PMLE Architecting Low-Code ML Solutions — 15 Questions

Question 1

A company wants to use BigQuery ML to train a DNN_CLASSIFIER model on a dataset with 100 million rows. They are concerned about training time and cost. Which approach can help optimize training performance while staying within BigQuery ML?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

BigQuery ML automatically optimizes training; no additional configuration needed. BigQuery ML automatically handles many training optimizations internally, including distributed training across slots and query optimization, so no additional configuration is strictly required to train a DNN_CLASSIFIER on 100 million rows. While options like sampling or limiting iterations can reduce cost, they change model quality and are not the 'optimize while staying within BigQuery ML' answer the exam expects. The correct answer reflects that BigQuery ML manages training performance natively.

Answer

Use OPTIONS('MAX_ITERATIONS' = 10) to limit training iterations

Answer

Use Vertex AI AutoML Tables instead of BigQuery ML

Answer

Train on a 10% random sample of the data to reduce cost

Question 2

A company needs to analyze customer feedback from app reviews to identify common themes and sentiment. They have millions of reviews in multiple languages. Which combination of pre-built APIs should they use?

Accepted Answer

Translation API then Natural Language API. The reviews are text in multiple languages, so the Translation API is needed to convert them into a single language (e.g., English) that the Natural Language API can analyze for sentiment and entity/theme extraction. The Natural Language API supports multiple languages for sentiment, but its entity and syntax analysis is optimized for English, and translation ensures consistent theme detection across all reviews. Combining Translation API then Natural Language API directly addresses the multilingual text requirement.

Answer

Speech-to-Text then Natural Language API

Answer

Natural Language API only

Answer

Vision API then Natural Language API

Question 3

A retail company uses Recommendations AI to power personalized product recommendations on their website. They notice that the 'frequently-bought-together' model is not capturing complementary items that are often purchased in the same session but not necessarily in the same transaction. Which TWO actions should they take to improve the model?

Accepted Answer

Enable the 'others-you-may-like' recommendation type in addition to 'frequently-bought-together'. Option B is correct because enabling the 'others-you-may-like' recommendation type complements 'frequently-bought-together' by surfacing related items based on broader user behavior and co-view/co-purchase signals, which helps capture complementary products that appear in the same session but not the same transaction. Option E is correct because Recommendations AI learns from user event data, and ingesting session-level events such as product detail views, add-to-cart actions, and page views within the same session gives the model the behavioral context needed to identify items frequently browsed together even when they are not bought in one transaction. The unmarked options do not belong: A would reduce historical signal and hurt learning, C introduces a separate custom modeling path rather than improving the existing Recommendations AI model, and D would restrict the model to only 'frequently-bought-together', which is exactly the limitation being addressed.

Answer

Decrease the event retention period to focus on recent purchases

Answer

Use AutoML Tables to build a custom recommendation model

Answer

Set the recommendation type to only 'frequently-bought-together'

Question 4

A retail company wants to build a recommendation system for their e-commerce website. They have user purchase history and product metadata. Which Google Cloud service is most suitable for building a 'frequently bought together' recommendation model with minimal custom ML development?

Accepted Answer

Recommendations AI with the 'frequently bought together' model type. Recommendations AI is a Google Cloud service specifically designed for building recommendation systems with minimal ML expertise. It offers pre-built model types, including 'frequently bought together', which directly addresses the requirement. This allows the retail company to deploy a recommendation model quickly without custom development.

Answer

Vertex AI Prediction with a custom TensorFlow model

Answer

BigQuery ML with MATRIX_FACTORIZATION model type

Answer

Vertex AI AutoML Tables

Question 5

A retailer wants to implement a recommendation engine that suggests products based on a user's current cart. They have limited ML expertise and want a quick deployment. Which Recommendations AI model type should they use?

Accepted Answer

Frequently bought together. The 'Frequently bought together' model type is the correct choice because it directly leverages collaborative filtering based on co-purchase patterns in historical transaction data, enabling the retailer to recommend items commonly purchased alongside the current cart contents. This model requires minimal ML expertise and can be quickly deployed using pre-built Recommendations AI templates, as it does not require user-level personalization or real-time session data.

Answer

Others you may like

Answer

Recently viewed

Answer

Recommended for you

Question 6

A company needs to detect objects in real-time from a live video feed. They want to use a pre-trained model with minimal setup. Which Google Cloud service should they use?

Accepted Answer

Video Intelligence API. The Video Intelligence API is the correct choice because it provides pre-trained models for real-time video analysis, including object detection, with minimal setup. It is designed to extract metadata from video streams without requiring custom training, making it ideal for real-time live video feeds. Vertex AI AutoML Video, in contrast, requires custom training and is not pre-trained for immediate use.

Answer

Vertex AI AutoML Video

Answer

Document AI

Answer

Cloud Vision API

Question 7

A data scientist wants to use BigQuery ML for time-series forecasting. They need to evaluate model accuracy and compare different models. Which TWO BigQuery ML functions should they use?

Accepted Answer

ML.PREDICT. For evaluating model accuracy and comparing different models in BigQuery ML, you use ML.EVALUATE to compute metrics and ML.PREDICT to generate forecasts. ML.TRAIN is not a valid function; training is done via CREATE MODEL. ML.FEATURE_IMPORTANCE and ML.EXPLAIN_PREDICT are for model interpretability, not for evaluation or comparison.

Answer

ML.EXPLAIN_PREDICT

Answer

ML.FEATURE_IMPORTANCE

Answer

ML.TRAIN

Question 8

A company needs to classify images of products into categories (e.g., electronics, clothing, food). They have labeled images and want to use a low-code solution on Google Cloud. Which service is suitable for this task?

Accepted Answer

AutoML Vision. AutoML Vision is the only low-code service among the options that allows you to train custom image classification models using labeled images without writing code. It directly matches the requirement to classify product images into categories. The other options are unsuitable: AutoML Tables is for tabular data, Cloud Vision API using product search is for finding similar products rather than custom classification, Vertex AI Workbench with a custom TensorFlow model requires custom coding, and Document AI is designed for document processing.

Answer

AutoML Tables

Answer

Cloud Vision API using product search

Answer

Vertex AI Workbench with a custom TensorFlow model

Answer

Document AI custom extractor

Question 9

A company wants to transcribe customer service calls in real-time. The audio is telephony quality (8 kHz). Which Speech-to-Text model should they specify?

Accepted Answer

telephony. The 'telephony' model in Google Cloud Speech-to-Text is specifically optimized for transcribing audio from telephone calls, which typically use 8 kHz sample rate and have narrowband characteristics. It is trained on telephony data to handle the unique challenges of phone audio, such as background noise, compression artifacts, and limited frequency range. For real-time transcription of customer service calls, this model provides the highest accuracy. Other models like latest_short or latest_long are designed for different audio types and may not perform as well on telephony audio.

Answer

latest_short

Answer

latest_long

Answer

default

Question 10

A company uses BigQuery ML with a remote model calling Vertex AI's pre-trained image classification model. They need to classify images stored in Cloud Storage buckets. What is the correct approach?

Accepted Answer

Create a remote model with model_type='VERTEX_AI' and use ML.PREDICT with image URIs.. BigQuery ML remote models allow you to invoke Vertex AI pre-trained models via the `model_type='VERTEX_AI'` setting. You can then use `ML.PREDICT` directly on Cloud Storage image URIs without needing to export or transform the image data, as BigQuery ML handles the URI resolution and passes the image to Vertex AI for classification.

Answer

Train a custom model in BigQuery ML with image data.

Answer

Export images to BigQuery as base64 and then use ML.PREDICT.

Answer

Use ML.PREDICT with IMAGE data type directly.

Question 11

A data scientist wants to use AutoML Tables to build a binary classification model for loan default prediction. The dataset has 200 features and 1 million rows, with highly imbalanced classes. Which TWO options should they consider? (Choose 2)

Accepted Answer

Set the target column to 'default'. Option B is correct because AutoML Tables requires the user to explicitly designate the target column, and in this loan default scenario the target column is 'default', which the model will learn to predict as a binary label. Option E is correct because with highly imbalanced classes, AUC (Area Under the ROC Curve) is a robust optimisation objective that evaluates ranking quality across all classification thresholds, making it far more suitable than accuracy, which can be misleadingly high for the majority class. Option A is not required or specifically indicated here, since AutoML Tables performs automatic feature engineering by default and the question asks what the data scientist should consider, not a default behaviour. Option C is wrong because disabling hyperparameter tuning would degrade model quality and is not a recommended cost-saving measure for a high-value imbalanced classification task. Option D is wrong because a manual data split is not necessary; AutoML Tables provides automatic train/validation/test splitting, and manual splitting is typically used only for specific temporal or grouped requirements not mentioned here.

Answer

Enable automatic feature engineering

Answer

Disable hyperparameter tuning to save cost

Answer

Use Manual data split

Question 12

A company needs to detect objects in live video streams from security cameras. They require low-latency predictions and want to minimise operational overhead. Which TWO services should they use? (Choose 2)

Accepted Answer

Cloud Run. Option C (Cloud Run) is correct because it provides a fully managed, serverless container platform that scales automatically and can host a custom low-latency inference service for processing live video frames, minimizing operational overhead since no infrastructure is managed. Option E (Video Intelligence API) is correct because it offers pre-trained machine learning models specifically for video analysis, including object detection and tracking, and can be invoked directly on streaming or stored video without building or maintaining any ML infrastructure. Option A (Live Stream API) is not correct because it is designed for transcoding and packaging live video into streaming formats for delivery, not for performing object detection. Option B (Vertex AI Predictions) is not correct because it is a general-purpose model-serving endpoint that would require the company to train and deploy their own object detection model, adding operational overhead. Option D (AutoML Video) is not correct because it is used to train custom video classification or object tracking models, which is a training service rather than a low-latency live-stream prediction service.

Answer

Live Stream API

Answer

Vertex AI Predictions

Answer

AutoML Video

Question 13

A company wants to use Document AI to process a large volume of invoices. They need to extract line items and also have a human review the extracted data for accuracy. Which THREE features should they use? (Choose 3)

Accepted Answer

Invoice Parser Processor. The Invoice Parser Processor (C) is a specialized Document AI processor designed to extract structured data from invoices, including line items, which directly meets the requirement for extracting line items from a large volume of invoices. It is pre-trained on invoice layouts and can handle complex table structures, making it the correct choice for this task.

Answer

Online Processing

Answer

Custom Extractor

Question 14

A data engineer is using BigQuery ML with a BOOSTED_TREE_CLASSIFIER model. After training, they want to evaluate the model and understand which features contribute most to predictions. Which THREE BigQuery ML functions should they use?

Accepted Answer

ML.EVALUATE. ML.EVALUATE (A) is correct because it computes evaluation metrics (e.g., accuracy, precision, recall, ROC AUC, log loss) for a trained BOOSTED_TREE_CLASSIFIER model, which is exactly what the engineer needs to assess model performance. ML.FEATURE_IMPORTANCE (D) is correct because it returns the relative contribution of each input feature to the boosted tree model's predictions, directly addressing the requirement to understand which features matter most. ML.PREDICT (E) is correct because it applies the trained model to input data to generate predictions and, for boosted trees, can also surface per-prediction explanation details useful for interpreting model behavior. ML.GLOBAL_EXPLAIN (B) is not among the marked answers; while it exists for explainable AI on supported models, it is not one of the three functions specified here. ML.TRAIN (C) is not marked because it is used to create/train the model, not to evaluate it or inspect feature contributions after training.

Answer

ML.GLOBAL_EXPLAIN

Answer

ML.TRAIN

Question 15

A company needs to build a custom model to classify images of products into categories. They have a large labeled dataset. They want to use AutoML but are unsure which options support image classification. Which TWO AutoML products support image classification?

Accepted Answer

AutoML Video. AutoML Vision (C) is correct because it is Google Cloud's dedicated AutoML service for training custom image classification and object detection models on labeled image datasets, exactly matching the scenario of classifying product images into categories. AutoML Video (B) is also correct because it supports image classification as part of its video content classification capabilities, allowing custom models to classify visual content, including per-frame image classification tasks. AutoML Natural Language (A) is incorrect because it handles text classification, entity extraction, and sentiment analysis, not images. AutoML Translation (D) is incorrect because it builds custom machine translation models between languages. AutoML Tables (E) is incorrect because it is designed for structured/tabular data, not image data.

Answer

AutoML Natural Language

Answer

AutoML Translation

Answer

AutoML Tables