Databricks ML Associate Practice Question 11: Runtime ML Cluster Choice
Databricks Certified Machine Learning Associate · Test 2 · question 1 of 10
A new team will train scikit-learn and XGBoost models in Databricks notebooks and track every run in MLflow. Clusters are terminated each evening and recreated each morning. The team does not want to reinstall the machine learning stack on every cluster start, and they do not want to add a %pip install cell to the top of each notebook.
Which compute choice meets these requirements with the least setup?
Show answer and explanation
Answer:
Databricks Runtime for Machine Learning is the standard Databricks Runtime plus a curated, version-compatible machine learning environment: scikit-learn, XGBoost, PyTorch, TensorFlow and MLflow are already present when the cluster starts. Selecting the ML runtime in the cluster's Databricks Runtime version dropdown is the entire setup, and it survives the nightly terminate-and-recreate cycle because the libraries are part of the image rather than something layered on afterwards.
Option A works but is the more laborious path the question rules out. An init script runs on every start, adding minutes to cluster launch, and the team now owns pinning and upgrading a set of library versions that the ML runtime already tests together.
Option C is the per-notebook installation the requirements explicitly exclude. Notebook-scoped libraries are also reinstalled on every attach, so the cost is paid repeatedly.
Option D misreads what a SQL warehouse is. SQL warehouses run SQL queries; they do not attach to notebooks for Python model training and provide no machine learning libraries.
Reference: Databricks Runtime for Machine Learning