How to monitor feature distribution without training data?

PMLE By islever Editorial Last updated Aug 31, 2026

Short answer

A 1. Upload the model to Vertex AI Model Registry, and deploy the model to a Vertex AI endpoint

Question

You work at a bank. You have a custom tabular ML model that was provided by the bank’s vendor. The training data is not available due to its sensitivity. The model is packaged as a Vertex AI Model serving container, which accepts a string as input for each prediction instance. In each string, the feature values are separated by commas. You want to deploy this model to production for online predictions and monitor the feature distribution over time with minimal effort. What should you do?

1. Upload the model to Vertex AI Model Registry, and deploy the model to a Vertex AI endpoint
Correct Answer
1. Upload the model to Vertex AI Model Registry, and deploy the model to a Vertex AI endpoint
1. Refactor the serving container to accept key-value pairs as input format
1. Refactor the serving container to accept key-value pairs as input format

Key Takeaways

This question tests the distinction between Training-Serving Skew and Feature Drift, specifically that Skew requires training data, while trapping candidates into unnecessary container refactoring.

To monitor feature distribution without training data, use Vertex AI Feature Drift monitoring instead of Training-Serving Skew. The community consensus confirms that refactoring the container is unnecessary and violates the minimal effort constraint.

Common mistake: Selecting Training-Serving Skew monitoring is a common error because candidates often overlook that this metric requires the original training dataset to be available.

Expert Analysis

Why the Answer Is Correct

Option A (Feature Drift) is the correct choice because the question explicitly states that training data is not available. Feature Drift monitors changes in the distribution of production data over time, which does not require access to the original training set. Furthermore, deploying the model as-is satisfies the "minimal effort" requirement, as Vertex AI Model Monitoring can handle the input format without needing code changes.

Why the Other Options Are Wrong

Options suggesting Training-Serving Skew monitoring (Option B) are incorrect because this specific type of monitoring calculates the divergence between production data and the training data. Since the training data is unavailable due to sensitivity, this calculation is impossible. Options involving refactoring the container (Options C and D) are incorrect because the requirement is for "minimal effort"; rewriting the container to accept key-value pairs is unnecessary when the existing container supports predictions and monitoring.

Community Comment Notes

Community comments emphasize the critical constraint that "Training data not available -> can't be skew, so it must be drift," effectively ruling out skew detection. Additionally, users noted that Vertex AI Model Monitoring is capable of parsing comma-separated feature values, which validates that refactoring the serving container is not required to meet the monitoring goals.

Official Reference

https://cloud.google.com/vertex-ai/docs/model-monitoring/overview

This link opens the official vendor documentation in a new tab.

Exam Strategy

Always verify data availability before selecting a monitoring type; if training data is missing, you must choose Feature Drift over Training-Serving Skew. Additionally, strictly adhere to constraints like "minimal effort" to immediately eliminate options involving code refactoring.

Community Votes

100%

100% of anonymous learners picked answer A. Votes are pick records left by other test-takers — they are not the verified answer.

Discussion 4 contributions

Compiled from public community threads and reader submissions, lightly edited for clarity. Handles and vote tallies are shown as recorded, and comments reflect the views of individual contributors.

From public community threads 4

b1a8fae 👍 5 Answered A

A. Minimum effort -> ditch refactoring (hopefully not needed) Training data not available -> can't be skew, so it must be drift
pinimichele01 👍 3 Answered A

Training data not available -> can't be skew, so it must be drift
pikachu007 👍 1 Answered A

Handles string input format: Vertex AI Model Monitoring can parse comma-separated feature values, avoiding the need to refactor the serving container. It directly monitors feature distribution over time, aligning with the goal of detecting potential drifts.
CHARLIE2108 👍 1

I have a doubt, could someone please help with this? While "drift" (Option A) might imply gradual changes, "skew" (Option B) is more suitable for sudden shifts in feature distributions, potentially relevant for sensitive data. Is option B better than A?

Reader comments & corrections

No reader corrections yet — spotted an error or have a note? Share it below.

Practice All PMLE Questions

Access 65 questions with complete answers and detailed explanations.

View Full PMLE Practice Test →

← Back to PMLE Study Guide