Official Google Cloud Certified Professional Machine Learning Engineer Study Guide

AppendixAnswers to Review Questions

Chapter 1: Framing ML Problems

1. A, B, D. First understand the use case, and then look for the details such as impact, success criteria, and budget and time frames. Finding the algorithm comes later.
2. B. Hyperparameters are variables that cannot be learned. You will use a hyperparameter optimization (HPO) algorithm to automatically find the best hyperparameters. This is not considered when you are trying to match a business case to an ML problem.
3. B. The input data is time‐series data and predicting for next 7 days is typical of a forecasting problem.
4. B. A prediction has only two outputs: either valid or not valid. This is binary classification. If there are more than two classes, it is multiclass classification. Linear regression is predicting a number. Option C is popular with support tickets to identify clusters of topics but cannot be used in this case.
5. C. When you are trying to identify an object across several frames, this is video object tracking. Option A is factually incorrect. Option B is for images, not video. Scene detection or action detection classifies whether an “action” has taken place in video, a different type of problem, so option D is also wrong.
6. D. Topic modeling is an unsupervised ML problem. Given a set of documents, it would cluster them into groups and also provide the keywords that define each cluster.
7. C. Precision is a metric for unbalanced classification problems.
8. A. The root‐mean‐squared error (RMSE) is the best ...