Question

You are experimenting interactively in an Azure Machine Learning notebook running on a compute instance. You train several scikit-learn models in different cells and want each training run automatically captured as a separate tracked run — including parameters, metrics, and the model artifact — in the workspace's MLflow tracking store, without adding explicit logging calls for every value. What is the most efficient way to achieve this?

Accepted Answer

Call mlflow.sklearn.autolog() before fitting the models, and wrap each model's training in its own mlflow.start_run() context.

Answer

Manually call mlflow.log_param() and mlflow.log_metric() after each fit, then mlflow.sklearn.log_model() to store the artifact.

Answer

Submit each cell as a separate command job to a compute cluster so the platform captures the runs automatically.

Answer

Enable diagnostic settings on the workspace to stream training telemetry to a Log Analytics workspace.