Question 13

Explanation

Correct Answer:
They could specify nested=True when initiating the child run for each unique combination of hyperparameter values.

Explanation:

To organize MLflow runs into a parent-child hierarchy (e.g., one parent run for hyperparameter tuning and child runs for each hyperparameter combination), use the nested=True flag when starting child runs. This ensures child runs are nested under the parent run, maintaining a clear structure in the MLflow UI.

Steps to Implement:

1. Start the parent run:

   
   

   with mlflow.start_run() as parent_run:
       # Parent run logic (e.g., tuning setup)

2. Start child runs inside the parent context:

   
   

   for params in hyperparameter_combinations:
       with mlflow.start_run(nested=True) as child_run:  # Critical: nested=True
           # Log hyperparameters and metrics for each child run
           mlflow.log_params(params)

Why This Works:

• nested=True: Explicitly declares the child run as nested under the active parent run.

• UI Organization: Child runs appear grouped under the parent in the MLflow Experiments page.

Why Other Options Are Incorrect:

1. Same experiment ID for all runs:

   • Does not create a hierarchy; runs are flat siblings.

2. Indented code block without nested=True:

   • Runs are siblings unless nested=True is specified.

3. Databricks Autologging:

   • Automates logging but does not control run hierarchy.

4. nested=True on the parent run:

   • nested=True is only valid for child runs (parent runs cannot be nested).

Key Takeaway:

For parent-child run organization:
✅ Use with mlflow.start_run(nested=True) for child runs.
✅ Parent runs must be active when child runs are created.

Example MLflow UI View:

Parent Run (Tuning)
├── Child Run (Params 1)
├── Child Run (Params 2)
└── Child Run (Params 3)