Google Professional Machine Learning Engineer Question 377
Select 3Google Cloud PlatformYou are tasked with deploying a third-party ML pipeline management tool, such as MLflow, on Google Cloud. The tool should enable tracking experiments, managing models, and serving them, while ensuring scalability and minimizing infrastructure management overhead. Which combination of services and strategies should you use?
- A
Deploy MLflow on a Google Kubernetes Engine (GKE) cluster with autoscaling enabled for flexible scaling.
- B
Use Cloud Storage to store MLflow artifacts such as models and experiment results.
- C
Host MLflow using a Compute Engine VM and manually configure load balancing for scalability.
- D
Use Cloud SQL to store MLflow's backend database for experiment tracking.
- E
Leverage BigQuery for storing MLflow artifacts instead of Cloud Storage.
Show answer and explanation
Correct answers: A, B, D
Explanation
To effectively host a third-party ML pipeline tool like MLflow on Google Cloud, you should prioritize scalability and minimize manual infrastructure management. GKE with autoscaling is ideal for hosting the application, Cloud Storage is best for storing artifacts, and Cloud SQL is a managed solution suitable for MLflow's backend database. This combination ensures a scalable, reliable, and low-maintenance deployment.
- A. Correct.
Deploying MLflow on a GKE cluster with autoscaling enabled is a recommended approach for scalability and minimizing manual infrastructure management. GKE provides container orchestration and allows you to scale workloads efficiently.
- B. Correct.
Using Cloud Storage for storing artifacts is a standard practice in Google Cloud. It is highly scalable, durable, and integrates well with MLflow for managing large volumes of data.
- C. Incorrect.
Hosting MLflow on a Compute Engine VM is not ideal for scalability as it requires significant manual effort for configuration, scaling, and maintenance. This approach defeats the purpose of minimizing infrastructure management overhead.
- D. Correct.
Using Cloud SQL as the backend database for MLflow ensures reliable and managed database services for experiment tracking. It simplifies setup and maintenance compared to self-hosting a database.
- E. Incorrect.
BigQuery is not suitable for storing artifacts like models and experiment results. While it is excellent for analytical queries, artifacts are better suited to object storage like Cloud Storage.