Google Professional Machine Learning Engineer Question 422
Select 4Google Cloud PlatformYou are designing a CI/CD pipeline to deploy a machine learning model to Google Cloud. The model is trained using TensorFlow and is intended to serve predictions using Vertex AI Prediction. Which of the following steps should you include in the pipeline to ensure a reliable and automated deployment process?
- A
Package the trained model into a format compatible with TensorFlow SavedModel.
- B
Use Cloud Build to automate the creation of a container image for the model.
- C
Directly deploy the raw TensorFlow model to Vertex AI Prediction without testing.
- D
Integrate unit tests and data validation checks before deploying the model.
- E
Use Vertex AI Model Monitoring to set up automated alerts for model performance.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
To implement a reliable CI/CD pipeline for model deployment on Google Cloud, the process must include packaging the model in the correct format, automating container image creation, and ensuring quality through testing and validation. Additionally, monitoring the model after deployment is crucial for maintaining performance and identifying issues proactively. Skipping testing or deploying raw models directly would compromise the reliability of the deployment process.
- A. Correct.
Correct: TensorFlow SavedModel is a standard format for deploying TensorFlow models and is required for compatibility with Vertex AI Prediction.
- B. Correct.
Correct: Cloud Build is commonly used in CI/CD pipelines for automating container image creation, which is necessary for deploying models to Vertex AI.
- C. Incorrect.
Incorrect: Deploying the raw TensorFlow model directly without testing can lead to errors or unreliable predictions. Proper testing and validation are critical.
- D. Correct.
Correct: Unit tests and data validation checks are essential to ensure the model and its input data are robust before deployment.
- E. Correct.
Correct: Vertex AI Model Monitoring allows you to monitor model performance post-deployment, which is a key component of a reliable CI/CD process.