Free Databricks ML Associate sample questions
Try these before reading the answers. The mix follows the weighting, so most come from the platform and model development sections.
Question 1. A data scientist has run twenty training experiments with different hyperparameters and needs to compare their metrics and recover the exact parameters that produced the best result. Which Databricks capability provides this?
- A. MLflow tracking
- B. Unity Catalog
- C. Delta Live Tables
- D. A cluster policy
Show answer
Answer: A
MLflow tracking records parameters, metrics and artefacts for each run and lets you compare them, which is exactly what comparing twenty experiments and recovering the winning configuration requires. Unity Catalog governs access and lineage, Delta Live Tables builds data pipelines, and a cluster policy controls compute configuration.
Question 2. A team wants a quick baseline model and a starting point they can inspect and modify, rather than an opaque result. What does Databricks AutoML provide?
- A. An opaque best model with no visibility into how it was produced
- B. Deep learning models only
- C. A baseline model plus generated notebooks the team can inspect and edit
- D. A deployed endpoint requiring no further evaluation
Show answer
Answer: C
AutoML produces a baseline model together with editable notebooks showing how it was built, so the team can inspect and modify the approach rather than accepting a black box. It does not hide its method, it is not limited to deep learning, and it does not remove the need for evaluation.
Question 3. A churn model is trained on data where 4% of customers churned. It reports 96% accuracy. What should the data scientist conclude?
- A. The model performs well and can be registered
- B. Accuracy is misleading at this class balance; evaluate precision and recall on the churn class
- C. More features should be added until accuracy exceeds 99%
- D. The cluster should be resized before re-evaluating
Show answer
Answer: B
With 4% positives, predicting no churn for everyone yields 96% accuracy while identifying nobody, so accuracy is misleading and precision and recall on the churn class are the informative metrics. The model is not necessarily good, more features may not help, and cluster size does not affect metric validity.
Question 4. An organisation must control which teams can access a registered model and trace what data it was built from. Which capability governs this?
- A. MLflow tracking
- B. Databricks AutoML
- C. A scheduled job
- D. Unity Catalog
Show answer
Answer: D
Unity Catalog provides centralised access control and lineage across data and models, which is what team-level permissions plus data provenance requires. MLflow tracking records runs, AutoML generates baselines, and a job schedule controls when work runs.
Question 5. A model must score a batch of two million records once per night. Latency per record does not matter and cost does. Which deployment approach fits?
- A. Batch inference run as a scheduled job
- B. A real-time serving endpoint held continuously available
- C. Structured streaming inference
- D. Running the model manually in an interactive notebook each night
Show answer
Answer: A
Batch inference on a schedule processes large volumes without maintaining a persistent endpoint, which is the cheapest fit when per-record latency is irrelevant. A real-time endpoint bills for capacity that is idle most of the day, streaming suits continuous arrival, and an interactive notebook is not a production deployment.
How did you do?
Notice that three of the five are about Databricks capabilities rather than machine learning theory. That reflects the weighting — Databricks Machine Learning is the largest section at 38%.
If you were unsure which component does what, see weeks 1 and 2 of the study plan. Knowing that MLflow tracks and Unity Catalog governs is worth a surprising number of marks.