Why Study with PlanetCert?
The Latest Questions
Practice questions and exam topics aligned with the current exam objectives.
Detailed Explanations
Go beyond the answer. Master the material with comprehensive learning and professional explanations for every concept.

AI-Powered Insights
Personalized preparation guidance that adapts to your performance and identifies weak spots automatically.
Exam Information
Official specifications published by Databricks
Exam Format
Registration
Validity
Skills this practice test covers
- Decide when a Spark ML model fits a use case better than a single-node model.
- Scale training and hyperparameter tuning with Spark ML, the Pandas Function API, Optuna and Ray.
- Track complex experiments in MLflow with nested runs, custom metrics and custom model objects.
- Keep feature lookups point-in-time correct and serve features to online applications.
- Define ML assets in Databricks Asset Bundles and test pipelines with unit and integration tests.
- Detect drift with Lakehouse Monitoring and roll out new model versions with Databricks Model Serving.
Explore related topics
ML-PRO exam content
3 domains · 11 topics · 197 of 205 questions sorted by domain
Model Development86 questions
This section is about building and scaling models on Databricks. Candidates decide when Spark ML fits, spread training and tuning across a cluster, track complex experiments in MLflow, and keep feature lookups correct in time.
Skills tested
- Choose a Spark ML model or a single-node model for batch, streaming or real-time inference.
- Build, tune and evaluate a Spark ML pipeline with the right estimators and transformers.
- Run distributed hyperparameter tuning with Optuna or Ray, and log each trial to MLflow.
- Train one model per group of data with the Pandas Function API.
- Build feature pipelines with the FeatureEngineering Client and serve on-demand features.
Questions by topic
- Using Spark ML22
- Advanced MLflow Usage16
- Advanced Feature Store Concepts19
- Scaling and Tuning29
MLOps78 questions
MLOps keeps a model healthy after its first release. Candidates design lifecycle pipelines, test them, define environments as code, retrain automatically and watch for drift with Lakehouse Monitoring.
Skills tested
- Design a deploy code pipeline that moves training code from development through test to production.
- Write unit tests for notebook functions and an integration test for the whole ML pipeline.
- Define serving endpoints, MLflow experiments and registered models in Databricks Asset Bundles.
- Trigger retraining from drift or performance alerts, and keep the best model that results.
- Build a monitor for a snapshot, time series or inference table, and alert on drift thresholds.
Questions by topic
- Environment Architectures17
- Automated Retraining10
- Drift Detection and Lakehouse Monitoring22
- Model Lifecycle Management13
- Validation Testing16
Model Deployment33 questions
This section covers putting models in front of users. Candidates compare rollout strategies for high-traffic applications and serve custom PyFunc models through Databricks Model Serving.
Skills tested
- Compare blue-green and canary rollouts for an application with heavy traffic.
- Implement a model rollout strategy with Databricks Model Serving.
- Register a custom PyFunc model and log its artifacts in Unity Catalog.
- Query a served model through the REST API or the MLflow Deployments SDK.
Questions by topic
- Deployment Strategies17
- Custom Model Serving16
Requirements
- No formal prerequisites: anyone can book the exam.
- Recommended: a year or more of hands-on Databricks work, plus the Machine Learning at Scale and Advanced Machine Learning Operations courses.
- Recommended: working knowledge of Python, scikit-learn, Spark ML and MLflow.
- Recommended: practical experience with Lakehouse Monitoring and Databricks Model Serving.
About the ML-PRO exam
The Machine Learning Professional exam (ML-PRO) asks whether an engineer can take a model from a Spark ML pipeline to a monitored serving endpoint on Databricks. Its outline has three sections: Model Development, MLOps and Model Deployment. MLflow runs through all three, and the MLOps and deployment work also relies on Unity Catalog, Databricks Asset Bundles and Lakehouse Monitoring. Passing earns the Databricks Certified Machine Learning Professional certification, which is valid for two years.
Model Development has the most questions in the PlanetCert practice test, with MLOps close behind, and Scaling and Tuning is its largest single topic. The explanations show why one approach suits production, for example why point-in-time lookups stop data leakage, or why a canary rollout suits a busy endpoint. Study Mode suits the drift questions, where the right statistical test depends on whether a feature is numerical or categorical. Brainy AI Assistant can walk through a bundle definition or compare Ray with Spark for one training job.
How to earn the certification
- Earns
- Databricks Certified Machine Learning ProfessionalProfessional
The certification is valid for two years; holders recertify by passing the version of the exam that is live at the time.
- Before
- ML-ASSOC Machine Learning Associate
Databricks' associate exam for machine learning, the level below this one. It covers the basics that this professional exam builds on.
- ML-ASSOC Machine Learning Associate
How to prepare for ML-PRO
- 1
Study the current three-section outline
Databricks changed this outline on September 30, 2025. Older notes built around Experimentation or Solution and Data Monitoring describe the earlier version, so check each topic against Model Development, MLOps and Model Deployment.
- 2
Build one model from end to end
In a Databricks workspace, train a Spark ML pipeline, log it with nested MLflow runs, register it in Unity Catalog and serve it. Then point a Lakehouse Monitoring monitor at its inference table.
- 3
Tune the same model two ways
Scaling and Tuning has the most practice questions of any topic. Run one hyperparameter search with Optuna and again with Ray, log both to MLflow, and note where each one fits better.
- 4
Package the project as a bundle
Describe the experiment, the registered model and the serving endpoint in Databricks Asset Bundles, deploy them to two targets, and add unit tests for the notebook functions.
- 5
Rehearse rollouts, then time a full run
Compare a blue-green and a canary rollout on a serving endpoint. Then take the ML-PRO practice test in Timed Mode, and ask Brainy AI Assistant about every MLOps question that went wrong.
Start a timed ML-PROrun
Databricks recommends its Machine Learning at Scale and Advanced Machine Learning Operations courses, taught by an instructor or taken self-paced in Databricks Academy.
Who ML-PRO is for
- Machine learning engineers who deploy and maintain models on Databricks.
- MLOps engineers who own testing, retraining and monitoring for production models.
- Data scientists who move models from experiments into production systems.
- Holders of Machine Learning Associate who want the professional-level certification.
Typical job titles: Machine learning engineer · MLOps engineer · Data scientist · ML platform engineer · AI/ML architect · Senior data engineer
ML-PRO glossary
18 terms from the exam objectives and the vendor's official material, each defined in a sentence.
- Blue-green deployment
- A release method that runs the old and the new model version side by side and then moves all traffic to the new one at once.
- Canary deployments
- Releases that give a new model version a small part of the traffic first, and more of it only while the new version performs well.
- Data parallelism
- A training strategy that gives each worker a full copy of the model and a different slice of the data. Model parallelism splits the model itself instead.
- Databricks Asset Bundles
- Project files that define Databricks resources, such as jobs, experiments and serving endpoints, as code, so the same project deploys the same way to each environment.
- Drift metrics table
- The Lakehouse Monitoring table that compares current data with a baseline or an earlier time window, using statistical tests to show drift.
- Feature Serving
- A Databricks endpoint that returns feature values, including on-demand features computed at request time, to applications.
- FeatureEngineering Client
- The Databricks Python client for creating feature tables, building training sets and scoring models with automatic feature lookups.
- Inference table
- A table that logs the requests a served model receives and the predictions it returns, so that its quality can be tracked over time.
- Lakehouse Monitoring
- The Databricks feature that profiles a table over time and writes data quality and drift metrics to metric tables.
- MLflow Deployments SDK
- The MLflow client that creates, manages and queries Databricks serving endpoints from code.
- MLflow flavors
- Formats that tell MLflow how to save and load a model built with a given library, such as scikit-learn or Spark ML.
- Nested runs
- MLflow runs started inside a parent run, used to group related trials such as the steps of a hyperparameter search.
- Online tables
- Read-only copies of feature data kept in a fast store, so real-time applications can look up features with low latency.
- Optuna
- An open-source library for hyperparameter tuning; on Databricks its trials can run in parallel and log to MLflow.
- Point-in-time correctness
- Joining each training row only to the feature values that existed at that row's timestamp, so no future data leaks into training.
- pyfunc
- MLflow's generic Python function flavor, which lets any model be loaded and called through one predict interface.
- Ray
- An open-source framework for distributed Python, used here to spread training and tuning work across a cluster.
- Unity Catalog
- The Databricks governance layer that holds registered models, tables and the permissions on them.