Professional Machine Learning EngineerDemo

You built a custom model that carries out several memory-intensive preprocessing tasks before making a prediction. You deployed the model to a Vertex AI endpoint and confirmed that results were returned within a reasonable time. After directing user traffic to the endpoint, you find that it does not autoscale as expected when it receives multiple requests. What should you do?

You need to create classification workflows across several structured datasets that are currently stored in BigQuery. Because you will perform the classification multiple times, you want to complete these steps without writing code: exploratory data analysis, feature selection, model building, training, hyperparameter tuning, and serving. What should you do?

You work for a magazine distributor and must build a model that predicts which customers will renew their subscriptions in the coming year. Using your company’s historical data as the training set, you created a TensorFlow model and deployed it to AI Platform. You need to identify which customer attribute has the greatest predictive influence for each prediction served by the model. What should you do?

You work for an online travel agency that also sells advertising placements on its website to other companies. You have been asked to predict the most relevant web banner that a user should see next. Security is important to your company. The model-latency requirement is 300ms@p99, the inventory contains thousands of web banners, and your exploratory analysis has shown that navigation context is a good predictor. You want to implement the simplest solution. How should you configure the prediction pipeline?

You created an ML pipeline that has several input parameters. You want to explore the tradeoffs among different parameter combinations. The parameter options are:

• Input dataset
• Maximum tree depth of the boosted tree regressor
• Optimizer learning rate

You need to compare pipeline performance across the parameter combinations using F1 score, training time, and model complexity. You want the approach to be reproducible and to track every pipeline run on the same platform. What should you do?

Your company recently migrated several of its ML models to Google Cloud. You have begun developing models in Vertex AI. You need to implement a system to track model artifacts and model lineage. You want to create a simple, effective solution that can be reused for future models. What should you do?

You are building an ML model in a Vertex AI Workbench notebook. You want to track artifacts and compare models while experimenting with different approaches. As you iterate on the model implementation, you need to move successful experiments to production quickly and easily. What should you do?

You are analyzing customer data for a healthcare organization that is stored in Cloud Storage. The data includes personally identifiable information (PII). You need to carry out data exploration and preprocessing while maintaining the security and privacy of sensitive fields. What should you do?

You work for a textile manufacturing company. The company has hundreds of machines, each with many sensors. Your team used the sensor data to build hundreds of ML models that detect machine anomalies. The models are retrained daily, and you must deploy them cost-effectively. They must run 24/7 without downtime and provide sub-millisecond predictions.

What should you do?

You are building an ML pipeline for data processing, model training, and model deployment that uses different Google Cloud services. You have developed code for every individual task and anticipate a high frequency of new files. You now need to add an orchestration layer above these tasks. The orchestration pipeline must run only when new files exist in your dataset in a Cloud Storage bucket. You also need to minimize compute-node costs. What should you do?

You work for a hospital that wants to improve how it schedules operations. You need to create a model using the relationship between the number of scheduled surgeries and occupied beds. You want to predict, in advance, how many beds patients will need each day based on the scheduled surgeries. You have one year of hospital data arranged in 365 rows.

The data contains these variables for each day:

• Number of scheduled surgeries
• Number of beds occupied
• Date

You want to maximize the speed of model development and testing. What should you do?

Your team is training many ML models that use different algorithms, parameters, and datasets. Some models are trained in Vertex AI Pipelines, while others are trained on Vertex AI Workbench notebook instances. The team wants to compare model performance across both services while minimizing the effort needed to store parameters and metrics. What should you do?

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

You recently deployed an ML model. Three months after deployment, you find that the model performs poorly for certain subgroups, potentially producing biased outcomes. You suspect this inequitable performance results from class imbalances in the training data, but you cannot gather additional data. What should you do?

You work for a telecommunications company and are creating a model to predict which customers might not pay their next phone bill. The model is intended to proactively offer at-risk customers support, such as service discounts and extensions to bill due dates. The data is stored in BigQuery, and the predictive features available for training the model are:

• Customer_id
• Age
• Salary (measured in local currency)
• Sex
• Average bill value (measured in local currency)
• Number of phone calls in the last month (integer)
• Average duration of phone calls (measured in minutes)

You must investigate and mitigate potential bias against disadvantaged groups while maintaining model accuracy.

What should you do?

You created a Vertex AI pipeline with two steps. The first step preprocesses 10 TB of data, completes in about 1 hour, and saves the result to a Cloud Storage bucket. The second step uses the processed data to train a model.

You need to update the model’s code so that you can test different algorithms. You want to reduce pipeline execution time and cost while minimizing changes to the pipeline. What should you do?

Your team is conducting an NLP research project to predict authors’ political affiliation from articles they have written. You have a large training dataset structured as follows:

You used the standard 80%-10%-10% distribution for the training, testing, and evaluation subsets. How should the training examples be allocated among the train-test-eval subsets while preserving the 80-10-10 proportion?

You work for an ecommerce startup and must create a customer-churn prediction model. Your company’s latest sales records are stored in a BigQuery table. You want to understand how the initial model makes its predictions, and you want to iterate on the model as rapidly as possible while keeping costs low. How should you build your first model?

A proof-of-concept (POC) deep learning model has recently been created. The overall architecture is satisfactory, but the values of a couple of hyperparameters still need to be determined. You want to use Vertex AI hyperparameter tuning to identify both the suitable embedding dimension for a categorical feature used by the model and the optimal learning rate. Configure these settings:

• For the embedding dimension, set the type to INTEGER, with a minValue of 16 and a maxValue of 64.
• For the learning rate, set the type to DOUBLE, with a minValue of 10e-05 and a maxValue of 10e-02.

You use the default Bayesian optimization tuning algorithm and want to maximize model accuracy. Training time is not a concern. How should the hyperparameter scaling for each hyperparameter and maxParallelTrials be set?

You are building an ML model that predicts house prices. During data preparation, you find that an important predictor variable—the distance to the nearest school—is frequently missing and has low variance. Every instance (row) in the data is important. How should the missing data be handled?

You are creating a process to train and run your custom model in production. You need to demonstrate lineage for the model and its predictions. What should you do?

You work for a mobile-gaming startup that develops online multiplayer games. Recently, the company has seen more players cheating in its games, resulting in lost revenue and a poor user experience. You built a binary classification model to identify whether a player cheated after a game session is completed, then send a message to downstream systems to ban a player who cheated. Your model performed well in testing, and you now need to deploy it to production. You want the serving solution to return classifications immediately after a completed game session to prevent further revenue loss. What should you do?

You are an AI architect at a widely used photo-sharing social media platform. Your organization’s content moderation team currently reviews user-uploaded images and manually removes explicit content. You want to implement an AI service that automatically stops users from uploading explicit images. What should you do?

You are building a recommendation engine for an online clothing store. Historical customer-transaction data is stored in BigQuery and Cloud Storage.

You need to perform exploratory data analysis (EDA), preprocessing, and model training. You plan to rerun these EDA, preprocessing, and training steps while experimenting with different algorithm types. You want to minimize the cost and development effort required to run these steps as you experiment. How should you configure the environment?