Databricks-ML-Pro ML Ops Practice Test 9 — 15 Questions

Question 1

A data science team trains a scikit-learn model on a Databricks cluster and needs the same feature-engineering logic to run identically in a nightly batch scoring job and in a real-time Model Serving endpoint. They want a single artifact that encapsulates preprocessing and the estimator. Which approach should they use?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Log the model with MLflow using the sklearn flavor and a custom pyfunc wrapper that includes the preprocessing steps.. Wrapping preprocessing and the estimator in a single pyfunc model lets MLflow serialize both as one artifact. Batch scoring loads that artifact with the pyfunc loader, and Model Serving uses the same artifact, so feature transformations are guaranteed identical and training-serving skew is avoided without duplicating logic.

Answer

Register the raw estimator in Unity Catalog and reimplement preprocessing separately in the batch job and endpoint code.

Answer

Package the preprocessing into a custom container image and reference the image from the model version metadata.

Answer

Log the model with the sklearn flavor and pass the preprocessing function name as a signature parameter.

Question 2

You are preparing a feature-engineering job that reads from a Delta table and writes a feature table. The job must run on a schedule and be idempotent so that reruns after a failure do not duplicate data. Which approach should you use?

Accepted Answer

Use a Databricks Job with a notebook that performs a MERGE INTO the feature table using a deterministic key and a timestamp condition.. A MERGE INTO operation keyed on a deterministic identifier and a condition ensures that repeated runs update existing records instead of inserting duplicates, satisfying idempotency. This is the recommended pattern for scheduled feature-engineering jobs in Databricks that must be resilient to retries. The other options either rely on downstream fixes, destroy incremental state, or use mechanisms not designed for idempotent batch writes.

Answer

Use a Databricks Job with a notebook that uses DataFrame.write.mode("overwrite") to replace the entire feature table on each run.

Answer

Use a Databricks Job with a Delta Live Tables pipeline that writes to a streaming table with a defined primary key and applies AUTO CDC.

Answer

Use a Databricks Job with a notebook that appends new rows and relies on a downstream deduplication step in the serving layer.

Question 3

A team wants a scheduled Databricks job to automatically retrain a demand-forecasting model whenever the upstream feature table receives new data, and to register the resulting model version only if validation metrics improve. Which Databricks capability should they use to orchestrate this?

Accepted Answer

A Databricks job with a table-update trigger that runs the training notebook and a conditional registration task.. Databricks jobs support table-update triggers, letting a run start when the feature table changes. A multi-task job can then train the model and use a conditional task to register the version only when validation metrics beat the current production baseline, all within native orchestration.

Answer

A cron expression on the training notebook that runs every minute to poll for new feature data.

Answer

A Delta Live Tables pipeline that materializes the feature table and automatically promotes models with the highest accuracy.

Answer

An MLflow webhook that fires when a new experiment run is created and calls the registration API.

Question 4

A data scientist has completed hyperparameter tuning with Hyperopt on Databricks and now needs to register the best-performing model to the MLflow Model Registry, including its signature and input example, so that downstream scoring jobs can validate incoming data. The training script uses MLflow autologging. Which approach most reliably captures the signature and input example during registration?

Accepted Answer

Explicitly call mlflow.log_model() (or the flavor-specific log_model) with the signature and input_example arguments inside the training function, then register that logged model.. The signature and input example are artifacts attached at model logging time, so the reliable path is to pass them explicitly to the flavor-specific log_model call used by the training function. Registering afterward from the run URI, editing tags, or hand-editing the MLmodel file does not produce the complete, tooling-readable artifacts needed for downstream data validation.

Answer

Use the MLflow Client's update_model_version() method after registration to attach the signature and input example as tags on the model version.

Answer

Call mlflow.register_model() directly on the run URI returned by the tuning run, because autologging automatically infers and stores the signature and input example.

Answer

Register the model from the MLflow experiment's artifact location by copying the MLmodel file and manually editing it to include the signature and input example fields.

Question 5

A data science team at a retail company has registered a demand forecasting model in the MLflow Model Registry. The model version is currently in the 'Staging' stage and has been validated by the QA team. Before promoting it to 'Production', the ML engineer wants to ensure that the model's performance does not degrade when serving live traffic. They decide to deploy the model to a small percentage of production traffic while continuing to serve the existing model. Which Databricks feature should they use to achieve this?

Accepted Answer

Databricks Model Serving with traffic splitting. Databricks Model Serving natively supports traffic splitting, which lets you distribute incoming requests across multiple model versions on the same endpoint. This is ideal for canary deployments: you can send a small fraction of traffic to the new version, monitor metrics, and then increase the percentage. Other options lack the real-time routing capability required to test a model under live production load without affecting all users.

Answer

Databricks Jobs with conditional task execution

Answer

MLflow Model Registry webhooks

Answer

MLflow Projects with Docker environments

Question 6

Your team is using Databricks Feature Store to serve features to a real-time model deployed on Databricks Model Serving. A data scientist updates the feature computation logic for one of the features and publishes a new version to the online store. However, the model endpoint continues to return stale feature values for that feature. What is the most likely cause?

Accepted Answer

The model was logged with a feature spec that references an older version of the feature table, so it continues to use the old feature values.. The model's feature spec records the versions of the feature tables used during training and logging. When a new feature table version is published, existing models still reference the old version, leading to stale values at inference. To resolve this, the model must be re-logged with an updated feature spec that includes the new version, and the endpoint must be updated with the new model version.

Answer

The model endpoint has not been restarted after the feature table update, so it is still using cached feature values.

Answer

The online store is not configured to automatically sync with the offline store, so the new feature values were never published.

Answer

The model was logged without the feature lookup package, so it cannot query the online store.

Question 7

You are building a Databricks job that trains a model and registers it to the MLflow Model Registry. After registration, you need to automatically transition the model version to 'Staging' only if its validation accuracy exceeds 0.95. The job should fail if the accuracy is below this threshold. Which approach should you implement?

Accepted Answer

Use the MLflow Client to register the model, then use a conditional step in the job to call the transition_model_version_stage method if the accuracy metric meets the threshold.. To conditionally transition a model version based on a metric, you need to programmatically retrieve the metric and then call the MLflow Client's transition method. This allows you to enforce a threshold and fail the job if the condition is not met. Webhooks and REST API calls lack the ability to evaluate metrics before transitioning, and setting the stage during registration does not incorporate a conditional check.

Answer

Configure the MLflow Model Registry webhook to automatically transition the model version to 'Staging' when a new version is registered.

Answer

Use the Databricks REST API to update the model version stage after the job completes, relying on the job's success or failure to indicate accuracy.

Answer

Set the model version's stage to 'Staging' during registration using the stage parameter in mlflow.register_model, and rely on the training script to skip registration if accuracy is low.

Question 8

A team runs a nightly Databricks job that retrains a demand-forecasting model and registers a new version in the MLflow Model Registry. Compliance requires that the model version used for scoring in production be immutably identified and that any subsequent retraining not silently change what production serves. Which practice best satisfies this requirement?

Accepted Answer

Assign a stage or alias to the specific model version and have the scoring job reference the model by that stage or alias, updating it only through a controlled promotion step.. Pinning production to a specific model version through a stage or alias gives an auditable pointer that only changes via an explicit promotion. Referencing the model name alone, inventing new names per run, or reading raw artifact paths all allow production to drift or lose lineage, so they fail the immutability and traceability requirement.

Answer

Register each retrained model under a new registered model name that encodes the training date, and update the scoring job to read the newest name each night.

Answer

Store the model artifact path from the training run in a Delta table and have the scoring job read the path directly, bypassing the Model Registry entirely.

Answer

Have the scoring job reference the model by its registered model name only, so that the latest version is always used after each nightly run.

Question 9

A machine learning engineer is using MLflow Tracking to log metrics and artifacts for a deep learning model. They notice that the training run logs a large number of metrics (e.g., loss per batch) and want to reduce the storage footprint and improve query performance. They also need to retain the ability to compare runs and reproduce results. Which of the following actions is most appropriate?

Accepted Answer

Log only the final epoch metrics and use MLflow's step parameter to log intermediate metrics less frequently.. MLflow's log_metric function can accept a step parameter to record metrics over time, but logging every batch creates a large number of entries. Reducing the logging frequency or only logging final epoch metrics significantly decreases storage and improves query speed while still providing enough data for comparison and reproducibility. Other options either lose MLflow's tracking capabilities or introduce unsupported configurations.

Answer

Disable metric logging entirely and rely on TensorBoard for visualization.

Answer

Use a different tracking server with a NoSQL backend to handle the high volume of metrics.

Answer

Store metrics in a separate file and log it as an artifact instead of using mlflow.log_metric.

Question 10

A data scientist has registered a new model version in MLflow Model Registry and wants to validate it against a golden dataset before promoting it. The validation notebook must run automatically whenever a new version is registered in the 'ChurnModel' registry model, and the result should block promotion if the validation fails. Which approach should the data scientist use?

Accepted Answer

Configure a webhook on the MLflow Model Registry that triggers a Databricks job running the validation notebook, and use the job's success/failure to gate promotion.. MLflow Model Registry webhooks are designed to notify external systems when registry events occur, such as a new model version being created. By configuring a webhook that calls a Databricks job, the validation notebook runs automatically. The job's success or failure can then be used to decide whether to promote the version, providing the required gating mechanism.

Answer

Create a Databricks job with a file arrival trigger that monitors the MLflow artifact store for new model files.

Answer

Enable automatic model promotion in the registry so that any newly registered version is immediately moved to Production after passing built-in validation.

Answer

Use MLflow Projects to package the validation notebook and run it manually each time a new version is registered.

Question 11

A machine learning engineer needs to automate retraining of a model whenever new data lands in a Delta table. The retraining must run on a schedule, use a specific cluster configuration, and send an email alert on failure. Which Databricks feature should be used to orchestrate this workflow?

Accepted Answer

Databricks Jobs with a notebook task, a schedule trigger, and an email notification on failure.. Databricks Jobs is the native orchestration service for scheduled notebooks and scripts. It supports specifying cluster configuration, scheduling, and failure notifications. The other options either lack scheduling, cannot run training code, or are designed for data pipelines rather than ML workflows, so they do not satisfy the combined requirements of schedule, cluster control, and alerting.

Answer

An MLflow Project run triggered by a file arrival event in DBFS using a filesystem watcher.

Answer

A Delta Live Tables pipeline with a continuously running mode that trains the model in a streaming table.

Answer

A Databricks SQL dashboard with a scheduled refresh that calls a stored procedure to retrain the model.

Question 12

You are implementing a CI/CD pipeline for a machine learning model on Databricks. The pipeline must automatically retrain the model when new data arrives, validate its performance, and promote it to production if it meets quality thresholds. Which TWO of the following steps are essential to include in the pipeline to ensure safe and automated deployment? (Choose two.)

Accepted Answer

Implement automated tests that evaluate the model on a holdout dataset and compare metrics against a baseline before promotion.. The essential steps for an automated CI/CD pipeline include registering the model in MLflow Model Registry with stage transitions and implementing automated tests on a holdout dataset to validate performance against a baseline. These steps ensure that only models meeting quality thresholds are promoted, enabling safe automation. Other steps like manual review or storing artifacts in Git are not suitable for automated pipelines.

Answer

Manually review the model's performance metrics in a notebook before allowing the pipeline to proceed.

Answer

Store the model artifacts in a Git repository alongside the code to ensure versioning.

Answer

Configure the pipeline to directly overwrite the production model endpoint with the newly trained model without validation to reduce latency.

Question 13

You are monitoring a model served on a Databricks Model Serving endpoint. You need to track the distribution of incoming request payloads to detect data drift. Which Databricks feature should you use to automatically capture and store inference logs for analysis?

Accepted Answer

Enable inference tables on the model serving endpoint.. Inference tables are a Databricks Model Serving feature that automatically logs request and response payloads to a Delta table. This enables easy querying and monitoring for data drift without writing additional code. Other options either do not capture payloads automatically or are not designed for this purpose.

Answer

Configure the endpoint to log to MLflow Tracking.

Answer

Use Databricks SQL dashboards to query the endpoint's access logs.

Answer

Enable model serving logs to be written to a cloud storage bucket.

Question 14

A machine learning engineer needs to schedule a Databricks job that runs a Python wheel task to execute a packaged training pipeline on a recurring basis. The pipeline code is built into a wheel and stored in Unity Catalog volumes. Which job configuration correctly executes this workload?

Accepted Answer

Create a job with a Python wheel task, specifying the wheel file location and the entry point as package_name.module_name, and attach a cluster or serverless compute.. A Python wheel task is the purpose-built job task type for running packaged Python code: you point it at the wheel and a fully qualified entry point, and Databricks handles installation and invocation. Notebook-based installs or subprocess calls, and Spark submit tasks, do not provide the same reproducible, entry-point-driven execution.

Answer

Create a job with a notebook task that runs %pip install on the wheel and then calls the pipeline via subprocess from within the notebook.

Answer

Create a job with a Spark submit task that passes the wheel as a --py-files argument and runs the training script as the main class.

Answer

Create a job with a notebook task that contains a single cell calling dbutils.library.install on the wheel path, then imports the training module.

Question 15

You maintain a Databricks ML pipeline that trains a model nightly and registers new versions in MLflow Model Registry. A downstream batch scoring job in another workspace loads the model by stage. Auditors require that every production scoring run can be traced back to the exact training data snapshot and code commit. Which approach best satisfies this requirement?

Accepted Answer

Use mlflow.log_input with a Delta table dataset, log the Git commit as a run tag, and register the model version from that run so the run ID links model, data, and code.. The requirement is an immutable, queryable chain from a production model version to the exact training data snapshot and code revision. Logging the dataset via mlflow.log_input and the Git commit as a run tag, then registering from that run, creates that chain because the model version stores the run ID, and the run stores the dataset digest and code tag. Manual descriptions, parameters, signatures, or notebook revisions are mutable or incomplete and cannot satisfy an audit.

Answer

Store the training data path in the model's signature and use the workspace notebook revision to identify the code commit.

Answer

Set the model version's description to the Git commit hash and rely on the registered model name to identify the data snapshot.

Answer

Log the training run with mlflow.log_param for the Git commit and log the Delta table version as a tag, then register the model version from that run.