Google Professional Machine Learning Engineer Question 363
Select 3Google Cloud PlatformYou are tasked with developing an end-to-end ML pipeline for a retail company to predict future sales. The pipeline must include data ingestion, preprocessing, model training, and deployment to production on Google Cloud. During pipeline design, which considerations are most critical to ensure scalability and reliability?
- A
Use Google Cloud Dataflow for distributed data preprocessing to handle large datasets.
- B
Train the model locally on a small subset of data to reduce costs.
- C
Implement automated retraining using Vertex AI Pipelines to accommodate data drift.
- D
Deploy the model using AI Platform (Vertex AI) with managed endpoints for scalable predictions.
- E
Schedule batch inference jobs manually to control costs.
Show answer and explanation
Correct answers: A, C, D
Explanation
When designing an end-to-end ML pipeline, scalability and reliability are paramount. Using Google Cloud services like Dataflow for preprocessing, Vertex AI Pipelines for automated retraining, and Vertex AI managed endpoints for deployment ensures the pipeline can handle large datasets, adapt to changing data patterns, and deliver predictions reliably at scale. Local training on small datasets and manual job scheduling do not meet these requirements and may introduce inefficiencies or risks in production.
- A. Correct.
Google Cloud Dataflow is a fully managed service designed for scalable and distributed data processing, making it a critical choice for preprocessing large datasets in an end-to-end ML pipeline.
- B. Incorrect.
Training the model locally on a small subset of data is cost-effective but does not align with scalability or production-readiness, as it may lead to underfitting and poor generalization.
- C. Correct.
Automated retraining using Vertex AI Pipelines is essential for maintaining model accuracy over time, especially when data drift occurs, which is common in production environments.
- D. Correct.
Deploying the model using Vertex AI managed endpoints ensures scalability and reliability, as it provides features such as autoscaling, monitoring, and version management.
- E. Incorrect.
Manually scheduling batch inference jobs is error-prone and does not align with best practices for automation and scalability in end-to-end ML pipelines.