A public-health notebook trains a simple clinic no-show model alongside ETL feature prep. Which runtime should the team select?

A public-health notebook that trains a simple clinic no-show model alongside ETL feature prep needs a machine learning–capable Databricks runtime that includes common ML libraries for that mixed notebook. ML runtimes ship with widely used Python ML packages and a Spark environment suited to feature engineering plus lightweight training in one civic notebook workflow, avoiding brittle manual package hunts on SQL-only or stripped images.

A SQL warehouse runtime that rejects Python machine learning libraries cannot train models or host typical ML notebook stacks; warehouses are for SQL/BI. A job compute image stripped of ML packages to force external training only conflicts with the stem’s in-notebook training requirement and adds unnecessary pipeline fragmentation for a simple no-show model. Photon-only warehouse compute with no notebook ML capabilities again selects a SQL acceleration path instead of an ML-capable notebook runtime.

Exam caveat: pick ML runtimes when notebooks mix Spark feature prep and sklearn/MLlib-style training; pick standard data-engineering runtimes when there is no ML. GPU ML runtimes are for deep learning / heavy GPU needs—not automatically required for a “simple” clinic no-show model unless the stem says so. SQL warehouses remain correct for pure dashboard SQL even if ML exists elsewhere in the workspace.

Operational check: select an ML Databricks runtime on the public-health all-purpose or job cluster, confirm required libraries import cleanly, keep production training reproducible with a pinned runtime version, and separate large-scale training to appropriately sized ML compute if the simple notebook model grows.