Vertex AI and BigQuery ML

ML platforms close to the data

Vertex AI is Google Cloud's machine-learning platform: training and tuning custom models, managed endpoints for online prediction, batch prediction, pipelines, a feature store and model monitoring. Its Model Garden and generative AI APIs give access to Google's Gemini models and many open and third-party models, with features such as grounding and evaluation, all governed by the same IAM, VPC controls and data-location settings as the rest of your project. BigQuery ML lets analysts create and use models with SQL (CREATE MODEL, ML.PREDICT) directly where the data lives, including calling remote models hosted on Vertex AI. For a student or a small team, the practical lesson is to start with managed APIs and BigQuery ML before building custom training infrastructure.

A model in BigQuery ML

Train a logistic regression to predict churn, then score current customers, all in SQL.

CREATE OR REPLACE MODEL shop.churn_model
OPTIONS (model_type = 'logistic_reg', input_label_cols = ['churned']) AS
SELECT days_since_last_order, orders_last_90d, avg_basket, support_tickets, churned
FROM shop.customer_features
WHERE snapshot_date = '2026-06-30';

SELECT customer_id, predicted_churned_probs
FROM ML.PREDICT(
  MODEL shop.churn_model,
  (SELECT * FROM shop.customer_features WHERE snapshot_date = CURRENT_DATE()));

Evaluate before you trust a model

Run ML.EVALUATE and compare against a simple baseline (for example "customers with no order in 60 days churn"). A model that cannot beat a one-line rule is not ready for production.

त्वरित जाँच: An analyst wants to train a simple model without moving data out of the warehouse. What should they use?

• BigQuery ML
• Cloud Tasks
• Memorystore
• Cloud NAT