Associate Data Practitioner · Sample questions

Associate Data Practitioner sample questions with answers and explanations

10 real questions from the CertifyDeck Associate Data Practitioner bank, spread across the domains of the official Google Cloud Certified Associate Data Practitioner skills outline, with the correct answer and a full explanation of why the other options are wrong. They are original questions written from the public syllabus, not exam dumps, and they are the same style you will find in the practice tests and mock exams.

1. Question 1Domain: Data Analysis and PresentationEasy

   A LookML view has a raw_event_stream dimension that a join's sql_on references, but business users find it confusing in the Explore field picker. You must keep the dimension available for modeling while removing it from the field picker. What should you do?

   • A.Delete the dimension from the view file and keep the join unchanged.
   • B.Add a label parameter that renames the dimension to Internal stream.
   • C.Add hidden: yes to the dimension definition in the view file.
   • D.Add a description parameter that tells users not to select the field.
   Show answer and explanation

   Correct answer: C. Add hidden: yes to the dimension definition in the view file.

   Explanation

   The hidden parameter keeps a field available for modeling but hides it from the field picker; it keeps the picker clean but is not a security feature. Deleting the dimension would break the join that references it, and a label or a description only changes how the field is named or described while it stays visible.

   Official references

   • hidden (for fields)
2. Question 2Domain: Data Analysis and PresentationMedium

   You are planning a classification model that flags payment transactions for manual fraud review. Missing a fraudulent transaction is very costly, while reviewing a legitimate transaction only costs a few minutes of an analyst's time. Which metric should you prioritize when you compare candidate models?

   • A.Mean absolute error, so that the average size of the errors is as small as possible.
   • B.Precision, so that most transactions sent for review are actually fraudulent.
   • C.The Davies-Bouldin index, so that the transaction groups are well separated.
   • D.Recall, so that as few fraudulent transactions as possible go undetected.
   Show answer and explanation

   Correct answer: D. Recall, so that as few fraudulent transactions as possible go undetected.

   Explanation

   When false negatives are the costly error, the documentation recommends optimizing for recall, which measures how many of the examples that should get the label actually get it. Precision is the choice when false positives are the expensive error, which isn't the case here; mean absolute error is a regression metric and the Davies-Bouldin index evaluates k-means clustering.

   Official references

   • AutoML beginner's guide
   • BigQuery ML model evaluation overview
3. Question 3Domain: Data Analysis and PresentationEasy

   Your team trains classification models in BigQuery ML and wants every new model to appear automatically in Model Registry once training finishes, so that it can be managed alongside other ML models. The models must not be exported from BigQuery ML. What should you do?

   • A.Create each model as a remote model with REMOTE WITH CONNECTION DEFAULT in its statement.
   • B.Run EXPORT MODEL to Cloud Storage after training and import the files into Model Registry.
   • C.Set MODEL_REGISTRY = 'VERTEX_AI' and a VERTEX_AI_MODEL_ID in the CREATE MODEL options.
   • D.Set ENABLE_GLOBAL_EXPLAIN = TRUE in the CREATE MODEL options so the model is published.
   Show answer and explanation

   Correct answer: C. Set MODEL_REGISTRY = 'VERTEX_AI' and a VERTEX_AI_MODEL_ID in the CREATE MODEL options.

   Explanation

   When you set the MODEL_REGISTRY option in CREATE MODEL, the model is registered and automatically displays in Model Registry on Gemini Enterprise Agent Platform (formerly Vertex AI Model Registry, the name in the exam guide) once training completes, without exporting it. Exporting and importing breaks the no-export requirement, ENABLE_GLOBAL_EXPLAIN only enables global explanations, and remote models can't be registered.

   Official references

   • Manage BigQuery ML models in Gemini Enterprise Agent Platform
4. Question 4Domain: Data Preparation and IngestionMedium

   Your company already runs mature Apache Beam pipelines that clean and aggregate order data before it reaches the warehouse. The data is moving to BigQuery, where reserved capacity must be kept mainly for the analysts' concurrent queries. You want to keep the existing transformation logic. Which approach should you use?

   • A.Use ELT: load the raw order data into BigQuery and clean it with scheduled BigQuery data preparations.
   • B.Use reverse ETL: export the raw order data from BigQuery and transform it in the source systems.
   • C.Use ELT: load the raw order data into BigQuery and rewrite the logic as Dataform SQL workflows.
   • D.Use ETL: run the existing Beam pipelines on Dataflow and write only the transformed data to BigQuery.
   Show answer and explanation

   Correct answer: D. Use ETL: run the existing Beam pipelines on Dataflow and write only the transformed data to BigQuery.

   Explanation

   The BigQuery documentation says ETL is beneficial when you already have a transformation process in place or want to reduce resource usage in BigQuery, and Dataflow runs Apache Beam pipelines as a managed ETL service. Both ELT options run the transformations inside BigQuery, consuming the capacity reserved for analysts and discarding the existing Beam logic. Reverse ETL moves results out of BigQuery and does not prepare incoming data.

   Official references

   • Introduction to loading, transforming, and exporting data
   • Dataflow overview
5. Question 5Domain: Data Preparation and IngestionMedium

   Your team builds a BigQuery orders table with Dataform. Every time the workflow updates the table, it must check that order_id is never null and never repeated, and the team must be alerted if the check fails. You want the simplest configuration. What should you do?

   • A.Add a rowConditions assertion with the expression order_id is not null to the config block.
   • B.Create a Dataform unit test that mocks the source tables and compares an expected result set.
   • C.Add nonNull and uniqueKey assertions for order_id to the config block of the table's SQLX file.
   • D.Create a Knowledge Catalog data profile scan on the orders table and review its distinct values.
   Show answer and explanation

   Correct answer: C. Add nonNull and uniqueKey assertions for order_id to the config block of the table's SQLX file.

   Explanation

   Built-in Dataform assertions in a table's config block run every time Dataform updates the workflow and alert you if they fail; nonNull asserts that columns are never null and uniqueKey asserts that no two rows share the same value. A rowConditions assertion with a not-null expression covers nulls but not duplicates. A unit test checks the transformation logic against mock inputs rather than the real table, and a profile scan reports statistics without failing the workflow. Knowledge Catalog was formerly called Dataplex, the name in the exam guide.

   Official references

   • Test data quality
6. Question 6Domain: Data Preparation and IngestionEasy

   You are a data analyst at a retail company. A BigQuery table of supplier records has inconsistent date formats, extra spaces and missing values. You want AI-generated cleaning suggestions that you can preview and apply in BigQuery Studio, with little code and no other service to provision. What should you do?

   • A.Open the table in BigQuery data preparation and apply the Gemini suggestions.
   • B.Write an Apache Beam pipeline that cleans the table and run it on Dataflow.
   • C.Create a Cloud Data Fusion instance and clean the table with Wrangler directives.
   • D.Create a Managed Service for Apache Spark cluster and run a cleaning job.
   Show answer and explanation

   Correct answer: A. Open the table in BigQuery data preparation and apply the Gemini suggestions.

   Explanation

   BigQuery data preparation uses Gemini in BigQuery to suggest cleaning, transformation and enrichment steps that you preview on sample data and apply from BigQuery Studio. Cloud Data Fusion Wrangler is also visual, but it requires provisioning a separate Cloud Data Fusion instance, which the analyst wants to avoid. An Apache Beam pipeline on Dataflow and a job on Managed Service for Apache Spark (formerly Dataproc, the name in the exam guide) both require code and their own resources.

   Official references

   • BigQuery data preparation overview
7. Question 7Domain: Data Pipeline OrchestrationMedium

   A Dataflow pipeline occasionally throws an exception in a transform that parses incoming records. You want to see the actual elements that were being processed when the exception occurred, without changing the pipeline code. What should you use?

   • A.The Recommendations tab, which lists insights on job performance.
   • B.The Throughput per step chart in the Job metrics tab.
   • C.Data sampling, with exception sampling of the related elements.
   • D.The Cost tab, filtered to the time window of the failures.
   Show answer and explanation

   Correct answer: C. Data sampling, with exception sampling of the related elements.

   Explanation

   Data sampling lets you observe the data at each step of a pipeline, and exception sampling, which is enabled by default, collects samples when the pipeline throws an exception so that you can see the elements involved. Recommendations give performance and cost insights, the Cost tab estimates resource cost, and the throughput chart shows volumes per step, but none of them shows the failing elements.

   Official references

   • Use data sampling to observe pipeline data
8. Question 8Domain: Data Pipeline OrchestrationMedium

   Your team needs to chain several HTTP-based Cloud Run services and API calls into a stateful sequence that runs thousands of times per day with millisecond latency between steps. The solution must be serverless, scale to zero and be billed per step executed. Which service should you use?

   • A.Managed Service for Apache Airflow, with one DAG for each sequence.
   • B.BigQuery pipelines, with a SQL query task for each service call.
   • C.Workflows, with the sequence of calls defined in YAML.
   • D.Managed Service for Apache Spark Workflow Templates on a cluster.
   Show answer and explanation

   Correct answer: C. Workflows, with the sequence of calls defined in YAML.

   Explanation

   Workflows orchestrates HTTP-based services into durable, stateful workflows with millisecond execution latency; it is serverless, scales down to zero and uses usage-based billing per step executed. Managed Service for Apache Airflow has latency of seconds between tasks and a provisioned scaling and billing model. Spark Workflow Templates run Spark and Hadoop jobs on clusters, and BigQuery pipelines run SQL, notebooks and data preparations rather than calls to HTTP services.

   Official references

   • Choose Workflows or Managed Airflow for service orchestration
9. Question 9Domain: Data ManagementMedium

   Field teams upload daily inventory files to a Cloud Storage bucket from a shared upload tool. The tool's service account must add new objects but must not read, overwrite or delete any existing file. Which role should you grant on the bucket?

   • A.Storage Legacy Bucket Writer (roles/storage.legacyBucketWriter).
   • B.Storage Object Admin (roles/storage.objectAdmin).
   • C.Storage Object Creator (roles/storage.objectCreator).
   • D.Storage Object User (roles/storage.objectUser).
   Show answer and explanation

   Correct answer: C. Storage Object Creator (roles/storage.objectCreator).

   Explanation

   Storage Object Creator allows a principal to create objects but not to view, delete or overwrite them, which is exactly what the upload tool needs. Storage Object User is the tempting choice, but it also grants read, update and delete access to objects. Storage Object Admin grants full control of objects, and Storage Legacy Bucket Writer allows creating, overwriting and deleting objects.

   Official references

   • IAM roles for Cloud Storage
10. Question 10Domain: Data ManagementMedium

    You publish a pricing dataset to customers through BigQuery sharing. Your contract allows customers to query the data but forbids them from copying, exporting or saving the shared tables into their own tables. How should you configure the listing?

    • A.Enable the data egress restriction options when you create the listing.
    • B.Grant customers BigQuery Data Viewer on the source dataset instead.
    • C.Enforce public access prevention on the project that holds the data.
    • D.Make the data exchange private so that only invited customers can see it.
    Show answer and explanation

    Correct answer: A. Enable the data egress restriction options when you create the listing.

    Explanation

    Data egress options on a listing let publishers stop subscribers from copying, cloning, exporting or snapshotting linked dataset tables, and also block CREATE TABLE AS SELECT and writing to destination tables. A private exchange only controls who can discover and subscribe, not what subscribers can do with the data. BigQuery Data Viewer includes permission to export table data, and public access prevention is a Cloud Storage setting.

    Official references

    • Introduction to BigQuery sharing
    • BigQuery IAM roles and permissions