PMLE practice questions 121–130

1. Question 121 · 364 in bank

   You recently developed a wide and deep model in TensorFlow. You generated training datasets using a SQL script that preprocessed raw data in BigQuery by performing instance-level transformations of the data. You need to create a training pipeline to retrain the model on a weekly basis. The trained model will be used to generate daily recommendations. You want to minimize model development and training time. How should you develop the training pipeline?

   • AUse the Kubeflow Pipelines SDK to implement the pipeline. Use the BigQueryJobOp component to run the preprocessing script and the CustomTrainingJobOp component to launch a Vertex AI training job.
   • BUse the Kubeflow Pipelines SDK to implement the pipeline. Use the DataflowPythonJobOp component to preprocess the data and the CustomTrainingJobOp component to launch a Vertex AI training job.
   • CUse the TensorFlow Extended SDK to implement the pipeline Use the ExampleGen component with the BigQuery executor to ingest the data the Transform component to preprocess the data, and the Trainer component to launch a Vertex AI training job.
   • DUse the TensorFlow Extended SDK to implement the pipeline Implement the preprocessing steps as part of the input_fn of the model. Use the ExampleGen component with the BigQuery executor to ingest the data and the Trainer component to launch a Vertex AI training job.

   Show answer

   Answer: A

   Explanation

   My Answer: A According with this documentation: https://cloud.google.com/vertex-ai/docs/tabular-data/tabular-workflows/overview A: CORRECT: BigQueryJobOp for running the existing preprocessing script that already resides there, CustomTrainingJobOp for launching custom training jobs on Vertex AI, which aligns with the requirement of using the pre-trained TensorFlow model. B: Not Correct: While DataflowPythonJobOp can be used for preprocessingthis increasing development time compared to the simpler BigQueryJobOp approach. C and D: Not Correct: While possible, using the TensorFlow Extended SDK with its components introduces unnecessary complexity for this specific scenario. For example, why use ExampleGen? Implementing preprocessing within the model's input_fn is generally not recommended due to potential efficiency drawbacks and training-serving skew issues.

   Source: Adapted from community discussion — guilhermebutzke

   References

   • Google Cloud documentation

   Next question

2. Question 122 · 364 in bank

   You are training a custom language model for your company using a large dataset. You plan to use the Reduction Server strategy on Vertex AI. You need to configure the worker pools of the distributed training job. What should you do?

   • AConfigure the machines of the first two worker pools to have GPUs, and to use a container image where your training code runs. Configure the third worker pool to have GPUs, and use the reductionserver container image.
   • BConfigure the machines of the first two worker pools to have GPUs and to use a container image where your training code runs. Configure the third worker pool to use the reductionserver container image without accelerators, and choose a machine type that prioritizes bandwidth.
   • CConfigure the machines of the first two worker pools to have TPUs and to use a container image where your training code runs. Configure the third worker pool without accelerators, and use the reductionserver container image without accelerators, and choose a machine type that prioritizes bandwidth.
   • DConfigure the machines of the first two pools to have TPUs, and to use a container image where your training code runs. Configure the third pool to have TPUs, and use the reductionserver container image.

   Show answer

   Answer: B

   Explanation

   GPUs for Training: Configure the first two worker pools with GPUs to leverage the hardware acceleration capabilities for your custom language model training code. Reduction Server without GPUs: The third worker pool should use the reductionserver container image. This image is pre-configured for Reduction Server functionality and doesn't require GPUs. High-Bandwidth CPU: Choose a machine type with high bandwidth for the third pool since Reduction Server focuses on communication and gradient reduction.

   Source: Adapted from community discussion — fitri001

   References

   • Google Cloud documentation
   • Google Cloud documentation

   Next question

3. Question 123 · 364 in bank

   You have deployed multiple versions of an image classification model on AI Platform. You want to monitor the performance of the model versions over time. How should you perform this comparison?

   • ACompare the loss performance for each model on a held-out dataset.
   • BCompare the loss performance for each model on the validation data.
   • CCompare the receiver operating characteristic (ROC) curve for each model using the What-If Tool.
   • DCompare the mean average precision across the models using the Continuous Evaluation feature.

   Show answer

   Answer: D

   Explanation

   Guys, I not sure about the answer D ... And maybe you could help me in my arguments. I think choose loss to compare the model performance is better than see for metrics. For example, when can build an image model classification that has good precision metrics, because the class in unbalanced, but the loss could be terrible because of kind of loss choose that penalizes classes. so, losses are better than metrics to available models, and the answer is in A or B. I thought that the A could be the answer because I see validation as a part of the training process. So, If we want to test the model performance over time, we have to use new data, which I suppose to be the held-out data.

   Source: Adapted from community discussion — guilhermebutzke

   References

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

   Next question

4. Question 124 · 364 in bank

   You have trained a model by using data that was preprocessed in a batch Dataflow pipeline. Your use case requires real-time inference. You want to ensure that the data preprocessing logic is applied consistently between training and serving. What should you do?

   • APerform data validation to ensure that the input data to the pipeline is the same format as the input data to the endpoint.
   • BRefactor the transformation code in the batch data pipeline so that it can be used outside of the pipeline. Use the same code in the endpoint.
   • CRefactor the transformation code in the batch data pipeline so that it can be used outside of the pipeline. Share this code with the end users of the endpoint.
   • DBatch the real-time requests by using a time window and then use the Dataflow pipeline to preprocess the batched requests. Send the preprocessed requests to the endpoint.

   Show answer

   Answer: B

   Explanation

   My Answer B: B. This option ensures that the preprocessing logic used during training, which has already been validated and tested, is applied consistently during real-time inference. By making the transformation code reusable outside of the batch pipeline and utilizing it in the endpoint, you ensure that the same preprocessing steps are applied to incoming data during inference, thus maintaining consistency between training and serving. A: While data validation is essential, it only ensures the format. It doesn't guarantee consistent preprocessing logic between training and serving. C: Sharing code with end-users might not be desirable for security or maintainability reasons. D: Batching introduces latency and might not be suitable for real-time needs. Additionally, using the entire Dataflow pipeline might be inefficient for individual requests.

   Source: Adapted from community discussion — guilhermebutzke

   Next question

5. Question 125 · 364 in bank

   You need to develop a custom TensorFlow model that will be used for online predictions. The training data is stored in BigQuery You need to apply instance-level data transformations to the data for model training and serving. You want to use the same preprocessing routine during model training and serving. How should you configure the preprocessing routine?

   • ACreate a BigQuery script to preprocess the data, and write the result to another BigQuery table.
   • BCreate a pipeline in Vertex AI Pipelines to read the data from BigQuery and preprocess it using a custom preprocessing component.
   • CCreate a preprocessing function that reads and transforms the data from BigQuery. Create a Vertex AI custom prediction routine that calls the preprocessing function at serving time.
   • DCreate an Apache Beam pipeline to read the data from BigQuery and preprocess it by using TensorFlow Transform and Dataflow.

   Show answer

   Answer: D

   Explanation

   My answer: D According to this documentation, it is very clear that using BigQuery is not a good approach for online prediction at the instance level. That's because we won't use the same code for both training and prediction serving. In the same documentation, the final table on the page recommends using Dataflow with TensorFlow Transform for instance-level data transformation. https://www.tensorflow.org/tfx/guide/tft_bestpractices

   Source: Adapted from community discussion — guilhermebutzke

   Next question

6. Question 126 · 364 in bank

   You are pre-training a large language model on Google Cloud. This model includes custom TensorFlow operations in the training loop. Model training will use a large batch size, and you expect training to take several weeks. You need to configure a training architecture that minimizes both training time and compute costs. What should you do?

   • AImplement 8 workers of a2-megagpu-16g machines by using tf.distribute.MultiWorkerMirroredStrategy.
   • BImplement a TPU Pod slice with -accelerator-type=v4-l28 by using tf.distribute.TPUStrategy.
   • CImplement 16 workers of c2d-highcpu-32 machines by using tf.distribute.MirroredStrategy.
   • DImplement 16 workers of a2-highgpu-8g machines by using tf.distribute.MultiWorkerMirroredStrategy.

   Show answer

   Answer: A

   Explanation

   B. TPU Acceleration: the question says that uses Tensorflow custom operations in the main loop and Google documentation literatelly says about TPU use: "Models with no custom TensorFlow/PyTorch/JAX operations inside the main training loop" C. High-CPU Machines: Make no sense because tell you to use a cpu (which does not help us in this case) So the correct answer is between A and D. However the question says that they are planning to use a large batch size so we need RAM. Therefore we should take the one with more. Correct answer: Option A

   Source: Adapted from community discussion — ccb23cc

   Next question

7. Question 127 · 364 in bank

   You are building a TensorFlow text-to-image generative model by using a dataset that contains billions of images with their respective captions. You want to create a low maintenance, automated workflow that reads the data from a Cloud Storage bucket collects statistics, splits the dataset into training/validation/test datasets performs data transformations trains the model using the training/validation datasets, and validates the model by using the test dataset. What should you do?

   • AUse the Apache Airflow SDK to create multiple operators that use Dataflow and Vertex AI services. Deploy the workflow on Cloud Composer.
   • BUse the MLFlow SDK and deploy it on a Google Kubernetes Engine cluster. Create multiple components that use Dataflow and Vertex AI services.
   • CUse the Kubeflow Pipelines (KFP) SDK to create multiple components that use Dataflow and Vertex AI services. Deploy the workflow on Vertex AI Pipelines.
   • DUse the TensorFlow Extended (TFX) SDK to create multiple components that use Dataflow and Vertex AI services. Deploy the workflow on Vertex AI Pipelines.

   Show answer

   Answer: D

   Explanation

   Airflow (A): While versatile, Airflow often requires more manual configuration and integration with ML services, potentially increasing maintenance effort. MLFlow (B): MLFlow focuses on experiment tracking and model management, lacking built-in pipeline components for data processing and model training. Kubeflow Pipelines (C): KFP is flexible but requires more setup and infrastructure management compared to TFX's managed services.

   Source: Adapted from community discussion — pikachu007

   Next question

8. Question 128 · 364 in bank

   You are developing an ML pipeline using Vertex AI Pipelines. You want your pipeline to upload a new version of the XGBoost model to Vertex AI Model Registry and deploy it to Vertex AI Endpoints for online inference. You want to use the simplest approach. What should you do?

   • AUse the Vertex AI REST API within a custom component based on a vertex-ai/prediction/xgboost-cpu image
   • BUse the Vertex AI ModelEvaluationOp component to evaluate the model
   • CUse the Vertex AI SDK for Python within a custom component based on a python:3.10 image
   • DChain the Vertex AI ModelUploadOp and ModelDeployOp components together

   Show answer

   Answer: D

   Explanation

   Built-in Functionality: Both ModelUploadOp and ModelDeployOp are pre-built components within Vertex AI Pipelines specifically designed for uploading models and deploying them to endpoints. Ease of Use: These components offer a user-friendly interface within the pipeline definition. You only need to specify essential details like the model path, container image URI (pre-built for XGBoost is available), endpoint configuration, etc. Reduced Code Complexity: Using these components eliminates the need for writing custom code within your pipeline for model upload and deployment, simplifying your pipeline logic.

   Source: Adapted from community discussion — fitri001

   References

   • Google Cloud documentation

   Next question

9. Question 129 · 364 in bank

   You work for an online retailer. Your company has a few thousand short lifecycle products. Your company has five years of sales data stored in BigQuery. You have been asked to build a model that will make monthly sales predictions for each product. You want to use a solution that can be implemented quickly with minimal effort. What should you do?

   • AUse Prophet on Vertex AI Training to build a custom model.
   • BUse Vertex AI Forecast to build a NN-based model.
   • CUse BigQuery ML to build a statistical ARIMA_PLUS model.
   • DUse TensorFlow on Vertex AI Training to build a custom model.

   Show answer

   Answer: C

   Explanation

   Ease of Use: BigQuery ML integrates seamlessly with BigQuery, allowing you to create and train models directly within SQL queries, eliminating the need for separate environments or coding. Statistical ARIMA_PLUS Strengths: This model is well-suited for time series forecasting, automatically handling seasonality, trends, and holidays, making it appropriate for monthly sales predictions. Minimal Effort: BigQuery ML handles model training and tuning, reducing the need for manual configuration or hyperparameter tuning. Fast Implementation: Model creation and training can be done in a few lines of SQL, enabling rapid deployment.

   Source: Adapted from community discussion — pikachu007

   Next question

10. Question 130 · 364 in bank

    You are creating a model training pipeline to predict sentiment scores from text-based product reviews. You want to have control over how the model parameters are tuned, and you will deploy the model to an endpoint after it has been trained. You will use Vertex AI Pipelines to run the pipeline. You need to decide which Google Cloud pipeline components to use. What components should you choose?

    • ATabularDatasetCreateOp, CustomTrainingJobOp, and EndpointCreateOp
    • BTextDatasetCreateOp, AutoMLTextTrainingOp, and EndpointCreateOp
    • CTabularDatasetCreateOp. AutoMLTextTrainingOp, and ModelDeployOp
    • DTextDatasetCreateOp, CustomTrainingJobOp, and ModelDeployOp

    Show answer

    Answer: D

    Explanation

    TextDatasetCreateOp: This component is specifically designed to handle text-based data like product reviews. It reads and prepares the text data for training the model. CustomTrainingJobOp: Since you want control over hyperparameter tuning, a custom training job is the most suitable option. This component allows you to define your training script using a framework like TensorFlow and configure hyperparameters for optimization. ModelDeployOp: After training, this component uploads the trained model to the Vertex AI Model Registry and deploys it to a Vertex AI Endpoint for serving predictions.

    Source: Adapted from community discussion — fitri001

    Next question

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

PMLE practice, question 121. 364 questions available.