ML Through Database Queries and UDFs

Abstract

This chapter explores various options for performing ML using a database system. Before we start, a valid question is why—at first glance, the relational model and query language seem to be a poor fit with ML, as most ML algorithms look very different from and oftentimes far more complicated than database queries. Thus, database systems have traditionally served as a data store for ML; the ML algorithm would pull the data out from the database, transform it into the appropriate format (e.g., matrices, tensors, or dataframes), and then analyze it using programs written in a different programming language. On the other hand, there are a number of compelling arguments for doing ML inside a database system.