Collaborating to manage data and models
Free Google Cloud Professional Machine Learning Engineer practice — 6 questions on Collaborating to manage data and models, with explanations. No sign-up. Full 12-question mixed test →
A fraud-detection model requires online feature lookups with p99 latency under 10ms, features are ingested continuously from Pub/Sub, and the training pipeline must generate point-in-time-correct feature snapshots to avoid label leakage. Which architecture BEST satisfies all these requirements on Google Cloud?
Vertex AI Feature Store is purpose-built for this pattern: streaming ingestion (via Dataflow/Pub/Sub) populates the online store optimized for single-digit-millisecond lookups, and its point-in-time lookup capability generates training data without leaking future information.
A financial services organization has ML training data scattered across BigQuery datasets, Cloud Storage buckets, and Dataproc-managed Hive tables across multiple projects. Auditors require centralized data discovery, automated data quality checks, and end-to-end lineage tracking for compliance, without migrating any underlying data. Which service should be used?
Dataplex provides a unified data governance layer across BigQuery, Cloud Storage, and Dataproc metastore sources in-place, offering data discovery/cataloging, automated data quality tasks, and lineage tracking without requiring data migration.
A team has 50TB of feature-engineering logic already implemented as PySpark jobs reading from Cloud Storage. They need to migrate to GCP with minimal code rewrite while orchestrating the jobs as steps within Vertex AI Pipelines, and without managing persistent Spark clusters. Which service should they choose for the Spark workload?
Dataproc Serverless for Spark runs existing PySpark code as-is without provisioning or managing clusters, and can be invoked as a step within Vertex AI Pipelines, satisfying the minimal-rewrite and no-cluster-management requirements.
Multiple versions of a churn-prediction model are registered in Vertex AI Model Registry. The team wants their CI/CD deployment pipeline to always deploy 'the current production model' by referencing a stable identifier, while still being able to instantly roll back by repointing that identifier to a previous version. What should they configure?
Vertex AI Model Registry supports aliases (e.g., 'production', 'default') that point to a specific model version; deployment pipelines can reference the alias, and rollback is achieved simply by reassigning the alias to a different version, with no pipeline code changes.
A data science team wants a fully managed, serverless notebook environment that requires no compute instance management, supports sharing via IAM permissions, and integrates natively with BigQuery Studio for querying without additional setup. Which Google Cloud notebook environment BEST fits these requirements?
Colab Enterprise is a serverless notebook environment tightly integrated with BigQuery Studio and Vertex AI, requiring no underlying VM management, and access is controlled entirely through IAM, matching all stated requirements.
A healthcare ML team must train a model using patient records in BigQuery containing PII (names, SSNs, medical record numbers). Regulations require that feature-engineering data scientists never see raw PII, but the training pipeline still needs a consistent identifier per patient to correctly join data across multiple tables. What is the BEST approach?
Cloud DLP's deterministic encryption/tokenization produces consistent pseudonymous tokens for the same input value, preserving referential integrity for joins, while column-level security or Dataplex policy tags on BigQuery ensure only authorized roles can ever view the raw PII values, satisfying both the technical and regulatory requirements.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$129.99 $34.99 with code FREETEST33 — valid through September 7.
Get my $34.99 deal →