21:15Comparing and analyzing MLflow runs
- 23:09Model registration & deployment
- 24:37Real-time predictions
- 26:41Docker model deployment
- 27:03Keras model switch
- 28:52Registry model updates
- 29:34Summary
related talks
30mMLflow: an Open Source Tool for Managing the ML Project Lifecycle
37mPlatform for Complete Machine Learning Lifecycle
24mMLflow: An open platform to simplify...
33mOperationalizing Model Lifecycle with MLFlow
57mManaging Machine Learning Lifecycle with MLflow
45mMachine Learning Lifecycle Made Easy with MLflow
17mIntroducing mlflow
55mOrganizing the Machine Learning Lifecycle with MLflow