Google Professional Machine Learning Engineer Question 434
Single answerGoogle Cloud PlatformYou are managing a machine learning project on Google Cloud and need to track and compare different model training runs. The goal is to identify the best-performing model based on evaluation metrics and ensure reproducibility for future experiments. Which approach should you use?
- A
Use Vertex AI Experiments to log training metadata, metrics, and artifacts for each run.
- B
Manually save model artifacts and metrics to a Google Cloud Storage bucket and compare them using a spreadsheet.
- C
Use BigQuery to store training metrics and write SQL queries to compare them.
- D
Leverage Vertex AI Pipelines exclusively for model comparison without additional tracking tools.
Show answer and explanation
Correct answer: A
Explanation
Vertex AI Experiments is purpose-built for tracking and comparing model artifacts, training runs, and evaluation metrics in an organized and efficient manner. It provides a centralized platform for experiment tracking, making it easier to identify the best-performing model and ensure reproducibility. Other approaches mentioned are either inefficient, incomplete, or not designed for this purpose.
- A. Correct.
Using Vertex AI Experiments is the correct approach as it is specifically designed for tracking, logging, and comparing training runs, including metrics and artifacts, ensuring reproducibility.
- B. Incorrect.
Manually saving data and comparing it using a spreadsheet is error-prone, inefficient, and does not provide reproducibility or robust version tracking.
- C. Incorrect.
While BigQuery can store metrics and allow comparisons, it does not natively manage artifacts or integrate with training workflows as efficiently as Vertex AI Experiments.
- D. Incorrect.
Vertex AI Pipelines is used to orchestrate machine learning workflows, but it does not natively provide a comprehensive solution for tracking and comparing model artifacts and metrics without additional tools like Vertex AI Experiments.