Databricks-ML-Assoc ML Workflows — 50 Questions

Question 1

Which TWO of the following statements regarding the MLflow Model Registry in Databricks are true?

Accepted Answer

Model versions are immutable once registered in the registry.. The MLflow Model Registry acts as a centralized repository for managing the full lifecycle of a machine learning model, including versioning, stage transitions, and annotations. Understanding the registry is crucial for MLOps, as it provides a single source of truth for production-ready models, enabling team collaboration and automated deployment pipelines by standardizing how models are promoted across development, staging, and production environments in a secure and audited manner.

Answer

Models can be registered from any local directory regardless of the tracking server.

Answer

Deleting a registered model version also deletes the raw training data used to build it.

Answer

The Model Registry automatically triggers retraining when a model version is updated.

Question 2

Refer to the exhibit. You are logging a model to the MLflow Model Registry. What is the primary purpose of providing the model signature and input example shown in the exhibit?

Accepted Answer

To enable the deployment platform to validate input types for inference requests.. Providing a model signature and input example allows MLflow to define the expected schema for the model's inputs and outputs. This metadata is used by Databricks Model Serving to validate incoming requests, ensuring that the input features match the types and order expected by the model. This prevents runtime errors and is a critical practice for maintaining stable production inference pipelines in a distributed, multi-user environment.

Answer

To increase the training speed of the model during the logging phase.

Answer

To automatically convert the model into a different framework like ONNX.

Answer

To reduce the physical storage size of the logged model artifacts.

Question 3

When using the Databricks Feature Store, what is the primary benefit of logging the feature table metadata when training a model?

Accepted Answer

It allows the system to automatically handle feature lookups during inference.. Logging feature store metadata allows for automatic tracking of feature lineage. When a model is logged with its feature dependencies, MLflow captures which features were used, enabling better traceability and auditability. This is critical for ML workflows to ensure consistency, as it allows the system to automatically lookup the correct feature values for inference, eliminating training-serving skew where different feature logic might accidentally be applied in production compared to development.

Answer

It automatically scales the compute cluster when the feature store is queried.

Answer

It enables the model to be trained on data that hasn't been processed yet.

Answer

It converts all features into a standard binary format for faster training.

Question 4

Your organization requires that all models deployed to production are signed off by a human reviewer. Which Databricks feature should be used to enforce this workflow?

Accepted Answer

Use MLflow Model Registry stage transitions with manual approval steps.. The MLflow Model Registry provides stage transition states (Staging, Production, Archived). By using permissions or integrating with external CI/CD tools, organizations can enforce a gatekeeping process where a human must manually approve a transition to 'Production'. This workflow is crucial for governance, ensuring that every production model has undergone a rigorous review process, which minimizes the risk of deploying flawed or insecure models into mission-critical business environments.

Answer

Use Delta Live Tables to restrict access to the production data table.

Answer

Enable 'Auto-Deploy' in the MLflow Model Registry settings.

Answer

Set the MLflow experiment status to 'ReadOnly' until a review is complete.

Question 5

Refer to the exhibit. A data scientist is setting up a Databricks Workflow task. Which statement accurately describes the function of the 'base_parameters' field?

Accepted Answer

It overrides the internal notebook code variables during job execution.. The 'base_parameters' field allows users to pass key-value pairs into a notebook at runtime. This capability is crucial for parameterizing workflows, enabling the same code to be reused with different configurations without code modifications. This promotes cleaner code management and supports hyperparameter tuning workflows where varying inputs are required for the same training script in different job runs, enhancing the modularity and reusability of the machine learning pipeline.

Answer

It defines the hardware specifications for the job cluster.

Answer

It restricts the notebook access to specific Databricks folders.

Answer

It automatically schedules the job to run every 10 minutes.

Question 6

When using the Databricks Feature Store, why is it critical to use the 'feature_lookup' feature when training a model?

Accepted Answer

It enforces point-in-time joins to prevent data leakage.. The 'feature_lookup' mechanism ensures that the exact same feature transformations are applied to both training and inference data. This eliminates training-serving skew, a common cause of poor model performance in production. By binding the features to the model artifact at training time, the Feature Store automatically retrieves the consistent, pre-computed feature values during inference, ensuring that the model receives data in the same format it was trained on.

Answer

It automatically generates synthetic training data for small datasets.

Answer

It allows the model to train directly on raw, unformatted JSON files.

Answer

It significantly reduces the latency of the training process itself.

Question 7

Refer to the exhibit. Why is this error occurring in the MLflow Model Registry workflow?

Accepted Answer

The registry workflow requires a mandatory approval step before promotion.. The error indicates that the organization has implemented a governance policy within the Model Registry. These policies are designed to enforce quality and security standards by preventing the automatic or unauthorized promotion of models to production. By requiring explicit approval, the platform ensures that only validated models are deployed, minimizing risk and ensuring compliance with organizational operational procedures, which is critical in regulated industries where model transparency is mandatory.

Answer

The model artifact file size exceeds the allowed limit for production.

Answer

The model version 1 is not compatible with the production compute type.

Answer

The user attempting to promote the model does not have Git access.

Question 8

Which TWO of the following statements are correct regarding the Databricks Model Registry?

Accepted Answer

Models can be transitioned through custom stages like 'Staging' and 'Production'.. The Model Registry provides a centralized platform for managing the lifecycle of ML models, including versioning and stage transitions. Understanding these features is vital for maintaining a robust MLOps pipeline where models transition from development to staging and finally to production. Proper stage management prevents unauthorized deployments and ensures that only validated, high-quality models reach the end-user environment, reducing the risk of production incidents.

Answer

The Registry automatically retrains models when data drifts beyond a specific threshold.

Answer

You can only register models that were trained using Scikit-learn or PyTorch.

Answer

The Registry allows multiple models to share the same name without versioning.

Question 9

Which component of MLflow provides a centralized UI to compare runs, visualize metrics, and manage model versions?

Accepted Answer

MLflow Tracking Server.. MLflow Tracking and the Model Registry work together within the Databricks platform to provide a single pane of glass for ML lifecycle management. The Tracking server records experiments, while the Registry manages the transition of artifacts to production. Having a unified UI is vital for teams to collaborate, share results, and ensure that the best models are promoted, reducing the overhead of managing disparate tools for tracking and deployment.

Answer

MLflow Projects.

Answer

MLflow Recipes.

Answer

MLflow Model Serving.

Question 10

Refer to the exhibit. What is the most likely cause of this error?

Accepted Answer

The feature 'user_age' was not logged as part of the model's feature dependencies.. The error suggests a discrepancy between the features defined in the model artifact and the features available in the Feature Store. This often occurs when a feature is deleted or renamed, or when the model is deployed to an environment that does not have the necessary Feature Store metadata correctly initialized. It highlights the importance of managing dependencies between model versions and their source feature engineering pipelines.

Answer

The model was trained on a cluster with insufficient RAM.

Answer

The input data provided to the inference request contains too many rows.

Answer

The user does not have permission to view the model artifacts in the Registry.

Question 11

You are training a model on a massive dataset and want to use Hyperopt to perform hyperparameter tuning across a distributed cluster. Which approach is most effective for scaling this process on Databricks?

Accepted Answer

Configuring the cluster to use high-concurrency mode with SparkTrials.. Using the `SparkTrials` class within Hyperopt allows for distributed hyperparameter tuning across worker nodes. This approach parallelizes the search process, enabling the evaluation of multiple hyperparameter combinations concurrently across the cluster. This is significantly faster than single-threaded tuning for large-scale models, as it leverages the full compute capacity of the Spark environment to explore the search space efficiently and converge on optimal parameters sooner.

Answer

Running standard trials inside a for-loop on a single-node driver.

Answer

Using the MLflow autologging feature to optimize parameters automatically.

Answer

Wrapping the training code in a pandas UDF for parallelization.

Question 12

A machine learning engineer is building a Databricks Job that trains a model and logs it with MLflow. The engineer wants to ensure that the exact library dependencies and environment used during training are captured so the model can be reliably reproduced later. Which approach should the engineer take?

Accepted Answer

Train the model inside an MLflow run created with mlflow.start_run(), which automatically logs the environment.. MLflow runs capture environment details automatically when created via mlflow.start_run(), including Python version, dependencies, and source. This ensures that a model logged within that run can be reproduced with the same libraries, which is critical for reliable retraining and audit. Manually logging environment files does not guarantee fidelity to the actual runtime.

Answer

Call mlflow.log_artifact() on the conda.yaml file produced by mlflow.run.

Answer

Use mlflow.set_experiment() to attach the environment metadata to the experiment.

Answer

Call mlflow.log_artifact() on the requirements.txt file after training completes.

Question 13

A data scientist is using Databricks Feature Store to create a feature table for a model that will be served in real time. The feature table will be used both for training and for online inference. Which TWO of the following statements are correct regarding the necessary configuration? (Choose two.)

Accepted Answer

The feature table must be published to an online store to support real-time serving.. To serve a model in real time with Feature Store, the feature table must be published to an online store so that low-latency lookups are possible. Additionally, the model must be logged with feature store metadata so that Model Serving knows which features to retrieve and how to join them. These two steps enable seamless online inference.

Answer

The feature table must be registered in the Databricks Model Registry.

Answer

The training dataset must be created using a Delta table that includes the primary key and timestamp columns.

Answer

The feature table must be stored in a Unity Catalog metastore.

Question 14

An ML engineer is building a training pipeline in Databricks that reads a large Delta table, applies a scikit-learn model, and logs parameters, metrics, and the model artifact using MLflow. The engineer notices that every run logs the full training dataset as an artifact, which is slow and unnecessary. The engineer wants to log only the model, metrics, and parameters while still enabling full reproducibility. Which MLflow feature should be used to capture the environment and code state without logging the dataset?

Accepted Answer

Enable MLflow autologging for the scikit-learn flavor, which automatically logs parameters, metrics, and model without logging the dataset.. MLflow autologging for scikit-learn automatically logs parameters, metrics, and the model artifact while avoiding logging the training dataset. It also captures the environment and source code, ensuring reproducibility without the overhead of storing large datasets. This is the intended use of autologging to streamline tracking and reduce storage costs.

Answer

Use mlflow.start_run() with the nested=True parameter to suppress artifact logging.

Answer

Use mlflow.log_artifact() to log only the model file and a requirements.txt manually.

Answer

Set the MLFLOW_TRACKING_URI to a Delta table location and log the dataset as a Delta table.

Question 15

An ML engineer wants to ensure that a model registered in the Databricks Model Registry automatically transitions to the Production stage after passing a validation notebook that computes accuracy on a holdout set. They want this transition to be triggered from a Databricks Job. Which approach correctly implements this?

Accepted Answer

Use the MLflow Client's transition_model_version_stage method in the validation notebook to move the model version to Production if accuracy exceeds the threshold.. Automating stage transitions requires calling the Model Registry API after validation. The MLflow Client's transition_model_version_stage method is the supported, permission-aware way to move a version to Production. Webhooks only notify; they do not transition. The MLmodel file and the update endpoint do not control stage. Thus the validation notebook should invoke the client method.

Answer

Configure a Model Registry webhook that listens for model version creation and automatically transitions to Production.

Answer

Set the model version's stage to Production directly in the MLmodel file before logging the model.

Answer

Use the Databricks REST API endpoint /api/2.0/mlflow/model-versions/update to change the stage after validation.

Question 16

An ML engineer is using MLflow Tracking to log experiments for a model. They want to ensure that the experiment can be reproduced later and that the best model can be registered to the Databricks Model Registry. Which TWO actions should they take? (Choose two.)

Accepted Answer

Log the model using mlflow.log_model with a signature that defines input and output schema, and register it to the Model Registry using the registered_model_name parameter.. To reproduce an experiment, you need to log the code version, hyperparameters, metrics, and the model artifact with its signature. Registering the model to the Model Registry requires logging it with a registered model name. Logging the Git commit hash and parameters ensures you can recreate the exact run, while the signature enables proper model serving and validation.

Answer

Store the training dataset in DBFS and log its path as a tag, but do not log the model itself to avoid large artifacts.

Answer

Use mlflow.autolog() to automatically log all metrics and parameters, and rely on it to register the model automatically.

Answer

Create a separate MLflow experiment for each run to isolate metrics and avoid confusion when comparing runs.

Question 17

A machine learning team is using Databricks Feature Store to build a training dataset for a fraud detection model. They want to ensure that the same feature computation logic is used during both training and online inference. Which TWO practices should they follow? (Choose two.)

Accepted Answer

Log the model with mlflow.pyfunc.log_model and include the feature_lookups parameter to automatically retrieve features at inference time.. Registering the feature table with proper keys and logging the model with feature_lookups are the two practices that ensure consistent feature computation between training and inference. The first centralizes feature definitions; the second embeds retrieval logic into the model so that online serving uses the same features and keys.

Answer

Configure the online store to automatically generate features on the fly from raw data using SQL expressions defined in the model.

Answer

Use a separate feature engineering notebook for online inference that reimplements the same transformations independently.

Answer

Manually copy the feature values into a CSV file and include it as an artifact with the model.

Question 18

An ML engineer is using MLflow Tracking with Databricks to log experiments for a fraud detection model. After several runs, they notice that the run's parameters, metrics, and artifacts are not appearing in the MLflow experiment UI. The engineer confirms that the code calls mlflow.start_run() and mlflow.log_param(). Which configuration is most likely missing to ensure logs are sent to the Databricks-hosted MLflow tracking server?

Accepted Answer

Set the MLFLOW_TRACKING_URI environment variable to databricks.. For MLflow on Databricks to log runs to the managed tracking server, the tracking URI must be set to 'databricks'. This can be done via the MLFLOW_TRACKING_URI environment variable or programmatically with mlflow.set_tracking_uri. Without it, MLflow defaults to a local store, and runs won't appear in the workspace experiment UI. The other options address different aspects but not the core connectivity.

Answer

Enable autologging by calling mlflow.autolog() before starting the run.

Answer

Install the mlflow-skinny package instead of the full mlflow library.

Answer

Create an MLflow experiment in the workspace and set it as the active experiment using mlflow.set_experiment().

Question 19

An ML engineer is deploying a model to production using Databricks Model Serving. The model was trained on a feature table in Databricks Feature Store. The engineer wants to ensure that the same feature transformations are applied during online inference as were used during training. Which approach correctly ensures consistency between training and inference?

Accepted Answer

Use the Feature Store's training_set() method to create a training set and then log the model with the feature table metadata using mlflow.log_model().. When a model is trained using Feature Store's training_set(), the resulting training set includes feature metadata. Logging the model with mlflow.log_model() stores this metadata, enabling Model Serving to automatically retrieve features from the online store and apply the same transformations. This guarantees consistency between training and inference.

Answer

Export the feature table to a CSV file and include it in the model artifact for the serving endpoint to read.

Answer

Log the feature table with the model using mlflow.log_artifact() and manually apply transformations in the serving endpoint.

Answer

Create a separate serving endpoint for each feature and use API calls to combine them at inference time.

Question 20

An ML engineer is evaluating a model with a custom Python function that logs a single numeric metric and a text summary. They want to compare this evaluation across multiple runs in the MLflow experiment UI. Which MLflow tracking function should they use to record both values in one call?

Accepted Answer

mlflow.log_metrics({"f1_score": 0.87}); mlflow.set_tag("summary", "good"). To record a numeric metric and a text summary for cross-run comparison, use mlflow.log_metrics for the number and mlflow.set_tag for the string. mlflow.log_metrics ensures the value appears in the metrics column for sorting and charting, while mlflow.set_tag stores the summary as searchable metadata. Other functions either reject strings or store data as artifacts, which are not directly comparable in the experiment UI.

Answer

mlflow.log_dict({"f1_score": 0.87, "summary": "good"}, "eval.json")

Answer

mlflow.log_metrics({"f1_score": 0.87})

Answer

mlflow.log_params({"f1_score": 0.87, "summary": "good"})