Google Professional Machine Learning Engineer Question 300
Select 2Google Cloud PlatformYour team is developing a machine learning model to predict customer churn for a subscription-based service. The model needs to make predictions in two scenarios: (1) real-time predictions when a customer interacts with the service and (2) daily batch predictions to analyze churn risk for all customers. Which combination of Google Cloud services would best meet these requirements?
- A
Use Vertex AI for real-time predictions and Dataflow for batch predictions.
- B
Use BigQuery ML for real-time predictions and Dataproc for batch predictions.
- C
Use Vertex AI for real-time predictions and BigQuery ML for batch predictions.
- D
Use Dataproc for real-time predictions and Dataflow for batch predictions.
Show answer and explanation
Correct answers: A, C
Explanation
This scenario requires a solution that addresses both real-time online inference and batch inference. Vertex AI is optimized for real-time online predictions as it allows for seamless deployment of machine learning models. For batch inference, Dataflow and BigQuery ML are both suitable options, depending on the data pipeline and storage setup. Vertex AI and BigQuery ML together provide a robust solution for such hybrid use cases, making options 1 and 3 the correct answers.
- A. Correct.
Vertex AI is designed for real-time online inference with pre-deployed models, while Dataflow is well-suited for batch processing pipelines. This combination satisfies both real-time and batch requirements effectively.
- B. Incorrect.
While BigQuery ML can be used for batch inference, it is not optimized for real-time predictions. Dataproc is also more suited for distributed data processing tasks rather than real-time predictions.
- C. Correct.
Vertex AI supports online inference for real-time predictions, and BigQuery ML is an excellent choice for batch predictions over large datasets stored in BigQuery. This is a valid solution for the given use case.
- D. Incorrect.
Dataproc is not designed for real-time predictions, and Dataflow is better suited for batch processing pipelines rather than real-time inference. This combination would not meet the requirements.