Skip to content

Running predictions against a model you already trained

BigQuery ML lets you generate predictions directly inside BigQuery using SQL, without moving data or provisioning serving infrastructure. The core mechanism is the ML.PREDICT function, which applies a trained (or imported) model to a table or query result and returns predictions alongside the original columns.

Must-know

  • ML.PREDICT is the standard syntax: SELECT * FROM ML.PREDICT(MODEL project.dataset.model_name, TABLE project.dataset.input_table), and it also accepts a subquery instead of a table.
  • Non-feature (passthrough) columns in the input are automatically preserved in the output alongside the prediction columns, so you don't need to manually rejoin results to source data.
  • For classification models, ML.PREDICT returns predicted_label and probability struct columns, and you can adjust the decision threshold using the threshold argument.
  • Time-series ARIMA_PLUS models use ML.FORECAST instead of ML.PREDICT to generate future forecasted values with confidence intervals.
  • ML.EXPLAIN_PREDICT returns the same predictions as ML.PREDICT plus feature attribution values, useful for explainability without retraining.
  • Inference works the same way for imported models (TensorFlow, ONNX, XGBoost) and remote models pointing to Vertex AI endpoints, and cost is based on bytes processed by the query, not a separate serving fee.
Check this objectiveFree · always available

A data analyst has trained a BigQuery ML logistic regression model named retail.churn_model on historical customer data that includes a churn label column. The analyst now has a new table, retail.active_customers, that does not contain a label column, and wants to generate a churn prediction for each row. Which query should the analyst run?

Your objective map0 tried · 0 right · 41 untouched

What you have tried across GCP ADP's objectives, not a readiness score.

Coverage checked against the published exam guide on Aug 12, 2026.

These are independent practice questions, written against this certification's published exam guide. They are not the certification vendor's own questions, and not the real exam.