Exploring and Preprocessing Data for ML PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 5)

Questions 21–24

1. 21application · medium

   A telecommunications company is building a churn prediction model. The dataset contains customer phone numbers and call detail records. The privacy team requires that phone numbers be masked so that the original number cannot be recovered, but the data must still be usable for aggregations like 'number of calls per customer'. What technique should the data engineer apply?

   • AReplace phone numbers with a deterministic hash using a secret salt
   • BRound the phone number to the nearest 1000 to generalize it
   • CRemove the phone number column entirely and use a random customer ID
   • DEncrypt phone numbers with a symmetric key and store the key in Cloud KMS
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A research team has a small dataset of 5,000 labeled images of skin lesions. They need to preprocess the images (resize, normalize, and augment) before training a convolutional neural network. The team is experimenting with different preprocessing strategies and needs fast iteration. What tool should they use?

   • AApache Spark on Dataproc for distributed image processing
   • BIn-memory Python with TensorFlow/Keras preprocessing utilities
   • CBigQuery ML to preprocess and train the model directly
   • DCloud Dataflow to create a batch pipeline for image transformation
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   When consolidating features into a feature store, which practice is most important to ensure consistency and reuse?

   • AEach team should create its own copy of features to avoid conflicts.
   • BFeatures should be stored only in the format used by the original team.
   • CFeatures should be computed on-the-fly for each model to ensure freshness.
   • DFeatures should be defined with clear, documented schemas and versioned to track changes.
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   A marketing analytics team has a dataset of 1 million customer records with mixed data types: numeric (age, income), categorical (education level, occupation), and a timestamp (last purchase date). They need to explore the data to understand distributions and correlations before building a customer segmentation model. The team is comfortable with SQL and wants to avoid managing infrastructure. What should they use?

   • ABigQuery to run exploratory queries and compute statistics
   • BCloud Dataflow to create a pipeline for exploration
   • CIn-memory Python with pandas and seaborn for visualization
   • DApache Spark on Dataproc for distributed exploration
   Reveal Solution Discussion Select an answer first