What's changed: Created Associate Data Practitioner Chapter 2 (Domain 2 "Analysis and presentation": analysis with BigQuery and notebooks = SQL/Colab Enterprise; dashboards with Looker = Looker/Looker Studio/LookML; using ML models = BigQuery ML/AutoML/LLM via remote connection/standard ML project/Model Registry).
2.3Using machine learning models
Understand creating ML models with BigQuery ML and AutoML, using pretrained large language models (LLMs) via remote connections in BigQuery, the steps of a standard ML project (data collection, training, evaluation, prediction), and organizing models in the Model Registry.
Data practitioners also use machine learning for prediction and classification. Even without being a specialist data scientist, you can apply ML via SQL and automated tooling.
2.3.1BigQuery ML and AutoML
BigQuery ML creates, trains, and evaluates ML models and runs predictions (inference) with just SQL, without moving data in BigQuery (e.g., demand forecasting, classification). AutoML builds models automatically with little code, making high-quality models accessible without deep ML expertise. Additionally, via a BigQuery remote connection you can call Google pretrained large language models (LLMs) to do text generation or summarization within SQL. Map "ML in SQL = BigQuery ML" and "automatic model building = AutoML."
2.3.2ML project and Model Registry
A standard ML project flows as data collection → training → evaluation → prediction (inference). Evaluation measures accuracy and more; if sufficient, the model is used for prediction in production. Organize and manage created models in the Model Registry to oversee versions and uses. Map "centrally manage models = Model Registry."
Common: requirement → means. E.g., "ML with SQL on BigQuery data without moving it" = BigQuery ML; "build a model automatically without code" = AutoML; "call a pretrained LLM from SQL" = LLM via remote connection; "centrally manage models" = Model Registry; "collect → train → evaluate → predict" = standard ML project.
Watch the mix-ups: (1) BigQuery ML (ML in SQL) vs AutoML (automatic build) differ in approach. (2) Pretrained LLMs are called from BigQuery via a remote connection. (3) An ML project uses evaluation before prediction (do not skip evaluation).
2.3.3Section summary
- BigQuery ML = ML in SQL without moving data; AutoML = automatic model building without code
- Call pretrained LLMs via a BigQuery remote connection
- Standard ML project = collect → train → evaluate → predict; manage models in the Model Registry
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. To create/predict ML models with just SQL on BigQuery data without moving it, which is best?
Q2. With little ML expertise, to automatically build high-quality models with little code, which is best?
Q3. To call Google pretrained LLMs from within BigQuery for text processing, what do you use?
Q4. To organize and manage created ML models with versions and uses, what do you use?
Q5. Which is the correct flow of a standard machine learning project?
Q6. Which correctly contrasts BigQuery ML and AutoML?