Original questions, written to the 2026 exam guide
Why these and not a dump
These are practice questions, not leaked exam content. Each one teaches the concept the real exam tests, with an explanation you can check against the Databricks docs. Why dumps fail →
Databricks Machine Learning
- During a custom evaluation job, an engineer saves a confusion-matrix image to a local path and wants it attached to the current MLflow run…
- A product manager asks why the ML team used AutoML before hand-tuning a custom pipeline. The team wants to explain a process-level…
- A regression experiment evaluates models with RMSE, and lower values are better. The engineer wants the single best active run using the…
- An engineer wants to register a trained model version in Unity Catalog using MLflow APIs from a notebook. The workspace is Unity Catalog…
- A user can read a registered model in Unity Catalog but cannot add or delete its tags. According to the current Databricks documentation…
- A fraud detection service needs sub-second feature retrieval for real-time scoring, while the data science team also needs full historical…
- A batch inference workflow loads a model with a URI like models:/prod.ml_team.fraud_detection@Champion. Why is this alias-based pattern…
- A production registered model in Unity Catalog already has version 3 marked as Challenger after validation. Version 2 currently serves as…
- A data scientist wants to create a new Unity Catalog feature table for customer features using SQL. Which requirement must be satisfied for…
- A team used AutoML for a churn classification problem and now wants to understand the exact transformations and training code behind the…
- A company has three Databricks workspaces for development, staging, and production. Multiple teams need to discover and reuse governed…
- A team wants full control over what is recorded during a training run rather than relying on autologging. Which pattern correctly performs…
- A streaming feature pipeline computes rolling session features from an event stream and needs to keep a Unity Catalog feature table updated…
- A batch inference job uses a model packaged with two feature lookups: account_creation_date and num_lifetime_purchases. The scoring…
- A recommender model needs customer and product features stored in Unity Catalog feature tables. The training dataset already contains…
- A foundation model fine-tuning job is extremely expensive, and retraining the same model separately in staging and production would add…
- A team computes updated customer aggregates every night and wants to refresh only the corresponding rows in an existing Unity Catalog…
Model Development
- A hospital readmission model will trigger costly interventions. The team wants the predicted probability itself to be trustworthy because…
- A forecasting team wants an evaluation metric that penalizes large prediction errors more strongly than small ones because a few bad misses…
- A linear baseline underfits both the training and validation sets on a complex nonlinear problem. Which change is most likely to reduce…
- An engineer has four hyperparameters, each with many plausible values, and only enough budget to test a limited number of configurations…
- A Hyperopt search space uses hp.choice to select among model options. After fmin finishes, the returned value for that choice is not the…
- A fraud dataset has only 1 percent positive examples. A data scientist wants the training procedure itself to place more emphasis on the…
- A model flags rare safety incidents. The team wants one metric that balances precision and recall so neither false positives nor false…
- A team wants a search strategy that uses results from earlier trials to guide which hyperparameter settings to test next instead of…
- A data scientist builds a scikit-learn Pipeline with several preprocessing steps and one final learning step. Which rule about pipeline…
- A regression target was log-transformed before training to stabilize variance. The team wants the training object itself to handle inverse…
- A business stakeholder wants a regression metric that is easy to explain as the typical absolute error in the same units as the target and…
- Why do practitioners often prefer k-fold cross-validation over a single train-validation split when estimating model performance during…
- A regressor was trained on log(y) instead of y. The analyst now wants to report RMSE in dollars because the business users think in…
- A classifier is being trained on an imbalanced medical dataset. The team wants each validation fold to preserve roughly the same class…
Data Processing
- A machine learning engineer is performing exploratory data analysis on a house_price column, which is a continuous numeric feature. They…
- A team is building a gradient boosting classifier (GBTClassifier in Spark MLlib) to predict customer churn. The dataset includes a zip_code…
- A data scientist is building a PySpark MLlib preprocessing Pipeline for a dataset with three columns: annual_salary (double)…
- A data engineer needs to remove outliers from a transaction_amount column in a Spark DataFrame using the Interquartile Range (IQR) method —…
- A data scientist wants to explore a large Databricks Spark DataFrame to understand numeric distributions, string column frequencies, and…
- A data scientist wants to clean a salary column in a large Spark DataFrame by removing rows where salary is more than 3 standard deviations…
- A data scientist is working with a dataset containing two columns with missing values: temperature_reading (a continuous numeric feature…
- A data scientist is preparing features for a linear regression model. The dataset contains a subscription_tier column with 3 nominal…
- A data scientist has a loan_amount column with 8% missing values. Exploratory analysis reveals that the distribution is heavily…
Model Deployment
- A team compares streaming inference with nightly batch scoring. What is the main advantage of the streaming design for event-driven use…
- A data scientist wants to score a stored MLflow model against a pandas DataFrame inside Python code, without calling a remote serving…
- 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…
- A retailer needs nightly predictions for 40 million rows and can tolerate results arriving hours later. The team wants the cheapest and…
- An operations team receives a steady stream of sensor events and wants model inference embedded inside a managed pipeline that processes…
Score it like the real exam
The same 45 questions, timed, with a section-by-section breakdown at the end. Free, no credit card.
Start the free ML Associate practice test →