GCP Professional machine learning engineer practice test.

Step into the world of a Professional Machine Learning Engineer. With Google Cloud tech as your ally, you’ll master vast datasets and create impactful models. Collaboration is key as you ensure fairness and efficiency in every project. From architecture to deployment, your expertise shines, making ML accessible to all.

Curious to test your knowledge? Take our quiz and get well-explained answers to deepen your understanding. Join us and shape a future where intelligence knows no bounds.

Welcome to your professionalmachinelearningengineer

Q1.A company that provides an online platform for employers to post open positions is building a machine learning model to predict the expected salary for a particular job posting. The team working on the model is in the exploratory phase and plans to implement a deep neural network (DNN) using TensorFlow for their machine learning algorithm. The training data set includes about 100,000 examples. The team needs to choose the appropriate hardware and platform for training the model while balancing performance and cost. Which of the following Google AI services and configurations should the team choose for this project?

A.Use Google AI Platform Training with scale tier = BASIC_GPU (a single worker instance with a standard machine type + 1 GPU)

B.Use Google AI Platform Training with scale tier = BASIC (a single worker instance with a standard machine type)

C.Use Google AI Platform with scale tier = CUSTOM (a single worker instance with a high CPU machine type)

D.Use AutoML Tables with a standard machine type and a 6-hour training budget.

Q2.A team of machine learning engineers is working on a project to build a CI/CD automation pipeline for a model that serves online predictions to a mobile application. New training data is available in a Cloud Storage bucket anywhere from every 14 to 28 days. The team is using Kubeflow to implement the training pipeline and needs to decide how often they should trigger a new training job. Which of the following strategies should the team use to trigger the training pipeline?

A.Use Cloud Scheduler to set up a cron job to retrain the model every 14 days.

B.Use Cloud Monitoring to track the cumulative count of predictions

C.Use Pub/Sub and Cloud Functions to trigger the pipeline and retrain the model when a new training data file is added to Cloud Storage.

D.Use Cloud Monitoring to trigger the pipeline and retrain the model when a new training data file is added to Cloud Storage.

Q3.You recently joined a start-up that developed a mobile application to allow users to upload / share short videos. As part of the machine learning engineering team, you are working on a solution to moderate and analyze the uploaded videos for filtering out inappropriate or misinformative content. How would you design the solution to test the results with less time and effort?

A.Cloud Pub/Sub, Cloud Function, Cloud Vision API, Cloud Logging

B.Cloud Pub/Sub, Dataflow, Cloud Vision API, Cloud Logging

C.Cloud Pub/Sub, Cloud Function, Video Intelligence API, BigQuery

D.Cloud Pub/Sub, Cloud Function, AutoML Video, BigQuery

Q4.You are asked to develop a solution to optimize the drop-off routes for after school buses. The students use a drop-off schedule application to confirm their drop-off location ahead of time. What’s your best approach to develop the solution?

A.Use historical application data to train a regression model to predict how many students to pick-up at each location and then draw the route on the map. Skip the location if the predicted student count is zero.

B.Use the live data collected from the drop-off schedule application to calculate the number of students at each location and then draw the route on the map. Skip the location if the student count is zero.

C.Use the historical application data to train a classification model to predict if a pick-up location should be skipped or not and then draw the route on the map with all the non-skipped locations

D.Use the historical and live application data to train a time series forecast to predict the number of students at certain pick-up time and then draw the route on the map. Skip the location if the student count is zero.

Q5.What are distinct ways to create a dataset?

A.A data source constructs a Dataset from data stored in memory or in one or more files and a data transformation constructs a dataset from one or more tf.data.Dataset objects.

B. A data source constructs a Dataset from data stored in memory or in one or more files.

C.None of the options are correct.

D.A data transformation constructs a dataset from one or more tf.data.Dataset objects.

Q6.Your team has used Google’s Vertex AI Platform to train a product recommendation model. The product catalog data changes monthly and the user activity data updates daily. You need to develop a testing strategy to make sure the update from an older version of the model to a newer one is as smooth as possible and without the need to provision any new infrastructure manually. How do you deploy the model on Vertex AI Platform?

A.Create a deployment endpoint and deploy both the old and new models to the same endpoint with 80:20 traffic split

B.Deploy a model per endpoint and configure an API Gateway to load balance the traffic with 80:20 split

C.Create a deployment endpoint and deploy both the old and new models to the same endpoint with 50:50 traffic split

D.Deploy a model per endpoint and configure an API Gateway to load balance the traffic with 50:50 split

Q7.A team of data analysts at a human resources consulting firm is building a machine learning model to predict the number of days it will take to fill a job opening the company manages for its customers. The team has structured data for its training data set, but does not have a lot of experience choosing machine learning algorithms. Which of the following Google services should the team use to build this machine learning model?

A.AutoML Tables

B.BigQuery ML

C.AI Platform Training

D.AI Platform Notebooks

Q8.You are using a Google AI Platform Notebook to develop a new machine learning model. Your dataset is in BigQuery, and you plan to write some preprocessing code to prepare the data using Python before running your training experiments. You need to decide the best way to load the data. Which of the following approaches should you use to load your data from BigQuery into your notebook?

A.Use the BigQuery Storage API to load the data directly into a Pandas DataFrame.

B.Export the data from BigQuery to Cloud Storage, and read in the data from Cloud Storage.

C.Use BigQuery connector to provide read access from within the AI Platform Notebook.

D.Export data to DataFlow and read in the data from DataFlow

Q9.Which of the following statements is not a feature of Analytics Hub?

A.Analytics Hub requires batch data pipelines that extract data from databases, store it in flat files, and transmit them to the consumer where they are ingested into another database.

B.Analytics Hub efficiently and securely exchanges data analytics assets across organizations to address challenges of data reliability and cost.

C. There are three roles in Analytics Hub – A Data Publisher, Exchange Administrator, and a Data Subscriber.

D.You can create and access a curated library of internal and external assets, including unique datasets like Google Trends, backed by the power of BigQuery.

Q10.Your company has developed a tabular classification model using AutoML. You have been asked by senior management to explain how the model works. You need to select a feature attribution method to help with the model explanation. Which method should you choose?

A.Sampled Shapley

B.XRAI (Explanation with Ranked Area Integrals)

C.Integrated Gradients

D.AutoML Feature Engineering

Q11.The learning rate is a hyperparameter that controls how much to change the model in response to the estimated error each time the model weights are updated. Choosing the learning rate is challenging. What can happen if the value is too large?

A.The model will not train..

B.If the learning rate value is too large, then the model will converge.

C.Training may take a long time.

D.A large learning rate value may result in the model learning a sub-optimal set of weights too fast or an unstable training process.

Q12.You are helping to develop a call center solution using Google’s CCAI (Contact Center AI). The number of human agents is very limited. You have received phone audio files from the last 3 years and were asked to design a solution to reduce the workload for the human agents, but still help the customers with questions and queries as soon as possible. How do you design the lowest effort solution that both fulfills the requirement and is quickest to market?

A.Use Speech-to-Text to convert audio files into text data, train an AutoML model to classify the text data and then build a CCAI Q&A knowledge base to be used by the chatbot

B.Use Speed-to-Text to convert audio files into text data, train a custom model to classify the text data and then build a CCAI Q&A knowledge base to be used by chatbot

C.Develop a CCAI chatbot to answer the questions as much as possible, use AutoML NLP sentiment analysis to detect the customer sentiment to decide if the request needs to be transferred to human agents or not

D.Develop a CCAI chatbot to answer the questions as much as possible, use NLP API sentiment analysis to detect the customer sentiment to decide if the request needs to be transferred to human agents or not

Q13.A team of data analysts at a company that manufactures agricultural equipment is working on a project to build a machine learning model to predict the likelihood of a component failure based on the past maintenance history for a given product. The team is familiar with SQL but does not have extensive experience with programming languages like Python or Java. The team is looking for the appropriate Google Cloud services to help prepare the training data and train the machine learning model. Which set of Google Cloud services should the team choose for this project?

A.Use BigQuery for data preparation and BigQuery ML to create the model.

B.Use Dataflow for data preparation and BigQuery ML to create the model.

C.Use Dataflow for data preparation and AI Platform Training to create the model.

D.Use BigQuery for data preparation and AI Platform Training to create the model.

Q14.You are training a machine learning model to predict the starting salary for recent graduates from various degree programs throughout the country using a deep neural network (DNN). After doing some initial data analysis on the training data set, you notice that one numeric feature is distributed relatively uniformly but contains a few extreme outliers. You are concerned that this feature could affect your model's performance and stability and want to take steps to avoid this. What can you do to ensure this feature does not negatively impact the training stability and model performance?

A.During data preparation, normalize the feature using linear scaling.

B.During data preparation, normalize the feature using clipping.

C.During model training, use L2 regularization

D.During model training, use a reLU activation function on all layers.

Q15.Model complexity often refers to the number of features or terms included in a given predictive model. What happens when the complexity of the model increases?

A.All of the options are correct.

B.Model will not figure out general relationships in the data.

C.Model is more likely to overfit.

D.Model performance on a test set is going to be poor.

Q16.Your team is preparing a data set to train a new machine learning model. A junior data analyst is responsible for cleaning up and analyzing the dataset before moving into BigQuery and does not have much coding experience. You are looking for a service that would allow the analyst to visually inspect the data, identify problems, and quickly transform the data without writing code. You would also like the output of this process to be something that is easily repeatable. Which of these Google Cloud services should you choose for this data preparation step?

A.Dataprep by Trifacta

B.Cloud Data Fusion

C.Cloud Composer

D.Dataflow

Q17.Which of the following algorithms is useful, if you want to specify a quantity of trials that is greater than the number of points in the feasible space?

A.Grid Search

B.Random Search

C.Bayesian Optimization

D.Manual Search

Q18.You are working with a recruiting company to develop a ML model for filtering applicants’ resumes automatically. The goal is to select the right applicant to interview by matching applicant skills and experience. After the model is deployed and used for a while, the recruiters found that all the applicants selected for IT jobs are male and White or Asian. You are asked to make changes to the model to make sure that all applicants are selected regardless of gender and race. What is the approach to take for improving the model?

A.Re-train the model with resumes, including more resumes from female, nonwhite, and non-Asian applicants

B.Re-train the model without any gender or race feature

C.Re-train the model with a bigger, more balanced dataset, which includes all kinds of applicants’ resumes

D.Train the model to recognize complex patterns, which will auto-correct to account for any apparent biases

Q19.Your team is building an image classification model to detect motor vehicles in images using Google AI Platform training. You are using TensorFlow to build the model, and for training, you use the BASIC_TPU scale tier, which includes a standard master virtual machine and a worker Cloud TPU (Tensor Processing Units) with eight TPU v2 cores. For the first training run, you set the batch size to 10 and the number of iterations per loop to 500. The training time is much slower than you expected, and the output from the training run does not provide any obvious clues. Which of the following is most likely causing training to run slower than anticipated?

A.The batch size is too small.

B.The batch size is too large.

C.The number of iterations per loop is too big.

D.The number of iterations per loop is too small.

Q20.You are working on a project to build a continuous integration/continuous delivery (CI/CD) process using Kubeflow to automatically deploy a newly trained machine learning model to Google AI Platform Prediction. Two primary triggers in the process can result in the pipeline delivering a new model to production: (1) a developer updates model or pipeline code (2) new training data is available. You are responsible for testing the scenario when new training data triggers the process. Which of the following actions should you expect to occur when new training data is available?

A.The new training data triggers the training pipeline to run and train a new model with the updated training data. If the model evaluation passes, the system deploys a new model.

B.The new training data triggers the system to deploy a new training pipeline. The system runs the training pipeline to train a new model with the updated training data. If the model evaluation passes, the system deploys a new model.

C.The new training data triggers the system to build the training pipeline. The system runs the training pipeline to train a new model with the updated training data. If the model evaluation passes, the system deploys a new model.

D.The new training data triggers the training pipeline to run and train a new model with the updated training data. The system runs automated tests on the training pipeline components and if tests pass, the system deploys a new model.

Q21.You are in the process of developing a new binary classification machine learning model using a deep neural network and need a baseline to easily test the model's quality. Since this is a new model, there is not an existing model to compare to the new model. Which of the following should you choose to use as a baseline to access your models quality?

A.Compare your model output against a model that always predicts the most common label.

B.Compare your model with one that predicts each label 50% of the time.

C.Compare your model output against a model generated with just one epoch.

D.Compare your model output against a model trained with no regularization.

Q22.You are hired to help automate invoice processing for a company. You need to design a solution to process the invoices as quickly and accurately as possible, and the invoice data can also be verified by the workers manually if needed. The invoice data is uploaded to Google Cloud Storage every night. How should you design the solution in Google Cloud?

A.Convert invoices to OCR images, use Google Vision API to extract the text from OCR images

B.Convert invoices to OCR images, use Google AutoML Vision to train a model to the text from OCR images

C.Use Document AI’s Invoice Parser with Human in The Loop for data verification and relabeling with Process API endpoint

D.Use Document AI’s Invoice Parser with Human in The Loop for data verification and relabeling using batchProcess API endpoint

Q23.Your company is developing a product recommendation service for your eCommerce site. You have been provided the product catalog data and historical user event data. The company wants to use this service to increase the sales revenue. How do you load the data and design a recommendation system using Google Retail AI?

A.Import both product catalog and user event data in parallel into Retail AI, then train a recommendation model optimized for click-through rate.

B.Import both product catalog and user event data in parallel Retail AI, then train a recommendation model optimized for conversion rate.

C.Import product catalog first and then load user event data into Retail AI, then train a recommendation model optimized for click-through rate.

D.Import product catalog first and then load user event data into Retail AI, then train a recommendation model optimized for conversion rate.

Q24.A company that provides computer vision and robotics to agricultural machinery is working on a new product that identifies and precisely treats plant species in the field. You are working on training a machine learning model for this project with roughly 100,000 labeled images of plants. You are using Google AI platform and are taking advantage of TPU accelerators to improve training times. You want to make sure reading in the large volume of images does not create a bottleneck during training. Which of the following solutions can help reduce bottleneck issues with data ingestion?

A.Package the images in several TFRecord files and store them on Google Cloud Storage. Read the images into a tf.data.TFRecordDataset for training.

B.Store each image file on Google Cloud Storage and read in the images during training using Dataflow.

C.Store each image file on Google Cloud Storage and read the images into a tf.data.Dataset for training.

D.Package the images in several compressed files and store them on Cloud Storage. Read the images into a tf.data.Dataset for training.

Q25.You are working on a machine learning model that uses PyTorch and are migrating the project from on-premise to the Google Cloud AI Platform. Your team is also using a few dependencies that are not related to machine learning. Your job is to recommend what steps the team needs to take to run this work on Cloud AI Platform Training. Which of the following solutions should you recommend for this machine learning project to run on Cloud AI Platform Training?

A.Create a custom container with the required dependencies.

B.Use the standard runtime container and list required dependencies in a YAML file.

C.Migrate your project to TensorFlow and use the standard runtime container.

D.Use an existing pre-built PyTorch container.