DBX-MLA Databricks Machine Learning Associate
- Vendor
- Databricks
- Level
- Associate
- Questions
- 48
- Time
- 90 min
- Domains
- 4
- In the bank
- 300
Practice 300 questionsTake a timed mock examDay-by-day study plan
Guides for this exam
- DBX-MLA Study Guide: Two Exams in One, a Model's Life in Unity Catalog and the Arithmetic of Machine Learning · 한국어
- 8 Mistakes That Repeat Across Every Databricks Exam, and How to Spot Each One in the Question · 한국어
The exam at a glance
| Level | Associate |
|---|---|
| Questions | 48 |
| Time | 90 minutes (about 1.9 minutes per question) |
| Passing mark | Not published by the vendor |
Domains and weights
| Domain | Weight | Practice questions here |
|---|---|---|
| D1 · Databricks Machine Learning | 38% | 114 |
| D2 · Data Processing | 19% | 57 |
| D3 · Model Development | 31% | 93 |
| D4 · Model Deployment | 12% | 36 |
51 of the 300 practice questions (17%) ask for more than one answer.
D1 · Databricks Machine Learning 38%
Platform questions ask which Databricks tool fits a step of the ML workflow: AutoML, Feature Store in Unity Catalog, MLflow tracking, the model registry or the ML runtime. The deciding detail is usually what the tool actually stores or speeds up. Strong options match the tool to its real job; weaker ones assume a feature, such as Photon or online sync, works where it does not.
What to watch for
- MLflow keeps a parameter once per run and a metric as a series over steps; runs, artifacts and models are separate things.
- Feature tables in Unity Catalog need a primary key, and online publishing has its own prerequisites such as change data feed.
- Photon speeds up Spark SQL and DataFrame work, not single-node libraries like scikit-learn or XGBoost on pandas.
Tasks in the official exam guide (18)
- 1.1 Identify best practices of an MLOps strategy
- 1.2 Identify the advantages of ML runtimes
- 1.3 Identify how AutoML facilitates model and feature selection
- 1.4 Identify the advantages AutoML brings to model development
- 1.5 Identify the benefits of Unity Catalog feature tables over workspace-level feature tables
- 1.6 Create a feature table in Unity Catalog
- 1.7 Write data to a feature table
- 1.8 Train a model with features from a feature table
- 1.9 Score a model using features from a feature table
- 1.10 Describe the differences between online and offline feature tables
- 1.11 Identify the best run using the MLflow Client API
- 1.12 Manually log metrics, artifacts and models in an MLflow run
- 1.13 Identify information available in the MLflow UI
- 1.14 Register a model with the MLflow Client API in the Unity Catalog registry
- 1.15 Identify the benefits of the Unity Catalog registry over the workspace registry
- 1.16 Identify when promoting code is preferred over promoting models and vice versa
- 1.17 Set or remove a tag for a model
- 1.18 Promote a challenger model to champion using aliases
D2 · Data Processing 19%
Data processing questions give a Spark or pandas DataFrame and ask for the right summary, filter or test. Read the data type and the exact need: exact versus approximate, per group versus overall, continuous versus categorical. Strong options use the API that returns exactly that; weaker ones compute something close but not equal.
What to watch for
- Group-wise statistics come from groupBy with agg, returned as one DataFrame.
- Approximate quantiles are fast; an exact median needs a relative error of 0 or an exact function.
- Choose the statistical test by the feature types: a t-test compares two group means, chi-square compares two categorical features.
Tasks in the official exam guide (9)
- 2.1 Compute summary statistics on a Spark DataFrame (.summary(), dbutils data summaries)
- 2.2 Remove outliers from a Spark DataFrame by standard deviation or IQR
- 2.3 Create visualizations for categorical or continuous features
- 2.4 Compare two categorical or two continuous features with the appropriate method
- 2.5 Compare imputing missing values with the mean, median or mode
- 2.6 Impute missing values with the mode, mean or median
- 2.7 Use one-hot encoding for categorical features
- 2.8 Identify the model types and data sets for which one-hot encoding is or is not appropriate
- 2.9 Identify scenarios where a log scale transformation is appropriate
D3 · Model Development 31%
Model development questions cover training, tuning and evaluation with Spark ML, Optuna or Hyperopt, and plain scikit-learn. Count carefully and look for leakage: many wrong answers come from evaluating on data the model has already seen. Strong options keep the validation set clean and the metric in the units the business uses.
What to watch for
- Split before you balance or transform, or duplicated rows leak into validation and inflate the score.
- A grid of m by n values with TrainValidationSplit fits m times n models plus one final refit; cross-validation multiplies by the folds.
- Load large data once outside the trial function; each distributed trial should get small inputs.
Tasks in the official exam guide (15)
- 3.1 Select the appropriate algorithm for a model scenario using ML foundations
- 3.2 Identify methods to mitigate data imbalance in training data
- 3.3 Compare estimators and transformers
- 3.4 Develop a training pipeline
- 3.5 Use Hyperopt fmin to tune hyperparameters
- 3.6 Perform random, grid or Bayesian search for hyperparameter tuning
- 3.7 Parallelize single-node models for hyperparameter tuning
- 3.8 Describe benefits and downsides of cross-validation over a train-validation split
- 3.9 Perform cross-validation as part of model fitting
- 3.10 Identify the number of models trained in grid search with cross-validation
- 3.11 Use common classification metrics (F1, log loss, ROC/AUC)
- 3.12 Use common regression metrics (RMSE, MAE, R-squared)
- 3.13 Choose the most appropriate metric for a scenario objective
- 3.14 Identify the need to exponentiate log-transformed variables before evaluating metrics or interpreting predictions
- 3.15 Assess the impact of model complexity and the bias-variance tradeoff
D4 · Model Deployment 12%
Deployment questions ask how predictions reach users: batch scoring, streaming or a Model Serving endpoint, and how versions move between them. The deciding detail is when the prediction is needed and who reads it. Strong options pick the simplest mode that meets the timing and refer to models by alias; weaker ones add a real-time endpoint nobody needs.
What to watch for
- Nightly scores read the next morning are a batch job writing to a Delta table, not a serving endpoint.
- Load a model by its Unity Catalog alias (for example @Champion) so jobs follow promotions without edits.
- Serving endpoints split traffic between served models; a configuration update keeps the old version serving until the new one is ready.
Tasks in the official exam guide (6)
- 4.1 Identify differences and advantages of batch, real-time and streaming serving
- 4.2 Deploy a custom model to a model endpoint
- 4.3 Use pandas to perform batch inference
- 4.4 Identify how streaming inference is performed with declarative pipelines (formerly Delta Live Tables)
- 4.5 Deploy and query a model for real-time inference
- 4.6 Split traffic between endpoints for real-time inference
A study order that works
The 19-day study plan for DBX-MLA takes the domains in this order. Each day has a short reading, key terms and a quiz from the bank.
- Days 1–7: D1 · Databricks Machine Learning
- Days 8–11: D2 · Data Processing
- Days 12–16: D3 · Model Development
- Days 17–18: D4 · Model Deployment
- Day 19: Final Review: Decision Cheat-Sheet Across the Four Domains
Recent changes to these questions
- : 1 answer choice set changed, 1 answer key corrected, 1 explanation rewritten
- : 60 questions added
- : 3 answer choice sets changed, 3 answer keys corrected, 3 explanations rewritten
- : Exam added to the bank (240 questions added)
About these questions
Every question is original, written from the public exam guide and checked against the vendor's own documentation. None comes from a real exam. Each one has an explanation of why the answer is right and why the other options are wrong. How the questions are written.