Databricks Certified Machine Learning Professional

Learning plan

About 75 hours of focused study, split into 5 phases. Pick your weekly pace and the timeline adjusts.

Estimated finish: 8 weeks

What the exam covers

Source: official exam guide. Weights can change, so check before you start.

Before you start

• Solid Python plus scikit-learn, PySpark DataFrames and MLflow basics. The Databricks ML Associate level is the natural step before this
• About a year of hands-on Databricks ML work is recommended by Databricks
• A Databricks workspace. Free Edition covers most labs; distributed Ray and online-table work may need a trial workspace

1. Weeks 1–2 · 18 h

   SparkML pipelines and distributed training

   • Decide when SparkML beats a single-node model, based on data size, model type and batch/streaming/real-time inference
   • Build a SparkML Pipeline with StringIndexer, OneHotEncoder, VectorAssembler and an estimator; tune it with CrossValidator; score in batch and streaming
   • Parallelise per-group training with the pandas Function API (applyInPandas) and pandas UDFs
   • Run distributed hyperparameter tuning with Optuna and with Ray, logged to MLflow, and compare Ray with Spark for training
   • Explain data parallelism vs model parallelism, and vertical vs horizontal scaling trade-offs

   Hands-on: On a public tabular dataset (for example NYC taxi or a loans dataset), build a SparkML pipeline with categorical encoding, tune it with CrossValidator, then train one model per region with applyInPandas. Log everything to MLflow using nested runs.

   Use: Exam guide (official PDF) · Optuna on Databricks · Ray on Databricks

2. Weeks 2–3 · 12 h

   Advanced MLflow and Feature Store

   • Use nested MLflow runs and log custom metrics, params and artifacts programmatically
   • Write a custom pyfunc model that does real-time feature engineering inside predict()
   • Build feature tables with the FeatureEngineeringClient and use point-in-time lookups to avoid label leakage
   • Configure online tables or feature serving for low-latency lookups and on-demand features

   Use: Feature engineering and serving docs · Databricks Academy – Machine Learning at Scale

3. Weeks 4–5 · 18 h

   MLOps, testing and environments

   • Explain the deploy-code pattern and map Databricks features to each model-lifecycle stage
   • Write unit tests for notebook functions and design an integration test that covers feature engineering, training, evaluation, deployment and inference
   • Define serving endpoints, MLflow experiments and registered models as Databricks Asset Bundles across dev, staging and prod
   • Trigger automated retraining from a drift or performance alert, and pick the champion model automatically

   Hands-on: Turn your training notebook into a Databricks Asset Bundle with dev and prod targets, plus a job and an MLflow experiment. Add a pytest unit test for one feature function and run `databricks bundle validate` and `deploy` from the CLI.

   Use: Databricks Asset Bundles docs · Databricks Academy – Advanced Machine Learning Operations

4. Weeks 5–7 · 15 h

   Monitoring and deployment

   • Build Lakehouse Monitoring monitors (snapshot, time series and inference profiles) and read the profile and drift metric tables
   • Choose the right statistical drift test for numerical vs categorical features and define custom metrics with slicing
   • Alert stakeholders when drift crosses a threshold, and track endpoint health (latency, error rate, CPU, memory)
   • Compare blue-green and canary rollouts and split traffic between model versions on Model Serving
   • Register a custom pyfunc with artifacts in Unity Catalog and query it through REST or the MLflow Deployments SDK

   Hands-on: Deploy two versions of your model behind one serving endpoint with a 90/10 traffic split, enable inference tables, then attach an inference-profile monitor and add a SQL alert on a drift metric.

   Use: Data quality / Lakehouse Monitoring docs · Model Serving docs

5. Weeks 7–8 · 12 h

   Practice and gap review

   • Answer the sample questions in the exam guide and write down the objective each one tests
   • Re-drill Model Development and ML Ops (88% of the exam combined), especially Feature Store point-in-time lookups and Lakehouse Monitoring table types
   • Complete one timed 59-question practice set in 120 minutes

   Use: Databricks AI Prep Guide · Udemy – Databricks ML Professional practice tests

You're ready to book when…

• You can build, tune and score a SparkML pipeline and explain when you would use Ray or applyInPandas instead
• You can deploy an ML project with Asset Bundles and explain the deploy-code promotion flow
• You can pick the right Lakehouse Monitoring profile type and drift test for a scenario
• You score 80% or higher on a timed practice set

Exam tips

• The September 2025 guide merged the old four sections into three, and monitoring now sits inside ML Ops. Older courses and practice tests list "Solution and Data Monitoring" as a separate section.
• You get 120 minutes for 59 scored questions, so there is time to reread long code-based questions.