Google Professional Machine Learning Engineer Question 436
Single answerGoogle Cloud PlatformYou are a machine learning engineer at a retail company working on a recommendation system. Your team is iterating on multiple models and needs a solution to track and compare model versions, hyperparameters, and evaluation metrics. Which approach should you take using Google Cloud services?
- A
Use Vertex AI Experiments to log and compare model artifacts, hyperparameters, and metrics across different runs.
- B
Manually save model metadata, hyperparameters, and metrics in a Google Cloud Storage bucket for tracking purposes.
- C
Deploy each model version to Vertex AI Model Registry and compare performance in production.
- D
Store model artifacts and metadata in BigQuery tables and write custom SQL queries to track and compare performance.
Show answer and explanation
Correct answer: A
Explanation
Vertex AI Experiments is the most suitable solution for tracking and comparing model artifacts, hyperparameters, and metrics. It is specifically built for this purpose, providing an integrated and efficient way to manage multiple training runs and their associated metadata, which is crucial in iterative model development processes.
- A. Correct.
This is the correct answer. Vertex AI Experiments is specifically designed to track, log, and compare model artifacts, hyperparameters, and evaluation metrics efficiently across multiple training runs.
- B. Incorrect.
Although storing metadata in a Cloud Storage bucket is possible, this approach lacks the automation and visualization capabilities provided by Vertex AI Experiments.
- C. Incorrect.
Deploying models to Vertex AI Model Registry is useful for versioning deployed models, but it doesn't provide tracking or comparison of training runs and artifacts.
- D. Incorrect.
BigQuery is a powerful analytics tool, but it requires significant custom work to store, query, and analyze model metadata, making it a less efficient solution compared to Vertex AI Experiments.