ASSOCIATE-DATA-PRACTITIONER practice question 4 of 25

You are predicting customer churn for a subscription-based service. You have a 50 PB historical customer dataset in BigQuery that includes demographics, subscription information, and engagement metrics. You want to build a churn prediction model with minimal overhead. You want to follow the Google-recommended approach. What should you do?

Full set: 106 questions
Answer options
A: Export the data from BigQuery to a local machine. Use scikit-learn in a Jupyter notebook to build the churn prediction model.
B: Use Dataproc to create a Spark cluster. Use the Spark MLlib within the cluster to build the churn prediction model.
C: Create a Looker dashboard that is connected to BigQuery. Use LookML to predict churn.
D: Use the BigQuery Python client library in a Jupyter notebook to query and preprocess the data in BigQuery. Use the CREATE MODEL statement in BigQueryML to train the churn prediction model. Correct answer
Correct Answer:
D. Use the BigQuery Python client library in a Jupyter notebook to query and preprocess the data in BigQuery. Use the CREATE MODEL statement in BigQueryML to train the churn prediction model.
Explanation

BigQuery ML (BQML) is the Google-recommended approach for building machine learning models on large-scale data stored in BigQuery with minimal overhead. The CREATE MODEL statement allows you to train models directly within BigQuery using SQL, eliminating the need to export data or manage separate compute infrastructure. This serverless approach is highly scalable and cost-effective for petabyte-scale datasets. Using the Python client library in a notebook is a common and efficient way to orchestrate these BQML queries and preprocessing steps.

Why Incorrect

A. Exporting 50 PB of data to a local machine is technically infeasible and impractical due to its massive size.
B. Using Dataproc requires managing a separate Spark cluster, which adds operational overhead and complexity compared to the serverless BQML approach.
C. Looker is a business intelligence and visualization tool; its LookML modeling layer is for defining metrics, not for training predictive machine learning models.

References

1. Google Cloud. (n.d.). Introduction to BigQuery ML. Google Cloud Documentation. Retrieved from https://cloud.google.com/bigquery/docs/bqml-introduction. (See "Why use BigQuery ML" section, which highlights benefits like no data movement and no need for a separate ML cluster).

2. Google Cloud. (n.d.). The CREATE MODEL statement. Google Cloud Documentation. Retrieved from https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-create. (This page details the core SQL statement for training models directly in BigQuery).

Scroll to Top