Google Professional Machine Learning Engineer Question 303
Select 3Google Cloud PlatformYou are a machine learning engineer at a retail company. Your team has trained a recommendation model to suggest products to users based on their browsing history. The model will serve millions of users daily and needs to provide personalized recommendations within 100 milliseconds for each request. Additionally, the team also wants to generate daily aggregate reports for trending product recommendations using the same model. Which combination of tools would be the most appropriate to address these requirements?
- A
Deploy the model as an online endpoint using Vertex AI for real-time predictions.
- B
Use BigQuery ML to serve the model for real-time predictions.
- C
Use Dataflow to batch-process user data daily for trending product reports.
- D
Deploy the model to Dataproc for real-time predictions.
- E
Use BigQuery to store logs of online predictions and generate daily aggregate reports.
Show answer and explanation
Correct answers: A, C, E
Explanation
To meet the requirements, you need to serve real-time predictions with low latency and also generate daily batch reports. Vertex AI is the best tool for deploying the model as an online endpoint for real-time predictions. Dataflow is suitable for batch processing data to generate daily reports. BigQuery can be used to store logs of predictions and perform SQL-based analytics for aggregate reporting. This combination of tools leverages Google Cloud's capabilities to handle both real-time and batch inference effectively.
- A. Correct.
Deploying the model as an online endpoint using Vertex AI is the most suitable option for serving real-time predictions with low latency, meeting the requirement of providing recommendations within 100 milliseconds.
- B. Incorrect.
BigQuery ML is designed for training and batch inference on large datasets but is not optimized for low-latency, real-time predictions, so this option is not suitable for the real-time prediction requirement.
- C. Correct.
Dataflow is well-suited for batch-processing large-scale data, making it an excellent choice for generating daily aggregate reports for trending product recommendations.
- D. Incorrect.
Dataproc is primarily used for big data processing and distributed computing frameworks like Hadoop or Spark. It is not optimized for low-latency, real-time predictions, so it is not suitable for serving the recommendations in this scenario.
- E. Correct.
BigQuery is an ideal choice for storing logs of online predictions and performing analytics, such as generating daily aggregate reports, due to its scalability and ability to handle large datasets efficiently.