Developing End-To-End ML Pipelines PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 4)

Questions 16–20

1. 16foundation · easy

   Which technique helps ensure consistent data preprocessing between training and serving?

   • AUsing the same preprocessing code and libraries in both training and serving
   • BTraining the model on raw data without any preprocessing
   • CUsing different preprocessing steps for training and serving to improve performance
   • DPreprocessing data manually before training and serving
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A startup is building their first ML pipeline for churn prediction. They have a tight deadline and a small team. They want to use a template that includes data validation, training, and evaluation components, and they want to deploy it to Vertex AI Pipelines. What is the most efficient way to get started?

   • AUse Cloud Composer to orchestrate a series of Cloud Functions for each step.
   • BUse a single Vertex AI Training job that includes all steps in one script.
   • CUse a pre-built pipeline template from the Vertex AI Pipelines template gallery and customize the training component for their model.
   • DBuild a custom pipeline from scratch using the Kubeflow Pipelines SDK to have full control.
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   Which managed service on Google Cloud provides a fully managed Apache Airflow environment for orchestrating workflows, including ML pipelines?

   • ACloud Composer
   • BDataflow
   • CAI Platform Training
   • DVertex AI Pipelines
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A logistics company runs a daily ML pipeline that trains a delivery-time prediction model. Recently, the pipeline failed because a new data source started sending a string value in a column that was previously numeric. The team wants to catch such issues early and prevent pipeline failures. What should they add to the pipeline?

   • AA manual review step where a data engineer inspects the data before each training run.
   • BA retry mechanism in the pipeline that automatically reruns the failed step.
   • CA monitoring alert on the training job's CPU utilization to detect when the pipeline is about to fail.
   • DA data validation step using TensorFlow Data Validation (TFDV) to check the schema and detect anomalies before training.
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   When should you consider building a custom pipeline solution instead of using a managed service or template?

   • AWhen you have unique orchestration requirements that are not supported by existing services or templates
   • BWhen you have a very simple pipeline that can be built in minutes
   • CWhen you want to avoid writing any code
   • DWhen you need a solution that is guaranteed to be bug-free
   Reveal Solution Discussion Select an answer first