PMLE practice questions 341–350

1. Question 341 · 364 in bank

   You are the lead ML engineer for a large ecommerce company. The Data Analytics team has prepared a dataset of 50 million transaction rows, which is stored in a partitioned BigQuery table. The team consists primarily of SQL experts with limited experience in Python or TensorFlow. You need to build a binary classification model to predict the likelihood of a customer making a purchase in the next seven days. You need to provide daily batch predictions. You need to minimize development effort, data movement, and the overall complexity of the solution. What should you do?

   • ACreate a model with BigQuery ML that includes the MODEL_TYPE=AUTOML_CLASSIFIER settings. Use the ML.PREDICT function to generate predictions as part of your nightly data warehouse jobs.
   • BUse the BigQuery ML TRANSFORM function to normalize and scale numerical features, one-hot encode categorical features, and train a MODEL_TYPE=LOGISTIC_REG mode. Use the ML.PREDICT function to generate predictions as part of your nightly data warehouse jobs.
   • CTrain a classification model with Agent Platform AutoML Tabular Workflows, deploy the trained model to an Agent Platform endpoint, and add a nightly Cloud Run job to trigger batch predictions.
   • DDevelop an XGBoost model on an Agent Platform Workbench notebook. Load the data into a pandas DataFrame, and pickle it to a Cloud Storage bucket. Create another notebook that loads the pickled model from Cloud Storage and saves predictions back to Cloud Storage. Schedule this notebook for nightly execution.

   Show answer

   Answer: A

   Next question

2. Question 342 · 364 in bank

   You have deployed a deep learning model to an Agent Platform endpoint using a machine type with NVIDIA GPUs. You initially configured the endpoint to autoscale based on a target CPU utilization of 60%. During a load test, you observe that prediction latency increases significantly as traffic rises, but the number of replicas remains constant. Cloud Monitoring indicates that CPU utilization stays below 40%, while the GPU utilization consistently exceeds 90%. You need to ensure that the endpoint scales efficiently to handle the increased load. What should you do?

   • AUpdate the autoscaling configuration to scale based on the GPU duty cycle metric.
   • BUpdate the autoscaling configuration to decrease the target CPU utilization to 20%.
   • CRedeploy the model to a machine type that includes a TPU.
   • DIncrease the minimum replica count for the endpoint to match the peak load that was observed during testing.

   Show answer

   Answer: A

   Next question

3. Question 343 · 364 in bank

   Your company manages an ecommerce platform and has a large dataset of customer reviews. Each review has a positive, negative, or neutral label. You need to quickly prototype a sentiment analysis model that accurately predicts the sentiment labels of new customer reviews while minimizing time and cost. What should you do?

   • AUse the Agent Platform Text Embeddings API to vectorize the text, and train a regression model by using Agent Platform AutoML to predict sentiment scores.
   • BUse Agent Platform AutoML to train a multi-class classification model that predicts sentiment labels based on the training data.
   • CUse the Natural Language API for real-time sentiment analysis.
   • DTrain a sentiment analysis model by using a BERT-based model, and fine-tune the model by using domain-specific customer reviews.

   Show answer

   Answer: B

   Next question

4. Question 344 · 364 in bank

   You are an enterprise AI architect in a large international company. An ML team has asked you to help with development of a pilot of a customer support bot. The ML team plans to use the DeepSeek-V3.2 model, and they expect a few hundred requests per day. You need to quickly deploy the model using a cost-efficient approach. What should you do?

   • AUse the DeepSeek-V3.2 model as a service (MaaS) in Model Garden. Send HTTP requests in the Agent Platform endpoint using the Python requests library.
   • BDownload the DeepSeek-V3.2 binaries from Model Garden, and deploy the model on a Google Kubernetes Engine (GKE) Autoscaling cluster. Send HTTP requests to the GKE endpoint using the Python requests library.
   • CNavigate to the DeepSeek-V3.2 model card in Model Garden. Click “Deploy model” to deploy it as an Agent Platform endpoint. Invoke the endpoint using the aiplatform SDK.
   • DUse the DeepSeek-V3.2 model as a service (MaaS) in Model Garden. Use a Python client from the openai SDK, and pass a custom endpoint URL to the client.

   Show answer

   Answer: D

   Next question

5. Question 345 · 364 in bank

   You have a high-traffic ecommerce application that uses a recommendation model deployed on an Agent Platform endpoint. During flash sales, traffic quickly spikes from 1,000 to 50,000 queries per second (QPS). During these spikes, the application returns 503: Service Unavailable errors because the autoscaler cannot provision new resources quickly enough to meet the sudden demand. Your need to eliminate service disruptions during the start of the sale. How should you configure the endpoint?

   • AChange the machine type to an a2-highgpu-1g instance with an NVIDIA A100 GPU.
   • BIncrease the min_replica_count to a baseline that covers a significant portion of the expected spike, and set the max_replica_count to 100.
   • CSet the min_replica_count to 1 and the max_replica_count to 100 in the deployed model’s automatic scaling configuration.
   • DDeploy the model to a private Agent Platform endpoint, and autoscale based on GPU utilization.

   Show answer

   Answer: B

   Next question

6. Question 346 · 364 in bank

   You lead an ML team at a financial technology company. Your team is building a system to detect potentially fraudulent credit card transactions in real-time. The model must return a prediction with very low latency (under 200ms) to approve or deny the transaction. The system must also scale to handle millions of transactions per day. Your dataset consists of structured, tabular transaction data stored in BigQuery. Your team is proficient in data science but wants to avoid the complexity of managing and optimizing a real-time model serving infrastructure. What should you do?

   • AIn an Agent Platform Workbench notebook, build a deep neural network (DNN) using PyTorch. Provision a cluster of GPU-enabled Compute Engine instances, and create an API to serve the model for real-time predictions.
   • BUse BigQuery ML to train a boosted tree classification model. Schedule a batch prediction job using the ML.PREDICT function that runs every minute and scores recent transactions.
   • CUse Forecasting with Agent Platform AutoML to train a time-series model on the sequence of transactions. Deploy the model as a tabular workflow to predict which future time windows are likely to contain fraudulent activity.
   • DUse Agent Platform AutoML to train using your tabular dataset from BigQuery. Deploy the trained model to an Agent Platform endpoint, and integrate the endpoint’s API into your transaction processing application.

   Show answer

   Answer: D

   Next question

7. Question 347 · 364 in bank

   You are training a ResNet-50 computer vision model on Gemini Enterprise Agent Platform using a custom PyTorch container. The training job is configured with a generic machine type and one NVIDIA A100 GPU. During the training, you monitor the resource utilization charts. You observe that the GPU utilization is volatile, frequently dropping to 0% for short intervals, while the CPU utilization remains consistently above 90%. You want to maximize GPU utilization to reduce training time. What should you do?

   • AIncrease the training batch size to saturate the GPU memory.
   • BUpdate the DataLoader utility to increase the num_workers parameter and enable pin_memory.
   • CSwitch to a machine type with a higher number of GPUs, and implement the torch.nn.parallel.DistributedDataParallel strategy in the training code.
   • DEnable automatic mixed precision (AMP) to speed up the matrix multiplications.

   Show answer

   Answer: B

   Next question

8. Question 348 · 364 in bank

   You work at a marketing research firm. Your team’s current project involves analyzing a large volume of customer surveys. Your team decides to use LLMs in Model Garden and track model training artifacts. You need to determine how to effectively share notebooks within your team and how to orchestrate the ML workflow. You want to use Google-managed services as much as possible. What should you do?

   • ADevelop your code in Agent Platform Colab Enterprise, and orchestrate your ML workflow by using Agent Platform Pipelines.
   • BDevelop your code in Agent Platform Colab Enterprise, and orchestrate your ML workflow by using Cloud Workflows.
   • CDevelop your code in Agent Platform Workbench, and orchestrate your ML workflow by using Cloud Build.
   • DDevelop your code in Agent Platform Workbench, and orchestrate your ML workflow by using Agent Platform Pipelines.

   Show answer

   Answer: A

   Next question

9. Question 349 · 364 in bank

   You are building an ML workflow that preprocesses data, trains a model with hyperparameter tuning, and deploys the best model to an Agent Platform endpoint. Your preprocessing step takes two hours and outputs a 100 GB dataset. You run experiments daily with different model architectures using the same preprocessing step. You need to automate this process and optimize the pipeline runs to minimize cost and time. What should you do?

   • AConfigure an Agent Platform Pipelines workflow with a single component for the preprocessing and training steps. Use Agent Platform hyperparameter tuning jobs within the component to identify optimal parameters.
   • BConfigure an Agent Platform Pipelines workflow with separate components for the preprocessing and training steps. Enable caching on the preprocessing component, and use Agent Platform Serverless Training custom jobs for the training steps.
   • CConfigure an Agent Platform Pipelines workflow with separate components for the preprocessing and training steps. Use a Cloud Run function to trigger the pipeline, and configure Agent Platform Serverless Training custom jobs for the training steps.
   • DConfigure an Agent Platform Pipelines workflow that starts with an Agent Platform Serverless Training custom job for the preprocessing step. Use a Cloud Run function to launch the hyperparameter tuning job that is triggered when the preprocessed data arrives in Cloud Storage.

   Show answer

   Answer: B

   Next question

10. Question 350 · 364 in bank

    You are an ML engineer at a retail company. You need to automate a recurring retraining workflow for a demand forecasting model. The workflow involves dependencies across hybrid environments and must execute the following sequential steps: 1. Trigger a shell script on an on-premises server to extract legacy sales data and send it to a Cloud Storage bucket. 2. Trigger a pipeline in Agent Platform Pipelines to train and validate the new model. 3. Push the validation metrics to a third-party API. You need a fully managed orchestration service that provides native connectors to handle these cross-platform dependencies. You want to use the most stable, robust solution. What should you do?

    • AUse Workflows to define the orchestration logic in YAML. Use the http.post connector to trigger the on-premises shell script and the Agent Platform API for the training job.
    • BUse Cloud Scheduler to trigger a Cloud Run function that executes the on-premises script using SSH, publishes a Pub/Sub message upon completion, and triggers the subsequent workflow steps.
    • CUse Managed Service for Apache Airflow to define a directed acyclic graph (DAG) in Python. Use Apache Airflow operators and sensors to orchestrate the workflow steps.
    • DUse Agent Platform Pipelines to develop a custom container component for each workflow step. Use the Kubeflow Pipelines (KFP) SDK to define the execution order.

    Show answer

    Answer: C

    Next question

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

PMLE practice, question 341. 364 questions available.