Databricks-ML-Assoc Databricks Machine Learning — 30 Questions

Question 1

Refer to the exhibit. An administrator is configuring a Unity Catalog access policy to restrict experiment deletion. Given the JSON snippet provided, what is the impact of this policy configuration?

Accepted Answer

It allows users to delete experiments.. The policy logic defines an 'allow' effect for the 'DeleteExperiment' action. In many security configurations, especially when using explicit deny-all defaults, this allows the designated action. It is critical to understand how Databricks permissions and Unity Catalog policies intersect with MLflow. Misconfiguring these policies can lead to unauthorized data destruction, which is why strict adherence to principle-of-least-privilege is required when defining experiment lifecycle management permissions.

Answer

It denies users the ability to delete experiments.

Answer

It restricts the policy only to the 'mlflow' namespace.

Answer

The policy is invalid and will be rejected by Databricks.

Question 2

When logging a model with MLflow, which THREE of the following are essential components of a model signature?

Accepted Answer

Input schema definition. Model signatures define the expected schema for inputs and outputs, which is vital for validation and deployment. By defining the input schema, output schema, and parameter types, MLflow can perform automatic type checking during inference requests. This prevents runtime errors in production when an API receives unexpected data types, ensuring the model behaves reliably when integrated into larger application architectures or served via a REST API endpoint.

Answer

Model training loss metrics

Answer

Model hyperparameters

Question 3

An ML engineer wants to compute features with point-in-time correctness to prevent data leakage when training a fraud detection model. Which Databricks Feature Store capability should be utilized?

Accepted Answer

Feature lookup function in training set creation using timestamp keys. Point-in-time joins are critical for preventing data leakage by ensuring that training datasets only include feature values that were known at the exact timestamp of each observation. The Databricks Feature Store automatically handles these temporal joins when creating training sets from feature tables containing timestamps.

Answer

Automated online feature sync via Databricks Jobs

Answer

Feature table time-travel queries using Delta Lake versions

Answer

Feature Store point-in-time training set creation API

Question 4

A team is designing a feature engineering pipeline in Databricks. Which TWO considerations are essential when using Feature Store to ensure data consistency between training and inference?

Accepted Answer

Defining a unique primary key for every feature table.. Feature Store ensures that the exact same transformation logic is applied during model training and real-time inference. By using a centralized feature table, the team avoids the 'training-serving skew' that occurs when features are calculated differently in offline and online environments. This consistency is vital for maintaining model performance and reliability, as it prevents the model from receiving features that diverge from the distribution it observed during training.

Answer

Re-calculating features manually in the inference code.

Answer

Always storing features as raw parquet files.

Answer

Ignoring the time-series nature of data.

Question 5

A machine learning engineer needs to tune a deep learning model using Hyperopt on Databricks. Which THREE steps are required to implement distributed hyperparameter tuning effectively?

Accepted Answer

Define a search space using Hyperopt's hp functions.. Distributed tuning with Hyperopt requires configuring the search space, the objective function, and the Trials object to handle parallelism across the cluster. By offloading individual trials to different workers, the process significantly reduces training time. Proper configuration ensures that the driver node orchestrates the search efficiently while workers execute the heavy computation, which is vital for complex models that would otherwise take too long to train on a single node.

Answer

Hard-code all hyperparameters within the training loop.

Answer

Restrict training to a single CPU core only.

Question 6

A team is building a real-time churn prediction system. Which THREE requirements are critical for the Online Store in this Databricks Feature Store architecture?

Accepted Answer

Sub-millisecond latency for feature lookups.. The Online Store must support low-latency lookups, high availability, and efficient point-in-time joins to be effective for real-time inference. These features ensure that when a customer triggers an event, the system can immediately fetch their current profile and history to generate an accurate, real-time prediction. Failure to meet these criteria results in unacceptable latency and inaccurate model scores, defeating the goal of a real-time service.

Answer

Daily batch processing capabilities for bulk updates.

Answer

Direct integration with local CSV storage.

Question 7

During a model deployment test, you notice that the model's inference performance is inconsistent. What is the most likely cause, and how can you mitigate it?

Accepted Answer

The model is not logged with its environment dependencies.. Performance inconsistencies often arise from environment differences between training and serving. Databricks handles this by capturing the exact environment configuration, including library versions, in the MLflow model artifact. Ensuring that these dependencies are explicitly logged and replicated in the production environment is key to consistent inference performance. This practice mitigates 'it works on my machine' issues and stabilizes the model's behavior across different computational stages in the MLOps lifecycle.

Answer

The model is too large for the network.

Answer

The input data has too many missing values.

Answer

The model was trained on a GPU cluster.

Question 8

A data scientist wants to share their notebook with a team member, but they want to restrict access so the team member can only view the results without modifying the code. What is the correct sharing permission setting?

Accepted Answer

Can Edit. The 'Can View' permission in Databricks allows users to execute the notebook and see results, but prevents them from changing the underlying code. This is essential for collaborative environments where stakeholders need to audit findings or review model performance without accidentally disrupting the development process. Understanding these permission levels is vital for maintaining security and integrity in shared workspaces during machine learning development projects.

Answer

Can Run

Answer

Can Manage

Answer

Owner

Question 9

An organization is preparing to deploy a machine learning model into production using Databricks Model Serving. Which TWO actions must the user perform to ensure the model is ready for high-availability inference?

Accepted Answer

Register the model in the MLflow Model Registry. To deploy models for production, the model must be registered in the Model Registry. Once registered, it must be transitioned to the 'Production' stage or specifically versioned for deployment. Databricks Model Serving handles the infrastructure, but the user must provide the environment configuration, such as defining compute resources and ensuring the model dependencies are correctly specified, to ensure consistent and scalable inference performance in a production environment.

Answer

Manually copy model artifacts to a persistent S3/DBFS bucket

Answer

Disable authentication on the serving endpoint

Answer

Convert the model to a flat JSON file

Question 10

A machine learning engineer has a scikit-learn model registered in Unity Catalog as `prod_ml.churn.final_model` with version 3 tagged `"champion"`. The batch scoring job reads the model with `mlflow.pyfunc.spark_udf(spark, model_uri="models:/prod_ml.churn.final_model/3")`. When a data scientist registers version 4 and sets the alias `champion` to it, the batch job starts scoring with version 4 without any code change. The engineer must guarantee the batch job always scores with the reviewed version 3 until a new version is explicitly approved. What should the engineer do?

Accepted Answer

Score with `models:/prod_ml.churn.final_model/3`, and only update the job to a new version after the version is approved and the alias is moved, keeping promotion as an explicit registry operation.. Immutable numeric version URIs prevent silent model changes, while aliases are movable pointers that can drift. To ensure the job always uses the reviewed version 3 until an explicit approval, the engineer should keep the version-pinned URI and make promotion a separate registry operation. This preserves reproducibility and makes the switch to version 4 an intentional, auditable step.

Answer

Change the model URI to `models:/prod_ml.churn.final_model/3` and rely on Unity Catalog permissions to block access to all later versions.

Answer

Keep the numeric version URI `models:/prod_ml.churn.final_model/3` and require an explicit code or configuration change to move the job to a newly approved version.

Answer

Replace the model URI with `models:/prod_ml.churn.final_model@champion` so the alias resolves consistently in every environment.

Question 11

A machine learning engineer is using Databricks Feature Store to create a training dataset. They define a feature table with a primary key and a timestamp column. They then use create_training_set to generate a training set. They notice that the training set includes features as of the latest timestamp, not as of the timestamp of each label event. What is the most likely cause of this issue?

Accepted Answer

The training set was created without specifying the timestamp column in the feature lookup, so point-in-time correctness was not applied.. Point-in-time correctness in Databricks Feature Store requires specifying the timestamp column when creating the training set. The create_training_set function uses this timestamp to perform time travel on feature tables, retrieving feature values as of each label event. If the timestamp is omitted, it defaults to using the latest feature values, which can cause leakage.

Answer

The label dataset does not have a timestamp column, so Feature Store defaulted to using the latest features.

Answer

The feature table was not registered with a primary key, so Feature Store could not perform time travel.

Answer

The feature table was created as a Delta table without enabling change data feed, so historical values are not available.

Question 12

A data scientist has registered a model in the Databricks Model Registry and wants to deploy it as a real-time endpoint using MLflow Model Serving. They need to ensure that the endpoint automatically scales based on traffic and that the model version is updated without downtime when a new version is registered. Which combination of features should they use?

Accepted Answer

Enable autoscaling on the endpoint and use the MLflow Model Registry webhooks to trigger an update of the served model version when a new version is registered.. Databricks Model Serving endpoints can be configured with autoscaling to handle varying traffic. To update the model version without downtime, you can use MLflow Model Registry webhooks to trigger an automatic update of the endpoint when a new model version is registered. This combination provides scalability and seamless updates.

Answer

Enable autoscaling on the served model and configure the endpoint to use the 'Production' stage; new versions in Production will be automatically picked up.

Answer

Use the Databricks Model Serving UI to create an endpoint with autoscaling enabled and specify the model version; for updates, create a new endpoint and switch traffic using a load balancer.

Answer

Configure the endpoint to serve the 'Staging' stage and manually promote to 'Production' after testing; autoscaling is not available for stage-based endpoints.

Question 13

A machine learning engineer is using Databricks Feature Store to create a feature table. They want to ensure that the feature table can be used for both model training and online serving. Which of the following is a required step when creating a feature table in Databricks Feature Store?

Accepted Answer

Define a primary key and a timestamp key.. When creating a feature table in Databricks Feature Store, you must specify a primary key and a timestamp key. These enable correct point-in-time lookups and ensure data consistency for training and serving. Other options are either optional or not required.

Answer

Enable change data feed on the Delta table.

Answer

Publish the feature table to an online store.

Answer

Partition the feature table by the primary key.

Question 14

A data scientist is training a model using Databricks AutoML. They notice that the generated notebooks include a step that performs cross-validation using a specific metric. They want to change the primary metric used for model selection to area under the ROC curve (AUC). Where should they specify this in the AutoML experiment configuration?

Accepted Answer

In the AutoML experiment setup, set the 'primary_metric' parameter to 'roc_auc'.. Databricks AutoML uses the 'primary_metric' parameter to specify the metric for model selection. To use AUC, set primary_metric='roc_auc'. Other parameter names are not valid and would not achieve the desired effect.

Answer

In the AutoML experiment setup, set the 'evaluation_metric' parameter to 'roc_auc'.

Answer

In the AutoML experiment setup, set the 'optimization_metric' parameter to 'roc_auc'.

Answer

In the AutoML experiment setup, set the 'target_metric' parameter to 'roc_auc'.

Question 15

A data scientist is using Hyperopt with SparkTrials on Databricks to tune a model. They notice that some trials fail due to out-of-memory errors on the driver. Which adjustment should they make to improve the tuning process?

Accepted Answer

Set the SparkTrials parameter to use a smaller number of concurrent trials.. Out-of-memory errors on the driver during SparkTrials hyperparameter tuning often occur when too many trials run concurrently, overwhelming the driver's memory as it collects results. Reducing the number of concurrent trials via the SparkTrials parameter lowers the memory pressure. This is a targeted fix that balances parallelism and resource constraints.

Answer

Increase the number of Spark executors to distribute the load.

Answer

Switch to using Trials instead of SparkTrials to run trials sequentially.

Answer

Increase the driver memory by selecting a larger driver instance type for the cluster.

Question 16

A machine learning engineer is using MLflow Tracking on Databricks to log experiments. They want to ensure that the experiments are reproducible and that all necessary information is captured. Which TWO of the following should they log to achieve this? (Choose two.)

Accepted Answer

Hyperparameters. To ensure reproducibility, logging hyperparameters and evaluation metrics is essential. Hyperparameters define the model configuration, and metrics quantify performance. Other items like commit hash or dataset are either automatically captured or impractical to log fully.

Answer

Source code version control commit hash

Answer

The model's learned weights

Answer

The entire training dataset

Question 17

A machine learning engineer is using Databricks AutoML to train a model on a dataset with a binary target. They want to ensure the model is optimized for recall. Which AutoML configuration should they use?

Accepted Answer

Set the primary metric to recall during AutoML experiment setup. To optimize for recall, you must set recall as the primary metric in Databricks AutoML. This directs the hyperparameter tuning and model selection to maximize recall. Other options either do not change the objective or are not available in AutoML's configuration, so they would not achieve the desired optimization.

Answer

Increase the number of trials in AutoML

Answer

Manually adjust class weights in the training code

Answer

Use the default primary metric (F1 score)

Question 18

A machine learning engineer is deploying a model to a Databricks model serving endpoint. They need to enable auto-scaling to handle variable traffic. Which configuration parameter should they set when creating the endpoint?

Accepted Answer

min_instances and max_instances. To enable auto-scaling for a model serving endpoint, the engineer must configure min_instances and max_instances. These parameters define the lower and upper bounds for the number of instances, allowing the endpoint to automatically scale based on traffic. Other parameters like scale_to_zero_enabled, instance_type, and workload_size do not provide dynamic scaling within a range. Therefore, setting min_instances and max_instances is the correct approach.

Answer

scale_to_zero_enabled

Answer

workload_size

Answer

instance_type

Question 19

A data scientist is building a Feature Store in Databricks. They need to ensure that features are available for both model training and online serving. They have created a feature table and want to publish it to an online store. Which TWO actions must they perform to make the features available for low-latency online inference? (Choose two.)

Accepted Answer

Publish the feature table to an online store using the Databricks Feature Store UI or API.. To serve features online with Databricks Feature Store, you must publish the feature table to an online store and enable Change Data Feed on the source Delta table. Publishing creates the online copy, and CDF allows Databricks to capture and propagate updates to keep the online store synchronized.

Answer

Configure a Databricks SQL warehouse to serve features via JDBC for online inference.

Answer

Register the feature table in the MLflow Model Registry to enable online serving.

Answer

Create a separate Delta table for online features and manually copy data using a scheduled job.

Question 20

A machine learning engineer is deploying an MLflow model to a Databricks model serving endpoint. The model was trained and logged with a custom Python function registered in the model's signature. When the endpoint is invoked, the request fails with an error indicating the custom function is not found. The engineer confirms the model artifact and its dependencies are correctly logged in the MLflow run. Which step should the engineer take to resolve this issue?

Accepted Answer

Package the custom function into a Python module, include it in the model's code directory when logging, and specify it in the model's requirements.. The error indicates the custom function is not found at serving time. MLflow models capture code dependencies only if they are included in the model's code directory or installed as packages. Packaging the function as a module and including it in the code directory ensures it is bundled with the model. Specifying it in requirements ensures installation. This resolves the missing function issue and allows the endpoint to invoke the model successfully.

Answer

Ensure the custom function is defined in the same notebook used for training and re-log the model with the function included as a code dependency.

Answer

Add the custom function to the model's conda environment YAML file and re-log the model.

Answer

Log the custom function as an artifact and update the model's signature to reference the artifact path.