Google Professional Machine Learning Engineer
Get started today
Ultimate access to all questions.
Ultimate access to all questions.
You are a machine learning engineer working at a bank. Your bank has obtained a custom tabular machine learning model from a third-party vendor. Due to the sensitive nature of the training data used, it is not available to you. The model is packaged in a Vertex AI Model serving container, designed to accept a string as input for each prediction instance, with the feature values separated by commas. Your task is to deploy this model for online predictions and ensure that you can monitor the feature distribution over time to detect any issues. Given that you need to achieve this with minimal effort, what should you do?
Explanation:
Option A is the correct answer. Since the training data is not available due to its sensitivity, feature skew detection is not applicable because it requires access to the training data distribution to compare against the incoming data distribution. Therefore, feature drift detection becomes the appropriate objective for monitoring the feature distribution over time. Additionally, Option A avoids unnecessary refactoring of the serving container, aligning with the requirement of achieving the task with minimal effort.