Google Professional Machine Learning Engineer Question 302
Select 2Google Cloud PlatformYou are working for an e-commerce company that wants to personalize product recommendations for users. The company has a pre-trained recommendation model deployed on Vertex AI. They want to provide real-time recommendations while a user is browsing the website and also generate daily batch recommendations for email campaigns. Which combination of Google Cloud services should you use to meet these requirements?
- A
Use Vertex AI Endpoint for online inference and Dataflow for batch inference.
- B
Use BigQuery ML for both online and batch inference.
- C
Use Vertex AI Endpoint for online inference and BigQuery ML for batch inference.
- D
Use Dataflow for online inference and Dataproc for batch inference.
- E
Use Vertex AI Endpoint for both online and batch inference.
Show answer and explanation
Correct answers: A, C
Explanation
The scenario requires both real-time (online) and batch inference capabilities. Vertex AI Endpoint is designed for low-latency, real-time predictions, making it suitable for online inference. For batch inference, Dataflow or BigQuery ML are appropriate choices. Dataflow can process large-scale pipelines, while BigQuery ML is suited for batch inference on structured data within BigQuery. Combining Vertex AI Endpoint with either Dataflow or BigQuery ML ensures both requirements are met.
- A. Correct.
Vertex AI Endpoint is designed for online (real-time) inference, while Dataflow can process large-scale data pipelines for batch inference. This combination meets the requirements for both real-time and batch recommendations.
- B. Incorrect.
BigQuery ML is primarily designed for training and batch inference directly on BigQuery data. It is not suitable for real-time (online) inference.
- C. Correct.
Vertex AI Endpoint is suitable for online inference, and BigQuery ML can handle batch inference. This combination is valid for the described use case.
- D. Incorrect.
Dataflow is not optimized for online (real-time) inference, and while Dataproc can be used for batch inference, it is not the best choice given the availability of more specialized tools like BigQuery ML.
- E. Incorrect.
Vertex AI Endpoint is not designed for batch inference. It is optimized for real-time predictions, so this option does not meet the batch processing requirement.