Architecting Low-Code AI Solutions - Quiz

Question 1 of 15

You are building a model to predict customer churn (binary outcome: yes/no) using BigQuery ML. You have a dataset with customer demographics, usage patterns, and historical churn data. Which BigQuery ML model type is most appropriate for this task?

CREATE MODEL with model_type='LINEAR_REG' CREATE MODEL with model_type='LOGISTIC_REG' CREATE MODEL with model_type='KMEANS' CREATE MODEL with model_type='ARIMA_PLUS'

Explanation

LOGISTIC_REG is the correct choice for binary classification problems like predicting customer churn. LINEAR_REG is for continuous numerical predictions, KMEANS is for clustering, and ARIMA_PLUS is for time series forecasting.

Question 2 of 15

Your company wants to forecast monthly sales for the next 6 months using historical sales data stored in BigQuery. Which BigQuery ML model type should you use?

BOOSTED_TREE_REGRESSOR DNN_CLASSIFIER ARIMA_PLUS MATRIX_FACTORIZATION

Explanation

ARIMA_PLUS is specifically designed for time series forecasting in BigQuery ML. It can capture trends, seasonality, and patterns in historical data to predict future values. The other options are not designed for time series analysis.

Question 3 of 15

You need to build a recommendation system in BigQuery ML to suggest products to customers based on their purchase history. Which model type is most suitable?

LINEAR_REG MATRIX_FACTORIZATION AUTOENCODER LOGISTIC_REG

Explanation

MATRIX_FACTORIZATION is designed for recommendation systems in BigQuery ML. It decomposes the user-item interaction matrix to find latent factors and make personalized recommendations. This is the standard approach for collaborative filtering.

Question 4 of 15

You are evaluating a binary classification model in BigQuery ML that predicts loan defaults. The model has high precision (95%) but low recall (40%). What does this indicate?

The model correctly identifies most actual defaults but has many false positives The model misses many actual defaults but rarely incorrectly predicts a default The model has balanced performance across all metrics The model is overfitting to the training data

Explanation

High precision with low recall means the model is very conservative in predicting defaults. When it predicts a default, it's usually correct (95% precision), but it misses many actual defaults (only catches 40% via recall). This could be problematic if catching all defaults is critical.

Question 5 of 15

In BigQuery ML, you want to perform feature engineering by creating polynomial features for your linear regression model. Which SQL clause allows you to transform features during model training?

TRANSFORM clause in the CREATE MODEL statement FEATURE_ENGINEERING clause in the CREATE MODEL statement You must manually create feature columns before training PREPROCESS clause in the ML.TRAIN_MODEL statement

Explanation

The TRANSFORM clause in BigQuery ML's CREATE MODEL statement allows you to specify feature transformations like polynomial features, scaling, encoding, and more. These transformations are automatically applied during both training and prediction.

Question 6 of 15

You trained a regression model in BigQuery ML and received an R-squared value of 0.85. What does this indicate about your model?

The model explains 85% of the variance in the target variable The model has 85% accuracy on the test set The model has 15% error rate 85% of predictions are within one standard deviation

Explanation

R-squared (coefficient of determination) measures the proportion of variance in the dependent variable that is explained by the model. An R-squared of 0.85 means the model explains 85% of the variance in the target variable, which generally indicates a good fit.

Question 7 of 15

You need to generate batch predictions for 10 million customer records using a trained BigQuery ML model. Which approach is most efficient?

Use ML.PREDICT with a JOIN to your customer table Export the model and run predictions in Python locally Create a Cloud Function to call the model for each record Use Vertex AI Batch Prediction service

Explanation

ML.PREDICT with a JOIN in BigQuery is the most efficient approach for batch predictions on large datasets already in BigQuery. It leverages BigQuery's distributed processing and eliminates data movement. The data stays in BigQuery, making it much faster than exporting or using external services.

Question 8 of 15

You want to automatically handle missing values and one-hot encode categorical variables in BigQuery ML. What feature should you use?

Manual preprocessing with SQL CASE statements before CREATE MODEL AUTO preprocessing option in CREATE MODEL OPTIONS TRANSFORM clause with explicit ML.STANDARD_SCALER and ML.ONE_HOT_ENCODER BigQuery ML automatically handles this without any configuration

Explanation

BigQuery ML automatically handles missing values and performs one-hot encoding for categorical variables (STRING type) without requiring explicit configuration. This automatic feature engineering is one of the key benefits of BigQuery ML for rapid model development.

Question 9 of 15

Your BigQuery ML classification model shows an F1-score of 0.72. The business stakeholder asks what this means. What is the best explanation?

The model is 72% accurate overall The model achieves a harmonic mean of 72% between precision and recall, balancing false positives and false negatives 72% of the data was used for training The model has a 28% error rate

Explanation

The F1-score is the harmonic mean of precision and recall, providing a balanced measure of model performance that accounts for both false positives and false negatives. An F1-score of 0.72 indicates reasonable but not excellent performance in balancing these metrics.

Question 10 of 15

You need to detect anomalies in manufacturing sensor data using BigQuery ML. Which model type is most appropriate?

LOGISTIC_REG AUTOENCODER BOOSTED_TREE_CLASSIFIER KMEANS with k=2

Explanation

AUTOENCODER models in BigQuery ML are specifically designed for anomaly detection. They learn to reconstruct normal patterns, and data points with high reconstruction error are flagged as anomalies. This is more effective than classification or clustering for unsupervised anomaly detection.

Question 11 of 15

You're building a model to predict house prices (continuous value) and want the best performance. Which BigQuery ML model type typically provides the highest accuracy for regression tasks with complex, non-linear relationships?

LINEAR_REG BOOSTED_TREE_REGRESSOR ARIMA_PLUS KMEANS

Explanation

BOOSTED_TREE_REGRESSOR typically provides the best performance for regression tasks with complex, non-linear relationships. It uses gradient boosting to combine multiple decision trees, capturing complex patterns better than linear regression. LINEAR_REG assumes linear relationships, and the other options aren't designed for this task.

Question 12 of 15

You trained a BigQuery ML model and want to deploy it for online predictions with low latency (< 100ms). What is the recommended approach?

Query ML.PREDICT directly from BigQuery for each prediction request Export the model to Vertex AI for online prediction serving Use BigQuery's streaming API with ML.PREDICT Create a Cloud Function that queries BigQuery ML.PREDICT

Explanation

For low-latency online predictions (< 100ms), you should export the BigQuery ML model to Vertex AI for online prediction serving. Direct BigQuery ML.PREDICT queries typically have higher latency suitable for batch predictions but not for real-time serving requirements.

Question 13 of 15

You're using BigQuery ML to classify customer support tickets into categories. You have 50,000 training examples and want to use deep learning. Which model type should you choose?

DNN_CLASSIFIER LOGISTIC_REG BOOSTED_TREE_CLASSIFIER AUTOML_CLASSIFIER

Explanation

DNN_CLASSIFIER (Deep Neural Network Classifier) is the appropriate choice when you want to use deep learning for classification in BigQuery ML. With 50,000 training examples, you have sufficient data for a DNN. AUTOML_CLASSIFIER is not a valid BigQuery ML model type.

Question 14 of 15

When evaluating a multi-class classification model in BigQuery ML, you notice the model performs well overall but poorly on one specific class. Which technique should you consider?

Increase the learning rate Use class weighting or oversampling for the underrepresented class Switch to binary classification Reduce the number of features

Explanation

When a model performs poorly on a specific class, it's often due to class imbalance. Using class weighting (via class_weight option in BigQuery ML) or oversampling the underrepresented class helps the model pay more attention to that class during training, improving its performance.

Question 15 of 15

You need to perform feature selection in BigQuery ML to reduce model complexity and improve interpretability. Which approach is recommended?

Use ML.FEATURE_IMPORTANCE to identify and select the most important features Manually remove features one by one and retrain Use PCA (Principal Component Analysis) to reduce dimensions BigQuery ML does not support feature selection

Explanation

ML.FEATURE_IMPORTANCE in BigQuery ML allows you to evaluate which features contribute most to your model's predictions. You can use these insights to select the most important features and retrain with a reduced feature set, improving model interpretability and potentially performance.

Welcome Back

Terms of Service & Privacy Notice

Create Account

Verify Your Phone Number

Reset Your Password

Login Required

Get in Touch

Developing ML Models by Using BigQuery ML