Home / Practice tests / ML Associate

Free · No credit card required

Databricks Machine Learning Associate Practice Test: Free Exam Questions & Answers

Realistic practice questions with worked explanations — Databricks ML, AutoML, MLflow tracking, Feature Store, model development and deployment.

45 questions on the real exam
90 min time limit
$200 per exam attempt

Question bank updated November 10, 2026

Exam blueprint

What's on the exam

The exam sections with their official weights — focus your study time where the points are.

Databricks Machine Learning

38%

ML runtime, AutoML, Feature Store, MLflow tracking and the model registry, experiment management.

Model Development

31%

scikit-learn and Spark ML pipelines, hyperparameter tuning with Hyperopt, evaluation metrics.

Data Processing

19%

Feature engineering with Spark and the pandas API, missing data, splits at scale.

Model Deployment

12%

Model Serving endpoints, batch vs. real-time inference, model versioning and stages.

Straight from our question bank

Try 4 real practice questions

Every question comes with a worked explanation — expand the answer when you're ready.

1 Databricks Machine Learning

A machine learning engineer already has a Unity Catalog Delta table with the correct columns, but it was created without a primary key. She wants to use it as a feature table rather than recreating it. What should she do?

  1. AClone the table into the workspace Feature Store, where tables are created in the form that the Feature Engineering UI and API recognize as feature tables
  2. BPublish the existing table to an online store with publish_table, which backfills the missing primary key metadata on the offline source table
  3. CConvert the Delta table into a notebook-scoped temporary view and call create_training_set, which recognizes feature tables dynamically at training time
  4. DRegister the existing table as a model artifact in MLflow, after which Catalog Explorer lists it as a feature table linked to an active experiment run
  5. EAlter the existing table to set the key columns to NOT NULL and add a primary key constraint, after which Unity Catalog treats it as a feature table
Show answer & explanation

Correct answer: E

WHY E: Databricks docs describe converting an existing Unity Catalog Delta table into a feature table by setting the primary key columns to NOT NULL and adding a primary key constraint. WHY NOT A: The workspace Feature Store is not required; any Unity Catalog Delta table with a primary key can serve as a feature table. WHY NOT C: Temporary tables or views are not the path to governed feature tables; recognition comes from the primary key in catalog metadata, not from training time. WHY NOT D: MLflow registration is unrelated to feature table discovery in Catalog Explorer. WHY NOT B: Online publishing does not create missing primary key metadata on the source table; the key has to exist first.

2 Model Development

A scikit-learn GridSearchCV job is running on one large machine with many CPU cores. The model itself is single node, but the engineer wants different parameter-fold evaluations to run concurrently on that machine. Which configuration is the standard approach?

  1. ASet random_state on the estimator to make the folds execute at the same time with identical pseudorandom seeds
  2. BSet n_jobs on GridSearchCV to run the candidate evaluations in parallel across the available local workers
  3. CSet cv to a single train_test_split on GridSearchCV to make the estimator parallelize over the remaining rows
  4. DSet verbose on GridSearchCV to a high value to make the search dispatch each fold to a separate CPU core
  5. ESet warm_start on the estimator to let later parameter-fold fits reuse earlier ones and overlap on the cores
Show answer & explanation

Correct answer: B

WHY B: GridSearchCV supports parallel execution through n_jobs, which runs parameter-fold fits concurrently on a single machine. WHY NOT A: random_state controls reproducibility, not parallelism. WHY NOT C: A single split does not create the same tuning parallelism; it only reduces the number of fits per candidate. WHY NOT D: verbose controls how much progress logging is printed; it is not the mechanism that schedules fits across cores. WHY NOT E: warm_start reuses the previous solution when the same estimator is refit; it does not run parameter-fold evaluations concurrently.

3 Data Processing

A data engineer is building a Spark MLlib Pipeline to prepare a categorical string column city for a logistic regression model. The city column contains string values like 'Seattle', 'Austin', 'Chicago'. Which sequence of preprocessing steps correctly implements one-hot encoding for this column so that it is suitable for a linear model?

  1. AApply OneHotEncoder directly to the string city column with handleInvalid='keep', then use VectorAssembler; OneHotEncoder accepts string columns and performs the string-to-index conversion internally
  2. BApply OneHotEncoder to the city column, then StandardScaler to the binary vector, then VectorAssembler; rescaling the indicator values is the step that makes encoded categories usable by logistic regression
  3. CApply StringIndexer to convert city strings to numeric indices and pass those indices to the model; LogisticRegression applies one-hot encoding internally to index columns produced by StringIndexer
  4. DApply StringIndexer to convert city strings to numeric indices, then OneHotEncoder to turn those indices into sparse binary vectors, then VectorAssembler to combine them with the other numeric features
  5. EApply HashingTF to hash each city string into a fixed-length sparse vector, then VectorAssembler; HashingTF is the recommended MLlib step for nominal string columns and replaces StringIndexer and OneHotEncoder
Show answer & explanation

Correct answer: D

WHY D: StringIndexer -> OneHotEncoder -> VectorAssembler is the standard PySpark MLlib pattern for categorical features in linear models. StringIndexer converts string categories to integer indices (required before OHE), OneHotEncoder converts those indices to sparse binary vectors (one indicator per category, dropLast=True by default to avoid multicollinearity), and VectorAssembler combines all features into the single 'features' vector expected by MLlib models. WHY NOT A: PySpark's OneHotEncoder does not accept string columns; it requires integer-indexed input produced by StringIndexer first, and passing strings directly raises a type error. WHY NOT B: OneHotEncoder cannot accept string inputs directly; in addition, StandardScaler is inappropriate for binary indicator features already on the [0,1] scale and is not a substitute for StringIndexer. WHY NOT C: LogisticRegression does not auto-apply one-hot encoding to StringIndexer output; it treats the integer indices as ordinal numeric values, introducing a false ordering between city categories. WHY NOT E: HashingTF is designed for text feature extraction: it hashes the words of a document into a frequency vector. It is not the recommended approach for a single nominal column like city, and it can introduce hash collisions.

4 Model Deployment

A team has written custom Python inference logic that is not covered by a built-in MLflow flavor. They want to deploy it with Databricks Model Serving. Which preparation path is correct?

  1. AZip the notebook source and upload it directly to a serving endpoint without using MLflow model packaging
  2. BSave only the fitted weights to DBFS, because endpoints infer the rest of the Python environment automatically
  3. CLog the model in MLflow format, typically with pyfunc for custom code, register it, and then create a serving endpoint
  4. DRegister a Delta table as a model version, because serving endpoints can execute SQL tables as Python models
  5. EExport the model to CSV and point the endpoint at the file path so Databricks can discover the predict function
Show answer & explanation

Correct answer: C

WHY C: Databricks custom model serving expects an MLflow-packaged model, often via pyfunc for arbitrary Python logic, followed by registration and endpoint creation. WHY NOT A: A raw notebook is not the required serving artifact. WHY NOT B: Dependencies and model packaging must be captured properly. WHY NOT D: Delta tables are not model artifacts. WHY NOT E: CSV files do not define deployable MLflow model logic.

Take the full practice test free →

Study by topic

Practice questions by exam section

Drill one section at a time — each page has fresh questions with full explanations.

Browse all 45 free Machine Learning Associate exam questions & answers → Every question of the free full-length test, each on its own page with the answer and explanation.

Why it works

Practice tests beat re-reading the docs

Find your weak spots in 20 minutes instead of 20 hours.

Take a timed test

Full-length, under exam conditions — no signup needed to try.

See your breakdown

Score plus a topic-by-topic analysis of where you lost points.

Study what matters

Focus on your two or three weakest areas — every explanation teaches the concept.

Retake until ready

Consistently above 80%? You're ready to book the real exam.

FAQ

Frequently asked questions


Is this Databricks Machine Learning Associate practice test free?

Yes. You can take a full-length Databricks Machine Learning Associate practice test on TestLogicHub without paying or entering a credit card.

How many questions are on the real Databricks Machine Learning Associate exam?

The exam has 45 multiple-choice questions and a 90-minute time limit. It costs USD 200 per attempt and is proctored.

What score do I need to pass?

Databricks does not publish an exact cut score. A safe target is to score consistently above 80% on full-length practice tests before booking the real exam.

Are these questions like the real exam?

The questions are mapped to the official exam guide sections, written in the scenario style of the real exam, and every answer comes with a full explanation.

Where can I find Databricks Machine Learning Associate exam questions and answers?

Every question in the free Databricks Machine Learning Associate practice test has its own page on TestLogicHub with the correct answer and a full explanation, organised by exam section. Searching for leaked real exam questions or dumps breaks the Databricks certification terms and can get a certification revoked, so practise with original questions instead.

Does TestLogicHub cover other Databricks certifications?

Yes — TestLogicHub has practice tests for the Databricks Data Engineer Associate and Professional, Data Analyst, Machine Learning Associate, and Generative AI Engineer certifications.

Ready to find your weak spots?

Take the free Databricks Machine Learning Associate practice test — timed, weighted, and explained like the real thing.

Start now — it's free →