Question

You are training a model in Azure Machine Learning using a training script that performs k-fold cross-validation. For each of the 5 folds you train a separate model and want to track each fold's metrics separately, while still grouping all folds under a single logical experiment run so you can compare aggregate performance. You are using MLflow for experiment tracking. Which approach lets you organize the fold-level tracking correctly?

Accepted Answer

Start a parent MLflow run, then start a nested run (nested=True) for each fold to log its metrics independently under the parent.

Answer

Start one MLflow run and log each fold's metrics with a step index, so all folds share the same run.

Answer

Submit 5 separate command jobs, one per fold, and rely on the workspace to group them automatically by experiment name.

Answer

Log all fold metrics as tags on a single run, since MLflow metrics cannot be separated by fold otherwise.