4.2 Preparing Data for AI and ML PROFESSIONAL-DATA-ENGINEER Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   When preparing data for batch predictions with a BigQuery ML model, what must be true about the feature columns in the prediction input table?

   • AThey must have the same names and compatible data types as the training data.
   • BThey must be encoded as one-hot vectors.
   • CThey must be in the same order as the training data columns.
   • DThey must be exactly the same rows as the training data.
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   A ride-sharing company has a BigQuery ML model that predicts trip duration. The model was trained on data with features like `distance_km`, `time_of_day` (encoded as sine/cosine), and `weather` (one-hot encoded). For online predictions, the serving data arrives as a JSON payload with raw values. The data engineer needs to ensure that the online prediction requests are processed correctly. What is the best way to handle this?

   • AUse a different model that does not require feature engineering.
   • BRetrain the model to accept raw JSON inputs.
   • CPass the raw JSON directly to the model and let BigQuery ML handle the transformation.
   • DCreate a preprocessing function that transforms the raw JSON into the exact feature format used in training, and call the model with the transformed data.
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   Which Google Cloud service is specifically designed to generate embeddings for text data using pre-trained models?

   • ACloud Translation API
   • BCloud Vision API
   • CVertex AI Embeddings for Text
   • DCloud Natural Language API
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A manufacturing company is building a BigQuery ML model to predict equipment failure. The dataset is time-series data collected from sensors. The data engineer needs to split the data into training and evaluation sets. Which splitting strategy is most appropriate?

   • ASplit chronologically, using the earlier 80% of timestamps for training and the later 20% for evaluation.
   • BSplit based on sensor ID, using some sensors for training and others for evaluation.
   • CUse the most recent 80% for training and the oldest 20% for evaluation.
   • DRandomly split the data into 80% training and 20% evaluation.
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A media company uses a BigQuery ML model to predict user engagement scores. The model was trained with a custom feature where `time_of_day` was transformed into sine and cosine components. For online predictions, the serving data includes a raw `timestamp` column. What should the data engineer do to prepare the serving data?

   • AConvert the timestamp to a string and pass it as a categorical feature.
   • BDrop the timestamp column from the serving data.
   • CPass the raw timestamp directly to the model.
   • DApply the same sine and cosine transformation to the timestamp in the serving data.
   Reveal Solution Discussion Select an answer first