Google Professional Machine Learning Engineer Question 195
Select 2Google Cloud PlatformYou are working on a machine learning project to build a model for predicting customer churn. You want to test different hyperparameter configurations, preprocessing steps, and model architectures to determine the optimal setup. You also need to ensure proper version control of your experiments for reproducibility. Which Google Cloud tools and features should you use to achieve this?
- A
Vertex AI Experiments to track and compare different training runs, including their configurations and performance metrics.
- B
Kubeflow Pipelines to orchestrate complex workflows and automate preprocessing, training, and evaluation steps.
- C
Vertex AI Feature Store to store and serve features for real-time predictions during the experimentation phase.
- D
BigQuery ML to directly train machine learning models on structured data without needing a custom pipeline.
- E
Vertex AI Model Registry to store trained models and manage their deployment lifecycle.
Show answer and explanation
Correct answers: A, B
Explanation
To effectively experiment with different configurations and maintain reproducibility in machine learning workflows, Vertex AI Experiments is essential for tracking and comparing runs, while Kubeflow Pipelines enables automation of complex workflows. These tools together provide a robust framework for managing experimentation. The other options, while useful in other contexts, do not specifically address the needs of experimentation and version control in this scenario.
- A. Correct.
Vertex AI Experiments is specifically designed to track, organize, and compare machine learning experiments, making it ideal for managing configurations and ensuring reproducibility.
- B. Correct.
Kubeflow Pipelines is a powerful tool for designing and automating end-to-end machine learning workflows, which is crucial for experimenting with different preprocessing steps and model architectures.
- C. Incorrect.
Vertex AI Feature Store is useful for managing features across teams and serving them in production, but it is not directly relevant to managing or tracking experimentation workflows.
- D. Incorrect.
BigQuery ML is useful for training models on structured data, but it does not provide the level of control or flexibility needed for experimentation with preprocessing or hyperparameter tuning.
- E. Incorrect.
Vertex AI Model Registry is designed for managing and deploying trained models, not for managing experimentation processes.