Experiment Tracking with MLflow on SageMaker AI
Explore how to implement experiment tracking with MLflow on SageMaker AI to log parameters, metrics, and models for reproducibility. Learn to set up managed MLflow Apps, configure training jobs, and maintain experiment metadata in AWS. Understand the differences between MLflow Apps and tracking servers and how to support collaborative workflows in SageMaker Studio.
We'll cover the following...
The previous lesson introduced SageMaker Experiments, with its experiments, trials, and trial components, as a way to organize training runs. AWS documentation now states that experiment tracking through the SageMaker Experiments Python SDK is available only in Studio Classic, and it recommends that new work in the current SageMaker Studio use the managed MLflow integration. Past Experiments runs stay viewable in Studio Classic. To continue that work in MLflow, update the training code to use the MLflow SDK and run it again; nothing is converted automatically.
This lesson covers what MLflow tracks, how managed MLflow on SageMaker AI is set up, and how to log a run so that a teammate can reproduce it. It also shows what the exam expects you to choose when a scenario asks for reproducible tracking. The next lesson covers SageMaker Model Registry for version control and deployment approval, and the lesson after it shows how models logged in MLflow sync into that registry. ...