PMLE practice questions 71–80

1. Question 71 · 364 in bank

   You are an ML engineer at a retail company. You have built a model that predicts a coupon to offer an ecommerce customer at checkout based on the items in their cart. When a customer goes to checkout, your serving pipeline, which is hosted on Google Cloud, joins the customer's existing cart with a row in a BigQuery table that contains the customers' historic purchase behavior and uses that as the model's input. The web team is reporting that your model is returning predictions too slowly to load the coupon offer with the rest of the web page. How should you speed up your model's predictions?

   • AAttach an NVIDIA P100 GPU to your deployed model’s instance.
   • BUse a low latency database for the customers’ historic purchase behavior.
   • CDeploy your model to more instances behind a load balancer to distribute traffic.
   • DCreate a materialized view in BigQuery with the necessary data for predictions.

   Show answer

   Answer: B

   Explanation

   I changed my mind. B: Im read a lot this page https://cloud.google.com/architecture/minimizing-predictive-serving-latency-in-machine-learning#online_real-time_prediction If the web team is reporting that the model is returning predictions too slowly to load the coupon offer with the rest of the web page, it suggests that the bottleneck might indeed be in the inference process rather than in data retrieval or processing. Given that the model is deployed on Google Cloud, choosing a low-latency database makes it suitable for scenarios where quick access to data is crucial, such as real-time predictions for web applications. Option D: While pre-aggregating data in BigQuery can improve query speed, it might not be as efficient as a low-latency database for frequently accessed data like customer purchase history.

   Source: Adapted from community discussion — guilhermebutzke

   References

   • Google Cloud documentation

   Next question

2. Question 72 · 364 in bank

   You work for a small company that has deployed an ML model with autoscaling on Vertex AI to serve online predictions in a production environment. The current model receives about 20 prediction requests per hour with an average response time of one second. You have retrained the same model on a new batch of data, and now you are canary testing it, sending ~10% of production traffic to the new model. During this canary test, you notice that prediction requests for your new model are taking between 30 and 180 seconds to complete. What should you do?

   • ASubmit a request to raise your project quota to ensure that multiple prediction services can run concurrently.
   • BTurn off auto-scaling for the online prediction service of your new model. Use manual scaling with one node always available.
   • CRemove your new model from the production environment. Compare the new model and existing model codes to identify the cause of the performance bottleneck.
   • DRemove your new model from the production environment. For a short trial period, send all incoming prediction requests to BigQuery. Request batch predictions from your new model, and then use the Data Labeling Service to validate your model’s performance before promoting it to production.

   Show answer

   Answer: B

   Explanation

   B can be effective in controlling resources available to the new model, ensuring that it is not delayed by the autoscaling trying to scale up from 0. Not A: there is no indication in the description that quota limits cause the slowdown and does not address issue where new model is performing poorly on canary testing. Not C : when you pull the new model from prod environment, you could affect end-user experience Not D: Same as C plus you rely on batch predictions which does not align with the need for online, real-time predictions in the prod environemnt. Data Labeling Service is more about assessing accuracy and less about resolving latency issues.

   Source: Adapted from community discussion — AnnaR

   References

   • Google Cloud documentation
   • Google Cloud documentation

   Next question

3. Question 73 · 364 in bank

   You want to train an AutoML model to predict house prices by using a small public dataset stored in BigQuery. You need to prepare the data and want to use the simplest, most efficient approach. What should you do?

   • AWrite a query that preprocesses the data by using BigQuery and creates a new table. Create a Vertex AI managed dataset with the new table as the data source.
   • BUse Dataflow to preprocess the data. Write the output in TFRecord format to a Cloud Storage bucket.
   • CWrite a query that preprocesses the data by using BigQuery. Export the query results as CSV files, and use those files to create a Vertex AI managed dataset.
   • DUse a Vertex AI Workbench notebook instance to preprocess the data by using the pandas library. Export the data as CSV files, and use those files to create a Vertex AI managed dataset.

   Show answer

   Answer: A

   Explanation

   A By writing a query that preprocesses the data using BigQuery and creating a new table, you can directly create a Vertex AI managed dataset with the new table as the data source. This approach is efficient because it leverages BigQuery’s powerful data processing capabilities and avoids the need to export data to another format or service. It also simplifies the process by keeping everything within the Google Cloud ecosystem. This makes it easier to manage and monitor your data and model training process.

   Source: Adapted from community discussion — 36bdc1e

   Next question

4. Question 74 · 364 in bank

   You developed a Vertex AI ML pipeline that consists of preprocessing and training steps and each set of steps runs on a separate custom Docker image. Your organization uses GitHub and GitHub Actions as CI/CD to run unit and integration tests. You need to automate the model retraining workflow so that it can be initiated both manually and when a new version of the code is merged in the main branch. You want to minimize the steps required to build the workflow while also allowing for maximum flexibility. How should you configure the CI/CD workflow?

   • ATrigger a Cloud Build workflow to run tests, build custom Docker images, push the images to Artifact Registry, and launch the pipeline in Vertex AI Pipelines.
   • BTrigger GitHub Actions to run the tests, launch a job on Cloud Run to build custom Docker images, push the images to Artifact Registry, and launch the pipeline in Vertex AI Pipelines.
   • CTrigger GitHub Actions to run the tests, build custom Docker images, push the images to Artifact Registry, and launch the pipeline in Vertex AI Pipelines.
   • DTrigger GitHub Actions to run the tests, launch a Cloud Build workflow to build custom Docker images, push the images to Artifact Registry, and launch the pipeline in Vertex AI Pipelines.

   Show answer

   Answer: D

   Explanation

   Choose D: GitHub Actions should be used to run tests and initiate the workflow upon code merges. Then, Cloud Build is a suitable service for building Docker images and handling the subsequent steps of pushing the images to Artifact Registry. So, Vertex AI Pipelines can be launched as part of the Cloud Build workflow for model retraining. In A Using Cloud Build directly from GitHub Actions would bypass GitHub Actions' capabilities for triggering and testing. In B, Cloud Run for building Docker images can introduce potential compatibility issues with Vertex AI Pipelines. In C, Skipping Cloud Build for image building limits the workflow's portability and integration with Vertex AI. https://cloud.google.com/vertex-ai/docs/pipelines/introduction https://medium.com/@cait.ray13/serving-ml-model-using-google-pub-sub-python-f569c46e7eb0

   Source: Adapted from community discussion — guilhermebutzke

   References

   • Google Cloud documentation
   • Google Cloud documentation
   • Google Cloud documentation

   Next question

5. Question 75 · 364 in bank

   You are working with a dataset that contains customer transactions. You need to build an ML model to predict customer purchase behavior. You plan to develop the model in BigQuery ML, and export it to Cloud Storage for online prediction. You notice that the input data contains a few categorical features, including product category and payment method. You want to deploy the model as quickly as possible. What should you do?

   • AUse the TRANSFORM clause with the ML.ONE_HOT_ENCODER function on the categorical features at model creation and select the categorical and non-categorical features.
   • BUse the ML.ONE_HOT_ENCODER function on the categorical features and select the encoded categorical features and non-categorical features as inputs to create your model.
   • CUse the CREATE MODEL statement and select the categorical and non-categorical features.
   • DUse the ML.MULTI_HOT_ENCODER function on the categorical features, and select the encoded categorical features and non-categorical features as inputs to create your model.

   Show answer

   Answer: B

   Explanation

   When the TRANSFORM clause is present, only output columns from the TRANSFORM clause are used in training. Any results from query_statement that don't appear in the TRANSFORM clause are ignored. https://cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-create#transform so if you want TRANSFORM then use TRANSFORM for both categorical and non-categorical features

   Source: Adapted from community discussion — BlehMaks

   References

   • Google Cloud documentation
   • Google Cloud documentation

   Next question

6. Question 76 · 364 in bank

   You need to develop an image classification model by using a large dataset that contains labeled images in a Cloud Storage bucket. What should you do?

   • AUse Vertex AI Pipelines with the Kubeflow Pipelines SDK to create a pipeline that reads the images from Cloud Storage and trains the model.
   • BUse Vertex AI Pipelines with TensorFlow Extended (TFX) to create a pipeline that reads the images from Cloud Storage and trains the model.
   • CImport the labeled images as a managed dataset in Vertex AI and use AutoML to train the model.
   • DConvert the image dataset to a tabular format using Dataflow Load the data into BigQuery and use BigQuery ML to train the model.

   Show answer

   Answer: C

   Explanation

   Which SDK use? • If you use TensorFlow in an ML workflow that processes terabytes of structured data or text data -> TFX • For other use-cases -> KFP

   Source: Adapted from community discussion — pinimichele01

   References

   • Google Cloud documentation

   Next question

7. Question 77 · 364 in bank

   You are developing a model to detect fraudulent credit card transactions. You need to prioritize detection, because missing even one fraudulent transaction could severely impact the credit card holder. You used AutoML to tram a model on users' profile information and credit card transaction data After training the initial model, you notice that the model is failing to detect many fraudulent transactions. How should you adjust the training parameters in AutoML to improve model performance? (Choose two.)

   • AIncrease the score threshold
   • BDecrease the score threshold.
   • CAdd more positive examples to the training set
   • DAdd more negative examples to the training set
   • EReduce the maximum number of node hours for training

   Show answer

   Answer: B

   Explanation

   B & D D. Add more negative examples to the training set: Fraudulent transactions are typically a minority compared to legitimate transactions. By increasing the number of negative examples (fraudulent transactions) in your training data, you provide AutoML with more information about the patterns of fraudulent activity. This can help the model better distinguish between legitimate and fraudulent transactions. B. Decrease the score threshold: The score threshold determines the level of suspicion assigned to a transaction by the model. A lower threshold means the model flags more transactions as suspicious, potentially catching more fraudulent activities. However, this might also lead to an increase in false positives (flagging legitimate transactions). You'll need to find a balance between fraud detection and acceptable false positive rates based on your business needs.

   Source: Adapted from community discussion — fitri001

   Next question

8. Question 78 · 364 in bank

   You are building a real-time prediction engine that streams files which may contain Personally Identifiable Information (PII) to Google Cloud. You want to use the Cloud Data Loss Prevention (DLP) API to scan the files. How should you ensure that the PII is not accessible by unauthorized individuals?

   • AStream all files to Google Cloud, and then write the data to BigQuery. Periodically conduct a bulk scan of the table using the DLP API.
   • BStream all files to Google Cloud, and write batches of the data to BigQuery. While the data is being written to BigQuery, conduct a bulk scan of the data using the DLP API.
   • CCreate two buckets of data: Sensitive and Non-sensitive. Write all data to the Non-sensitive bucket. Periodically conduct a bulk scan of that bucket using the DLP API, and move the sensitive data to the Sensitive bucket.
   • DCreate three buckets of data: Quarantine, Sensitive, and Non-sensitive. Write all data to the Quarantine bucket. Periodically conduct a bulk scan of that bucket using the DLP API, and move the data to either the Sensitive or Non-Sensitive bucket.

   Show answer

   Answer: D

   Explanation

   I would say D because if the data is considered non sensitive there is no reason to recheck if it's sensitive so you move to a non sensitive bucket. Data that is sensitive will be directed to the Sensitive bucket that has limited access just like the Quarantine bucket that should have limited access policy.

   Source: Adapted from community discussion — mousseUwU

   References

   • Google Cloud documentation

   Next question

9. Question 79 · 364 in bank

   You need to deploy a scikit-leam classification model to production. The model must be able to serve requests 24/7, and you expect millions of requests per second to the production application from 8 am to 7 pm. You need to minimize the cost of deployment. What should you do?

   • ADeploy an online Vertex AI prediction endpoint. Set the max replica count to 1
   • BDeploy an online Vertex AI prediction endpoint. Set the max replica count to 100
   • CDeploy an online Vertex AI prediction endpoint with one GPU per replica. Set the max replica count to 1
   • DDeploy an online Vertex AI prediction endpoint with one GPU per replica. Set the max replica count to 100

   Show answer

   Answer: B

   Explanation

   B. Deploy an online Vertex AI prediction endpoint. Set the max replica count to 100: This option provides a higher number of replicas (100) to handle the expected high volume of requests during peak hours. While it might result in increased costs, it provides the necessary scalability to manage the incoming traffic efficiently. During non-peak hours, you can consider scaling down the replicas to reduce costs, as Vertex AI allows dynamic scaling based on demand.

   Source: Adapted from community discussion — pikachu007

   Next question

10. Question 80 · 364 in bank

    You work with a team of researchers to develop state-of-the-art algorithms for financial analysis. Your team develops and debugs complex models in TensorFlow. You want to maintain the ease of debugging while also reducing the model training time. How should you set up your training environment?

    • AConfigure a v3-8 TPU VM. SSH into the VM to train and debug the model.
    • BConfigure a v3-8 TPU node. Use Cloud Shell to SSH into the Host VM to train and debug the model.
    • CConfigure a n1 -standard-4 VM with 4 NVIDIA P100 GPUs. SSH into the VM and use ParameterServerStraregv to train the model.
    • DConfigure a n1-standard-4 VM with 4 NVIDIA P100 GPUs. SSH into the VM and use MultiWorkerMirroredStrategy to train the model.

    Show answer

    Answer: D

    Explanation

    Given the need to balance ease of debugging and reduce training time for complex models in TensorFlow, option D - "Configure an n1-standard-4 VM with 4 NVIDIA P100 GPUs. SSH into the VM and use MultiWorkerMirroredStrategy to train the model" appears to be more suitable. This setup utilizes NVIDIA P100 GPUs for computational power and employs MultiWorkerMirroredStrategy, which can distribute the workload across GPUs efficiently, potentially reducing training time while maintaining a relatively straightforward environment for debugging.

    Source: Adapted from community discussion — pikachu007

    References

    • Google Cloud documentation

    Next question

Keyboard:J Kmove between questionsA–Dselect an answerRshow or hide the answer← →change page

PMLE practice, question 71. 364 questions available.