GCP-PMLE practice questions

#261

You're tasked with creating a machine learning recommendation model for your company's e-commerce website using Recommendations AI. What is the best approach to develop recommendations that boost revenue while adhering to established best practices?

#262Pilih 2

You have recently deployed a machine learning model, and it has come to your attention that after three months of deployment, the model is exhibiting underperformance on specific subgroups, raising concerns about potential bias in the results. This inequitable performance is suspected to be linked to class imbalances in the training data, and the option of collecting additional data is not available. In this situation, what steps should you take? (Select two options.)

#263

You work on a team that builds state-of-the-art deep learning models by using the TensorFlow framework. Your team runs multiple ML experiments each week, which makes it difficult to track the experiment runs. You want a simple approach to effectively track, visualize, and debug ML experiment runs on Google Cloud while minimizing any overhead code. How should you proceed?

#264

As an ML engineer at a regulated insurance firm, you've been tasked with creating a model to approve or reject insurance applications. What key considerations should you take into account before developing this model?

#265

You've recently set up a new Google Cloud project and successfully tested the submission of a Vertex AI Pipeline job from Cloud Shell. Now, you're attempting to run your code from a Vertex AI Workbench user-managed notebook instance, but the job fails with an insufficient permissions error. What action should you take?

#266

You are in the process of building an ML model that analyzes segmented frames extracted from a video feed and generates bounding boxes around specific objects. Your goal is to automate various stages of your training pipeline, which include data ingestion and preprocessing from Cloud Storage, training the object model along with hyperparameter tuning using Vertex AI jobs, and ultimately deploying the model to an endpoint. To orchestrate the entire pipeline while minimizing the need for cluster management, which approach should you adopt?

#267

You've created a custom model in Vertex AI to predict user churn rate for your application. Vertex AI Model Monitoring is used for skew detection, and your training data in BigQuery includes two types of features: demographic and behavioral. Recently, you found that two separate models, each trained on one of these feature sets, outperform the original model. Now, you want to set up a new model monitoring pipeline that directs traffic to both models while maintaining consistent prediction-sampling rates and monitoring frequencies. You also aim to minimize management overhead. What should be your approach?

#268

You have a custom job that runs on Vertex AI on a weekly basis. The job is implemented using a proprietary ML workflow that produces the datasets, models, and custom artifacts, and sends them to a Cloud Storage bucket. Many different versions of the datasets and models were created. Due to compliance requirements, your company needs to track which model was used for making a particular prediction, and needs access to the artifacts for each model. How should you configure your workflows to meet these requirements?

#269

You are in the process of creating an ML model that aims to classify X-ray images to assess the risk of bone fractures. You've already trained a ResNet model on Vertex AI using a TPU as an accelerator, but you're not satisfied with the training time and memory usage. Your goal is to rapidly iterate on the training code with minimal code modifications and without significantly affecting the model's accuracy. What steps should you take to achieve this?

#270

You are profiling the performance of your TensorFlow model training time and have identified a performance issue caused by inefficiencies in the input data pipeline. This issue is particularly evident when working with a single 5 terabyte CSV file dataset stored on Cloud Storage. What should be your initial action to improve the efficiency of your pipeline?

#271

While executing a model training pipeline on Vertex AI, it has come to your attention that the evaluation step is encountering an out-of-memory error. Your current setup involves the use of TensorFlow Model Analysis (TFMA) within a standard Evaluator component of the TensorFlow Extended (TFX) pipeline for the evaluation process. Your objective is to address this issue and stabilize the pipeline's performance without compromising the quality of evaluation, all while keeping infrastructure overhead to a minimum. What course of action should you take?

#272

You work for an organization that operates a cloud-based communication platform combining chat, voice, and video conferencing. The platform stores audio recordings with an 8 kHz sample rate, all lasting over a minute. You are tasked with implementing a feature that automatically transcribes voice call recordings into text for future applications like call summarization and sentiment analysis. How should you implement this voice call transcription feature according to Google-recommended best practices?

#273

You have previously deployed an ML model into production, and as part of your ongoing maintenance, you collect all the raw requests directed to your model prediction service on a monthly basis. Subsequently, you select a subset of these requests for evaluation by a human labeling service to assess your model's performance. Over the course of a year, you have observed that your model's performance exhibits varying patterns: at times, there is a significant degradation in performance within a month, while in other instances, it takes several months before any noticeable decrease occurs. It's important to note that utilizing the labeling service incurs significant costs, but you also want to avoid substantial performance drops. In light of these considerations, you aim to establish an optimal retraining frequency for your model. This approach should enable you to maintain a consistently high level of performance while simultaneously minimizing operational costs. What steps should you take to achieve this?

#274

You've developed unit tests for a Kubeflow Pipeline, which depend on custom libraries. To automate these unit tests' execution following each new push to the development branch in Cloud Source Repositories, what steps should you take?

#275

You recently used Vertex AI Prediction to deploy a custom-trained model in production. The automated re-training pipeline made available a new model version that passed all unit and infrastructure tests. You want to define a rollout strategy for the new model version that guarantees an optimal user experience with zero downtime. What should you do?

#276

You have developed an ML model to detect the sentiment of users’ posts on your company's social media page to identify outages or bugs. You are using Dataflow to provide real-time predictions on data ingested from Pub/Sub. You plan to have multiple training iterations for your model and keep the latest two versions live after every run. You want to split the traffic between the versions in an 80:20 ratio, with the newest model getting the majority of the traffic. You want to keep the pipeline as simple as possible, with minimal management required. What should you do?

#277

You've trained a model that necessitated resource-intensive preprocessing operations on a dataset. Now, these preprocessing steps must be replicated during prediction. Given that the model is deployed on AI Platform for high-throughput online predictions, what is the most suitable architectural approach to use?

#278

You are developing a linear model that involves more than 100 input features, all of which have values ranging from -1 to 1. You have a suspicion that many of these features do not provide valuable information for your model. Your objective is to eliminate the non-informative features while preserving the informative ones in their original state. What technique is most suitable for achieving this goal?

#279

You are working at a hospital and have received approval to collect patient data for machine learning purposes. Using this data, you trained a Vertex AI tabular AutoML model to predict patient risk scores for hospital admissions. The model has been deployed, but you are concerned that over time, changes in patient demographics might alter the relationships between features, potentially affecting prediction accuracy. To address this, you need a cost-effective way to monitor for such changes and understand feature importance in the predictions. What should you do?

#280

You recently developed a deep learning model. To test your new model, you trained it for a few epochs on a large dataset. You observe that the training and validation losses barely changed during the training run. You want to quickly debug your model. What should you do first?