Google Professional Machine Learning Engineer Question 138
Select 2Google Cloud PlatformYou are tasked with building a machine learning pipeline in Google Cloud that ingests large volumes of text documents from a Cloud Storage bucket into Vertex AI for real-time inference. The model expects preprocessed input in JSON format. Which steps should you include to correctly configure the pipeline?
- A
Use Cloud Storage triggers to automatically preprocess the text documents and save them in a JSON format.
- B
Set up a Dataflow pipeline to preprocess the text documents and output them to a Cloud Storage bucket in JSON format.
- C
Configure a Vertex AI Endpoint to directly read text documents from the Cloud Storage bucket and preprocess them dynamically during inference.
- D
Deploy a custom preprocessor as a separate service and integrate it with the Vertex AI Endpoint to handle preprocessing before inference.
- E
Use Vertex AI's built-in preprocessing capabilities to convert text documents into JSON format automatically.
Show answer and explanation
Correct answers: B, D
Explanation
To ingest and preprocess text documents for inference in Vertex AI, it is essential to preprocess the raw data into the required JSON format. Dataflow is a scalable option for preprocessing large data volumes efficiently. Alternatively, deploying a custom preprocessing service provides flexibility and ensures the input adheres to the model's requirements. Vertex AI itself does not provide built-in preprocessing for raw text data, and Cloud Storage triggers require additional services for actual processing.
- A. Incorrect.
Cloud Storage triggers cannot perform preprocessing directly. While they can initiate workflows, you would still need an external service like Cloud Functions to handle the actual preprocessing.
- B. Correct.
Dataflow is an appropriate tool for preprocessing large-scale data, including text documents, and can output the processed data in the required JSON format.
- C. Incorrect.
Vertex AI Endpoints are not designed to directly preprocess raw input data such as text documents. Preprocessing should be handled before sending data for inference.
- D. Correct.
Deploying a custom preprocessor as a service allows for flexibility in preprocessing and ensures that the data is in the correct format before it is sent to the Vertex AI Endpoint for inference.
- E. Incorrect.
Vertex AI does not provide built-in preprocessing capabilities to convert raw text documents into JSON format. Preprocessing must be explicitly implemented outside of Vertex AI.