8 of the 22 questions in this set, with the correct answer marked and every option explained.
1. You work on a team where the process for deploying a model into production starts with data scientists training different versions of models in a Kubeflow pipeline. The workflow then stores the new model artifact into the corresponding Cloud Storage bucket. You need to build the next steps of the pipeline after the submitted model is ready to be tested and deployed in production on AI Platform. How should you configure the architecture before deploying the model to production?
✓Deploy model in test environment -> Validate model -> Create a new AI Platform model version
the model can be validated after it is deployed to the test environment, and the release version is established before the model is deployed in production
✗Create a new AI Platform model version -> Validate model -> Deploy model in test environment
Registers the production version before anything has been exercised in test. The promotion path runs backwards, so the gate has nothing to gate.
✗Validate model -> Deploy model in test environment -> Create a new AI Platform model version
Validation has to happen against something running. Placing it first leaves nothing to validate against.
✗Create a new AI Platform model version - > Deploy model in test environment -> Validate model
Creates the production version first and only tests afterwards. An unverified model would already exist in the registry before anyone confirmed it works.
2. You work for a manufacturing company that owns a high-value machine which has several machine settings and multiple sensors. A history of the machine’s hourly sensor readings and known failure event data are stored in BigQuery. You need to predict if the machine will fail within the next 3 days in order to schedule maintenance before the machine fails. Which data preparation and model training steps should you take?
✗Data preparation: Daily min value feature engineering with DataPrep; Model training: Logistic regression with BQML and AUTO_CLASS_WEIGHTS set to True
Reducing a day of hourly readings to one minimum throws away the trajectory. Gradual degradation shows up as a trend across many readings, which a rolling average preserves.
✗Data preparation: Rolling average feature engineering with DataPrep; Model training: Logistic regression with BQML and AUTO_CLASS_WEIGHTS set to False
Gets the features right and then undoes the benefit. Failures are rare events, so without automatic class weighting the model can predict 'no failure' every time and still score well.
✗Data preparation: Daily max value feature engineering with DataPrep; Model training: AutoML classification with BQML
A single daily peak loses the same trend information as the daily minimum, and it discards the class-imbalance handling the scenario needs.
✓Data preparation: Rolling average feature engineering with DataPrep; Model training: Logistic regression with BQML and AUTO_CLASS_WEIGHTS set to True
D is correct because it uses the rolling average of the sensor data and balances the weights using the BQML auto class weight balance parameter.
3. You are an ML engineer at a media company. You need to build an ML model to analyze video content frame-by-frame, identify objects, and alert users if there is inappropriate content. Which Google Cloud products should you use to build this project?
✗Pub/Sub, Cloud Function, Cloud Vision API
The Vision API works on still images. Analysing footage frame by frame, with temporal context, is what Video Intelligence provides.
✗Pub/Sub, Cloud Function, AutoML Video Intelligence, Cloud Logging
AutoML is for labels you define and train yourself. Explicit-content detection already ships in the pre-built API, so custom training is effort spent on a solved problem.
✓Pub/Sub, Cloud Function, Video Intelligence API, Cloud Logging
C is correct as Video Intelligence API can find inappropriate components and other components satisfy the requirements of real-time processing and notification
Two mismatches: Cloud IoT handles device telemetry rather than video ingestion, and Vision API again cannot process video.
4. You need to create a model that predicts stock prices given a variety of factors. What type of problem are we modeling for?
✗Clustering
An unsupervised method that groups records by similarity. There is no target to predict, whereas here the price is known and being modelled.
✗Classification
Assigns an input to one of a fixed set of categories. A share price is a continuous quantity with no discrete classes.
✗Linear
Names a family of model, not a category of problem. Linear regression is one technique for solving a regression problem.
✓Regression
Regression is the supervised learning task for modeling and predicting continuous, numeric variables. Examples include predicting real-estate prices, stock price movements, or student test scores.
5. What are wide neural networks good for, compared deep neural networks?
✓Wide neural networks are better for memorization.
Deep networks are better at generalization.
✗Wide networks are an older style neural network.
Width and depth are design choices rather than eras, and wide-and-deep architectures deliberately combine both to get memorisation and generalisation together.
✗Wide neural networks are better for generalization.
Generalisation is what depth buys. Wide layers excel at recording specific feature combinations they have seen.
✗Wide networks are better at more complex problems.
Hierarchical complexity favours depth, where each layer builds on features learned below it.
6. You are developing an application that will process thousands of images and scan for explicit content. You need to develop your learning model quickly, and are not familiar with working in Tensorflow. How can you complete this task as quickly as possible while saving on costs?
✗Develop your training model in Cloud ML Engine, then have your application call on the trained model.
Building a custom vision model means gathering labelled data, training and tuning — slower and more expensive than calling an API that already detects explicit content. (Cloud ML Engine is now part of Vertex AI.)
✗Hire a consultant to assist with training your machine learning model to process your images.
Adds cost and delay while still leaving you with a bespoke model to run and maintain.
✗Use Cloud Datalab to develop your learning model in an interactive format, to make the process easier.
A pleasant development environment does not remove the development. You would still be building the model yourself. (Datalab is deprecated; Vertex AI Workbench succeeded it.)
✓Have your application call on the Cloud Vision API and pass your images to it via a Cloud Storage URI.
The Cloud Vision API is able to scan for explicit images, making this the quickest and most efficient answer.
7. You need to give a team member the ability to use a training model for predictions, but not have the ability to create or delete models. What IAM role should you assign to achieve this task with the minimum necessary permissions?
✓Model User
The Model User role allows a user to read models and use them for prediction, but nothing else.
✗Project Editor
A sweeping project-wide role covering almost every resource. Granting it for prediction access alone is a clear least-privilege violation.
✗Cloud ML Engine Developer
Includes the ability to create and train models — more than the task calls for when the requirement is minimum necessary permissions.
✗Model Owner
Carries full control including deletion, which is precisely the capability being withheld.
8. What is a difference between example (training) data and test data?
✓Example (training) data uses its labels to train the machine learning model, while test data uses its labels to validate the model’s accuracy.
Examples have a correct label attached to them, which an ML model uses to learn how to associate features with the label. While test data also has labels, it is used to validate accuracy of the model, but not to train it.
✗Examples are used for feature engineering, and test data is used for hyperparameter tuning.
Both halves misassign the role. Feature engineering applies across the dataset, and hyperparameter tuning belongs to a validation split — using the test set for it contaminates your final estimate.
✗Examples are used to validate training models, and test data is used to train the learning model.
The two purposes are reversed. Training on the held-out set destroys the only unbiased measurement you have.
✗Test data have labels attached and examples do not.
In supervised learning both are labelled. What differs is how the labels are used — to fit the model, or to score it.
14 more questions in the app
Practise the full 22-question set with a timer, scoring and progress tracking.
