Databricks Certified Machine Learning Associate

About the Databricks Machine Learning Associate exam

The Databricks Certified Machine Learning Associate exam checks that you can do everyday machine learning work on Databricks: explore and prepare data, train and tune a model, track it with MLflow, register it in Unity Catalog and get it scoring data in batch or behind an endpoint. It suits data scientists and ML engineers who are new to the platform but not to machine learning. Databricks suggests at least six months of hands-on experience with the tasks in the exam guide.

The exam mixes two kinds of question. Some are about general ML practice: picking a metric for an imbalanced dataset, avoiding data leakage in cross-validation, or deciding how to impute a skewed feature. Others are about the platform itself: which MLflow call finds the best run, how a feature lookup works at scoring time, or how to promote a model with an alias. ML code in the questions is Python, and data manipulation code may be SQL.

Exam format

• Questions: 48 scored multiple-choice questions. The exam may also include unscored items, with extra time allowed for them.
• Time: 90 minutes.
• Cost: 200 USD.
• Delivery: Proctored, online or at a test center. No test aids are allowed.
• Languages: English, Japanese, Portuguese (Brazil) and Korean.
• Validity: 2 years. To recertify you take the current version of the exam again.

Databricks does not publish a passing score on the exam page.

What the exam covers

The exam outline has four sections:

• Databricks Machine Learning (38%). The largest section, about the platform's ML tools: choosing a Databricks Runtime for Machine Learning cluster, MLflow tracking and autologging, experiments and nested runs, AutoML and the notebooks it generates, feature tables in Unity Catalog, and the model registry in Unity Catalog with aliases and model signatures.
• ML Workflows (19%). Data exploration and preparation: summarising data, handling missing values, encoding categorical columns in the right order, scaling features, reproducible train-test splits, and spotting features that leak the outcome.
• Model Development (31%). Training and tuning: cross-validation versus a single validation split, grid versus random search, reading the gap between training and validation scores, choosing an interpretable model, feature importance, and choosing and reading evaluation metrics.
• Model Deployment (12%). Batch scoring with a Spark UDF from MLflow, scoring with feature lookups, and Model Serving endpoints, including splitting traffic between model versions.

How the practice questions here are organised

There are three 10-question practice tests. Each one mixes the four sections, with more questions on MLflow, feature engineering in Unity Catalog and model development, because those carry most of the exam's weight. Topics are labelled more finely than the four sections, for example MLflow Tracking, MLflow Model Registry, AutoML, Model Evaluation, so you can see exactly which areas you miss. Each question explains why the right answer is right and why each other option is wrong, with a link to the Databricks or MLflow documentation it is based on.

A study plan that works

• Take Test 1 without preparing, to find out whether your gaps are in ML theory or in the Databricks platform. Most people find it is one more than the other.
• If the platform is the gap, do the work in a workspace: run AutoML on a small dataset, open the generated notebooks, log runs by hand and with autologging, then register the best model in Unity Catalog and give it an alias.
• If ML theory is the gap, go back to the basics of leakage, validation strategy and metrics. The exam asks about them in plain terms, not deep mathematics.
• Read the explanation of every question you missed or guessed, then move on to Tests 2 and 3. Retake any test you scored under 70% a few days later.
• Learn the current names. Workspace model registry and feature store questions now point to their Unity Catalog versions, so older study material may teach the wrong answer.

These questions are original, written from the public Databricks and MLflow documentation and the official exam guide. They are not real exam questions, and this site is not affiliated with Databricks.