Week 04 — Experiment Tracking — MLflow, Weights & Biases

Six months from now, which hyperparameter combination produced your best result?

Lecture

The experiment-tracking problem · MLflow (open source) · Weights & Biases (managed) · Neptune, Comet, ClearML · the run / artifact / metric model · structured logging · dashboard comparison · promotion to a model registry.

Read before the lecture

• MLflow documentation: Tracking

Code lab

Lab 3 — Instrumented training loop

Take any training script from a prior course. Add MLflow tracking (parameters, metrics, artifacts). Run a hyperparameter sweep. Promote the best model to the registry.

Notebook: lab03-tracking.ipynb  ·  Dataset: Same as Lab 2.

Reference text for this week: chapter 04 of the bilingual notes — EN PDF · FR PDF.