Google Professional Machine Learning Engineer Question 360
Select 3Google Cloud PlatformYou are tasked with developing an end-to-end machine learning pipeline on Google Cloud to classify images into categories. The pipeline must handle data ingestion, preprocessing, model training, evaluation, and deployment. Which considerations are MOST critical to ensure the pipeline is scalable, maintainable, and fault-tolerant?
- A
Use Google Cloud Dataflow for distributed data preprocessing and transformation tasks.
- B
Train the model on a single local machine to simplify the development process.
- C
Use Vertex AI Pipelines to orchestrate and monitor the workflow stages.
- D
Manually deploy the trained model to a Compute Engine instance for prediction.
- E
Incorporate automated error handling and retry mechanisms for failed pipeline stages.
Show answer and explanation
Correct answers: A, C, E
Explanation
To build a scalable, maintainable, and fault-tolerant end-to-end ML pipeline, leveraging managed services like Google Cloud Dataflow for data preprocessing and Vertex AI Pipelines for orchestration is essential. These tools provide scalability and simplify the management of complex workflows. Additionally, incorporating automated error handling ensures reliability, especially in production-grade pipelines. Avoiding manual deployments and single-machine training helps maintain scalability and reduces operational overhead.
- A. Correct.
Google Cloud Dataflow is a managed service for distributed data processing and is well-suited for scalable and reliable preprocessing in an ML pipeline.
- B. Incorrect.
Training a model on a single local machine is not scalable and does not leverage Google Cloud's distributed training capabilities.
- C. Correct.
Vertex AI Pipelines is specifically designed for building and managing end-to-end ML pipelines, making it a critical component for scalability and maintainability.
- D. Incorrect.
Manually deploying the model to a Compute Engine instance is error-prone and not maintainable for production environments. Using Vertex AI Model Deployment or other managed services is a better practice.
- E. Correct.
Automated error handling and retry mechanisms help build a fault-tolerant pipeline, ensuring that transient failures do not disrupt the entire workflow.