GCP-PMLE-Ext sample questions

Question 1

You are designing a demand forecasting solution for a global retail chain. The data resides in BigQuery and consists of 5 years of historical sales data across 50,000 SKUs. The business requires a solution that minimizes development time while handling seasonality and holiday effects automatically. You need to create a model that forecasts sales for the next 30 days. Which BigQuery ML statement should you execute?

Accepted Answer

CREATE MODEL `project.dataset.sales_forecast` OPTIONS(model_type='ARIMA_PLUS', time_series_timestamp_col='date', time_series_data_col='sales', time_series_id_col='sku_id', horizon=30) AS SELECT date, sales, sku_id FROM `project.dataset.sales_history`

Answer

CREATE MODEL `project.dataset.sales_forecast` OPTIONS(model_type='DNN_REGRESSOR', input_label_cols=['sales']) AS SELECT * FROM `project.dataset.sales_history`

Answer

CREATE MODEL `project.dataset.sales_forecast` OPTIONS(model_type='BOOSTED_TREE_REGRESSOR', input_label_cols=['sales']) AS SELECT * FROM `project.dataset.sales_history`

Answer

CREATE MODEL `project.dataset.sales_forecast` OPTIONS(model_type='LINEAR_REG', input_label_cols=['sales']) AS SELECT * FROM `project.dataset.sales_history`

Question 2

Your organization wants to build an internal knowledge base search application using Generative AI. The application must answer employee questions based *strictly* on internal PDF policy documents stored in Cloud Storage. External knowledge should be minimized to prevent hallucinations. You need to implement this using Vertex AI Agent Builder. Which configuration should you choose?

Accepted Answer

Create a Vertex AI Search app, create an unstructured data store, import the PDFs from Cloud Storage, and configure the search widget to use 'Answer' mode with the data store as the grounding source.

Answer

Use the Document AI API to parse the PDFs, store the text in BigQuery, and use BigQuery ML to generate embeddings for vector search.

Answer

Create a Vertex AI Search app, create a structured data store, import the PDFs, and enable the 'Enterprise Web Search' feature.

Answer

Create a Vertex AI Conversation app, configure a generative fallback, and fine-tune a Gemini Pro model using the PDF text.

Question 3

You are using BigQuery ML to build a propensity to purchase model. Your dataset contains a `user_age` column where null values indicate users who haven't provided their birthdate. You want to replace these nulls with the average age of known users during the training process, but you want to ensure the exact same mean value is used during prediction to avoid training-serving skew. Which approach should you use?

Accepted Answer

Use the `TRANSFORM` clause inside the `CREATE MODEL` statement: `TRANSFORM(IFNULL(user_age, ML.MEAN_IMPUTER(user_age) OVER()) AS user_age, ...)`.

Answer

Pre-calculate the mean age in a separate SQL query, update the table to fill nulls, and then run `CREATE MODEL`.

Answer

Handle the nulls in the client application code before sending data to BigQuery for training or prediction.

Answer

Use the `OPTIONS(impute_nulls=TRUE)` parameter in the `CREATE MODEL` statement.

Question 4

A financial services company needs to train a fraud detection model using a highly imbalanced dataset (0.1% fraud cases). They want to use Vertex AI AutoML for Tabular Data. Which configuration step is most critical to ensure the model learns effectively from the minority class?

Accepted Answer

Select the 'AUC-PR' (Area Under the Precision-Recall Curve) as the optimization objective during model training configuration.

Answer

Manually downsample the majority class in BigQuery before importing the dataset to Vertex AI.

Answer

Increase the budget node hours to the maximum allowable limit.

Answer

Use the 'RMSE' optimization objective to minimize the error variance.

Question 5

You are building a movie recommendation system using the Vertex AI Retail API. You have a catalog of 10,000 movies and user event data (views, purchases). You want to ensure that the recommendations include a healthy mix of popular movies and niche content to encourage discovery. Which configuration should you adjust?

Accepted Answer

Enable the 'diversity' optimization setting in the serving config.

Answer

Use the 'price reranking' control to boost lower-priced items.

Answer

Filter the recommendations to exclude items with more than 1000 views.

Answer

Train a separate custom matrix factorization model in BigQuery ML.

Question 6

You are processing 50 TB of raw log data daily stored in Cloud Storage. The logs need to be cleaned, transformed, and aggregated into features for a fraud detection model. The transformation logic involves complex session windowing (aggregating events within 15-minute user sessions). Which Google Cloud service and tool combination is optimal for this preprocessing pipeline?

Accepted Answer

Use Dataflow with Apache Beam's session windowing capabilities.

Answer

Use Cloud Functions triggered by file uploads to process logs individually.

Answer

Use Dataprep by Trifacta to visually design the transformation recipe.

Answer

Use BigQuery SQL with standard aggregations.

Google Cloud Professional Machine Learning Engineer Free Sample Questions

Question 1

Question 2

Question 3

Question 4

Question 5

Question 6

Register free for 6 more questions

GCP-PMLE-Ext Sample Questions

Register free for 6 more questions

Related practice tests