Exploring and Preprocessing Data for ML PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A healthcare company is building a model to predict patient readmission risk. The training data includes patient names, email addresses, and medical records. The data science team needs to use this data for model training, but the compliance team requires that patient identity cannot be recovered from the training dataset. What should the team do before training?

   • ARemove the name and email columns and replace patient IDs with pseudonymous identifiers
   • BUse the data as-is but restrict access to the training pipeline with IAM roles
   • CEncrypt the entire dataset with a customer-managed key and store it in Cloud Storage
   • DAggregate the data at the zip code level and remove all individual patient records
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   A machine learning engineer needs to preprocess a small dataset (a few hundred MB) that fits in memory on a single machine. The preprocessing involves custom Python transformations that are not easily expressed in SQL. Which tool is most appropriate for this task?

   • ABigQuery
   • BIn-memory Python (e.g., pandas)
   • CApache Spark
   • DDataflow
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   Which of the following is a key principle of differential privacy?

   • ARestricting access to the dataset to authorized users only.
   • BRemoving all personally identifiable information from the dataset.
   • CAdding random noise to query results to protect individual records.
   • DEncrypting the dataset with a strong encryption algorithm.
   Reveal Solution Discussion Select an answer first
5. 14application · medium

   A retail company has a 2 TB dataset of customer purchase history stored in Cloud Storage as CSV files. The data science team needs to compute daily aggregate statistics (total sales, unique customers, and average order value) across 3 years of data to understand seasonality trends before building a forecasting model. The team is comfortable with SQL and wants to minimize infrastructure management overhead. What should they use?

   • AUse Cloud Dataflow with a Python pipeline to read and aggregate the CSVs
   • BRead the CSVs into a pandas DataFrame in a single VM and compute aggregates in Python
   • CLoad the CSVs into a BigQuery table and run SQL aggregation queries
   • DRun an Apache Spark job on a Dataproc cluster to compute the aggregates
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   A financial services company has multiple teams building machine learning models. The fraud detection team has engineered a 'transaction frequency in last 24 hours' feature that is proving highly predictive. The marketing team wants to use the same feature for a customer churn model. Both teams currently maintain separate copies of the feature computation logic in their training pipelines. What should the ML platform team do to ensure consistency and reduce duplication?

   • AAsk the marketing team to copy the fraud team's feature engineering code into their pipeline
   • BHave the marketing team retrain their model using the fraud team's training dataset
   • CStore the feature values in a shared Cloud Storage bucket as a Parquet file
   • DCreate the feature in Gemini Enterprise Agent Platform Feature Store and have both teams read from it
   Reveal Solution Discussion Select an answer first