Chapter 9. Machine Learning in BigQuery

Artificial intelligence (AI) is the domain of computer science focused on building computational systems that are capable of acting autonomously. Over the years, many different subfields have arisen in AI, but an approach that has proven successful in recent years has been the idea of using large datasets to train general-purpose models (such as decision trees and neural networks) that can solve complex problems with great accuracy.

Teaching a computer based on examples is called supervised machine learning, and it can be carried out in BigQuery with the data remaining in place. In this chapter, we look at how to solve a wide variety of machine learning problems using BigQuery ML. Even though machine learning can be carried out in BigQuery, being able to use powerful, industry-standard machine learning frameworks such as TensorFlow on the data in BigQuery can give us access to a much wider variety of machine learning models and components. Hence, in this chapter we also look at the connections that exist between BigQuery and full-fledged machine learning frameworks.

What Is Machine Learning?

If we have collected historical data (and what is a data warehouse for, if not precisely this?), and the historical data contains the correct answers (called the “label”), we can train machine learning models on this data to predict the outcome for cases where the label is not yet known. For example, if we have a historical dataset of actual sales figures, ...

Get Google BigQuery: The Definitive Guide now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.