HikeLetter | PMP Training & Certification Online Course | Project Management Learning

1

For a customer-risk scoring service, a production model's offline metrics are strong, but online business outcomes are poor. What should the team investigate first?

A. Differences between offline evaluation and production data, labels, and serving behavior B. The notebook's font size C. Whether the model name is short D. Whether the dashboard uses a dark theme

2

When reviewing a model release, a production model uses sensitive features. What should be addressed before launch?

A. Data minimization, access control, approved use, and applicable privacy/governance requirements B. Publish raw features in a public notebook C. Grant all workspace users access to the feature table D. Assume model deployment removes data obligations

3

In a governed enterprise ML environment, a registry workflow requires review before a candidate model becomes production. What should be configured?

A. A controlled promotion process with permissions and approval criteria B. Automatic promotion of every new run C. Shared credentials for all team members D. Manual overwriting of production artifacts without versioning

4

When monitoring a deployed model, a model endpoint receives requests with feature values outside the training range. What is the most robust response?

A. Monitor input drift and define validation, fallback, or retraining procedures appropriate to the risk B. Assume predictions remain reliable because the endpoint is online C. Disable request logging and monitoring D. Automatically grant callers model-registry admin access

5

In a regulated analytics use case, a team needs to audit which model version produced a prediction. What design helps?

A. Record a model version or immutable identifier with prediction metadata, subject to privacy and retention policies B. Record only the endpoint's friendly display name C. Discard all request and version metadata D. Use the current date as the model identifier

6

For a customer-risk scoring service, a team wants to detect degradation in a live model whose labels arrive weeks later. What approach is useful in the interim?

A. Monitor data quality, input drift, prediction distributions, and service metrics until outcome labels arrive B. Wait silently and collect no operational metrics C. Treat prediction confidence as proof of correctness D. Delete the training dataset

7

When reviewing a model release, a feature pipeline is recomputed differently in training and online inference. What risk does this create?

A. Training-serving skew B. Guaranteed improved fairness C. Lower latency in every case D. Automatic prevention of data leakage

8

In a governed enterprise ML environment, a production model's offline metrics are strong, but online business outcomes are poor. What should the team investigate first?

A. Differences between offline evaluation and production data, labels, and serving behavior B. The notebook's font size C. Whether the model name is short D. Whether the dashboard uses a dark theme

9

When monitoring a deployed model, a production model uses sensitive features. What should be addressed before launch?

A. Data minimization, access control, approved use, and applicable privacy/governance requirements B. Publish raw features in a public notebook C. Grant all workspace users access to the feature table D. Assume model deployment removes data obligations

10

In a regulated analytics use case, a registry workflow requires review before a candidate model becomes production. What should be configured?

A. A controlled promotion process with permissions and approval criteria B. Automatic promotion of every new run C. Shared credentials for all team members D. Manual overwriting of production artifacts without versioning

Databricks Certified Machine Learning Engineer Professional

Overview

Course Description

Course Questions