Preparing and using data for analysis and ML
Free Google Cloud Certified - Professional Data Engineer practice — 6 questions on Preparing and using data for analysis and ML, with explanations. No sign-up. Full 12-question mixed test →
A retail company wants to build an image classification model to detect defective products from photos taken on an assembly line. They have 500,000 labeled images stored in Cloud Storage. Which approach BEST meets this requirement?
Vertex AI natively supports image classification through AutoML Vision or custom training jobs using pretrained CNN architectures, which is the correct tool for complex unstructured image workloads at this scale.
A data engineer creates a BigQuery ML model with the following statement:
CREATE MODEL mydataset.my_model
TRANSFORM(
ML.STANDARD_SCALER(feature1) OVER() AS scaled_feature1,
label
)
OPTIONS(model_type='linear_reg', input_label_cols=['label']) AS
SELECT feature1, label FROM mydataset.training_table;
What is the primary benefit of using the TRANSFORM clause here?
The TRANSFORM clause captures preprocessing logic inside the model definition, so ML.PREDICT automatically re-applies the exact same transformation to new input data, guaranteeing training/serving consistency and reproducibility.
A machine learning team trains fraud-detection models in Vertex AI using features computed daily in BigQuery (for example, rolling 7-day transaction counts per user). At serving time, an online fraud-scoring API needs these same feature values with p99 latency under 10 milliseconds. Which solution should they implement?
Vertex AI Feature Store is purpose-built to bridge batch-computed offline features (e.g., from BigQuery) with a low-latency online serving store, enabling millisecond-level key lookups required for real-time inference.
A healthcare analytics company wants to share a curated, continuously updated BigQuery dataset with 12 external hospital partners so each partner can run their own queries, without giving them direct access to the source GCP project or physically copying the data into each partner's project. Which approach is correct?
Analytics Hub is designed for exactly this scenario: a publisher exposes a listing, and subscribers create read-only linked datasets in their own projects that stay continuously in sync with the source data, without direct project access or data duplication.
Which statement about the BigQuery ML ARIMA_PLUS model type is correct?
ARIMA_PLUS automatically detects seasonal patterns, handles missing and irregular time points, performs anomaly detection, and can adjust for holiday effects via the holiday_region option, minimizing manual feature engineering.
An executive dashboard built in Looker Studio, connected directly to a BigQuery table via a live connection, consistently shows data that is 12 hours stale, even though the underlying table is updated every 15 minutes by a Dataflow pipeline. Viewers are not manually refreshing the page. What is the MOST likely cause?
Looker Studio caches query results according to a configurable data freshness setting at the data source or chart level; if that setting is configured for a 12-hour interval, viewers will see stale data regardless of how frequently the underlying BigQuery table updates.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$129.99 $34.99 with code FREETEST33 — valid through September 16.
Get my $34.99 deal →