Developing End-To-End ML Pipelines PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard

   A media company needs to build an ML pipeline that processes video files, extracts audio, transcribes it, and then runs sentiment analysis. The pipeline has a mix of long-running GPU jobs (transcription) and lightweight CPU tasks (sentiment analysis). They need fine-grained control over resource allocation and want to avoid vendor lock-in. They have a small platform team that can maintain infrastructure. Which orchestration approach is most appropriate?

   • AUse a self-managed Airflow deployment on GKE with custom Kubernetes operators for each step.
   • BUse a custom Python script with Cloud Functions for each step, triggered by Cloud Storage events.
   • CUse Cloud Composer with standard Airflow operators and a single Dataproc cluster for all processing.
   • DUse Vertex AI Pipelines with pre-built components for transcription and sentiment analysis.
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   Which technique involves splitting the training data into multiple subsets to train and evaluate the model multiple times, reducing variance in the performance estimate?

   • ACross-validation
   • BRegularization
   • CFeature scaling
   • DData augmentation
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   What is the purpose of using a holdout set in model validation?

   • ATo perform feature engineering on the training data
   • BTo evaluate the model's performance on data it has not seen during training
   • CTo store the model's predictions for later analysis
   • DTo increase the size of the training dataset
   Reveal Solution Discussion Select an answer first
5. 9foundation · easy

   What is training-serving skew?

   • AA difference in the hardware used for training versus serving
   • BA difference in the model's accuracy on training data versus validation data
   • CA difference between how data is preprocessed during training and how it is preprocessed during serving
   • DA difference in the number of training examples versus serving examples
   Reveal Solution Discussion Select an answer first
6. 10foundation · easy

   What is the primary purpose of data validation in an ML pipeline?

   • ATo monitor the runtime performance of a deployed model
   • BTo ensure that the data meets expected schema, range, and distribution requirements before it is used for training or serving
   • CTo select the best-performing model from a set of candidates
   • DTo transform raw data into a format suitable for model training
   Reveal Solution Discussion Select an answer first