Professional Machine Learning Engineer · 14% of the exam

Collaborating within and across teams: free practice questions

5 sample questions from our 35-question bank for this domain — answers and explanations included. These are the same scenario-based style as the real Google Cloud exam.

1. A data scientist at a retail company is working in Vertex AI Workbench and wants to ensure that all team members can reproduce her Python environment exactly, including specific library versions used for feature engineering. She wants the solution to integrate natively with Vertex AI and require minimal manual configuration for her teammates. What should she do?

  • A. Export a requirements.txt file and commit it to the shared Git repository so teammates can run pip install manually.
  • B. Use a Vertex AI Workbench managed notebook with a custom container image that includes all required dependencies, and share the image via Artifact Registry.✓ Correct
  • C. Document the library versions in a README file stored in the notebook's home directory.
  • D. Install all dependencies in a persistent disk attached to the Workbench instance and share a snapshot of the disk.
Explanation

Using a custom container image stored in Artifact Registry is the recommended, cloud-native approach. It encapsulates all dependencies and configurations, and any team member can launch an identical Workbench instance from the same image, ensuring full reproducibility with minimal manual effort. Option A (requirements.txt) is fragile because pip may resolve slightly different transitive dependency versions across environments and requires manual steps. Option C (README documentation) is purely informational and does nothing to enforce reproducibility. Option D (disk snapshot) is operationally heavy, not version-controlled, and does not integrate cleanly with Vertex AI's managed notebook workflow.

2. Your data engineering team has built a feature engineering pipeline in Dataflow that transforms raw customer events into 50 aggregated features for a churn prediction model. The team wants to ensure these features are computed identically during model training (batch) and serving (real-time). Which approach best addresses this requirement?

  • A. Use Vertex AI Feature Store to register the 50 features with their transformations; create a Dataflow job for batch ingestion and configure Feature Store online serving✓ Correct
  • B. Write the feature transformations once in Python and execute them separately in Dataflow for training and in a Cloud Run service for serving
  • C. Export the feature definitions as a Pandas DataFrame and replicate it across training and serving environments
  • D. Use BigQuery views to materialize features for training and manually recreate the same logic in Vertex AI Predictions for serving
Explanation

Vertex AI Feature Store is purpose-built to ensure feature consistency between batch and online serving by storing feature definitions, transformations, and computed values in a centralized managed service. Option 0 is correct because Feature Store handles both batch ingestion (via Dataflow) and online retrieval with guaranteed consistency. Option 1 is wrong because executing the same logic separately in different systems (Dataflow vs. Cloud Run) risks subtle divergences and lacks versioning. Option 2 is wrong because exporting to Pandas doesn't scale to real-time serving and offers no consistency guarantees. Option 3 is wrong because BigQuery views and manual replication in Predictions duplicates logic and is error-prone.

3. Your team uses Colab Enterprise notebooks for interactive data exploration and feature engineering. A senior data scientist creates a notebook that performs complex feature transformations and wants to package this logic for use in a production Vertex AI Pipelines job. What is the best practice approach?

  • A. Copy the notebook code directly into a Python script for the pipeline component
  • B. Export the notebook as a Python module, add unit tests and input validation, then use it as a reusable component in the pipeline✓ Correct
  • C. Keep the notebook as-is and have the pipeline call the notebook directly via a custom component
  • D. Manually rewrite the feature engineering logic in Go for better performance in production
Explanation

Option B is correct. Converting notebook code into a tested, modular Python module is the production-ready approach. This ensures reproducibility, testability, and reusability. Option A skips testing and introduces technical debt. Option C adds unnecessary complexity and makes the pipeline dependent on notebook infrastructure, which is not designed for orchestration. Option D introduces unnecessary complexity and language switching without clear benefits. Professional ML engineering requires separating exploration (notebooks) from production (versioned modules).

4. A team is implementing a feature for responsible AI that explains model predictions to non-technical stakeholders. They want to provide simple, interpretable insights about which input features most influenced a specific prediction. Which approach is most appropriate?

  • A. Use SHAP values or feature attribution to identify the contribution of each feature to individual predictions; present results with visualizations suitable for non-technical audiences✓ Correct
  • B. Rely solely on global feature importance from the model, which may not explain individual predictions accurately
  • C. Ask stakeholders to manually trace through the model logic to understand predictions
  • D. Avoid explaining predictions to reduce model complexity
Explanation

Option A is correct. SHAP or other feature attribution methods provide instance-level explanations showing how each feature contributed to a specific prediction. Visualizations (e.g., waterfall plots, force plots) make this accessible to non-technical stakeholders. Option B uses global importance, which doesn't explain individual predictions. Option C is not scalable. Option D ignores explainability requirements. Model explainability is a cornerstone of responsible AI, and feature attribution is the standard approach for instance-level interpretability.

5. Your team is using Dataflow to preprocess a large historical dataset for model training. The preprocessing job runs successfully, but a junior engineer realizes they need to modify the filtering logic and re-run the entire job. The engineer wants to minimize compute costs while ensuring data consistency. Which approach is best?

  • A. Re-run the entire Dataflow job with modified code
  • B. Manually edit the output data files to apply the new filtering logic
  • C. Implement the preprocessing logic as a reusable, parameterized Dataflow template; version the template in source control; run a new job instance with updated parameters, which is more cost-effective for iterative changes✓ Correct
  • D. Use a spreadsheet to manually filter and clean the data instead of Dataflow
Explanation

Option C is correct. Dataflow templates allow parameterization and reuse without modifying code, making iterative preprocessing more maintainable and cost-effective. Versioning templates ensures reproducibility. Option A is inefficient when parameters change. Option B introduces data integrity risks and is not reproducible. Option D is not scalable for large datasets. Parameterized templates are the professional approach to iterative data preprocessing in production pipelines.

30 more questions in this domain

Practice the full bank with instant grading, flashcards, and a timed mock exam.

Start practicing free