Google Professional Machine Learning Engineer Question 380
Select 3Google Cloud PlatformYou are tasked with setting up a CI/CD pipeline to automate the deployment of machine learning models on Google Cloud. The pipeline requires steps to preprocess data, train the model, and deploy it to an endpoint. You decide to use Cloud Build for automation, but you need to configure the pipeline with appropriate triggers and compute resources. Which components should you configure as part of this setup?
- A
Define a Cloud Build trigger that activates when new training data is uploaded to Cloud Storage.
- B
Specify the machine type and compute resources for the training job in the Cloud Build configuration.
- C
Use Cloud Build for deploying the trained model directly to a Vertex AI endpoint.
- D
Set up a Cloud Scheduler job to periodically trigger the pipeline regardless of data changes.
- E
Integrate Cloud Build with BigQuery to preprocess the training data within the pipeline.
Show answer and explanation
Correct answers: A, B, E
Explanation
To set up a CI/CD pipeline for ML models using Cloud Build, you need to configure triggers to start the pipeline (e.g., when new data is uploaded), define compute resources for the training step, and ensure preprocessing is integrated into the pipeline. Cloud Build triggers and BigQuery integration can help achieve these requirements, while deploying models is better handled by Vertex AI Deployment.
- A. Correct.
Defining a Cloud Build trigger based on a new data upload to Cloud Storage is a valid approach to automate the pipeline's start when new data becomes available.
- B. Correct.
Specifying machine types and compute resources for the training job ensures that the pipeline has the necessary compute power to execute the training step within Cloud Build.
- C. Incorrect.
While Cloud Build can automate many steps, deploying models to Vertex AI endpoints is typically handled by Vertex AI Deployment, not Cloud Build directly.
- D. Incorrect.
A Cloud Scheduler job can trigger pipelines on a schedule, but it does not address event-driven triggers based on data changes, which is a more efficient option in this scenario.
- E. Correct.
Integrating Cloud Build with BigQuery enables preprocessing steps to occur within the pipeline, leveraging BigQuery's capabilities for querying and preparing training data.