A clerk's office wants scikit-learn training scripts to record metrics automatically in MLflow so interns are not responsible for manual screenshots. What should they enable?

Enabling MLflow autologging from the scikit-learn training scripts so metrics are captured automatically removes intern screenshot duty and writes structured metrics into Azure Machine Learning experiment tracking. Autologging hooks common training frameworks so parameters and scores land with the run without manual logging calls on every line. Clerk-office reproducibility standards depend on that automatic capture.

Enabling MLflow autologging for the PyTorch integration in the training scripts to capture run history fails because the stem concerns scikit-learn training, and PyTorch autologging does not automatically capture scikit-learn metrics or remove the interns' screenshot duty. Configuring Azure Monitor logging on the training compute to record CPU and memory usage during training fails the stem's goal because it tracks infrastructure health rather than MLflow experiment metrics and scores from scikit-learn training. Registering the scikit-learn model in the MLflow Model Registry to version model files for review fails because model registration versions artifacts after training; it does not automatically record metrics during training to relieve interns of manual screenshots.

Exam caveat: When stems ask for automatic metric capture from scikit-learn into Azure ML, MLflow autologging is the intended feature. Operational check: enable autolog in the training script, submit a job, confirm metrics appear on the run without screenshots, and verify interns are not in the critical path for metric capture.