Large-scale in-database machine learning with pure SQL

A computer-implemented method to perform in-database machine learning, the method comprising:

• generating, by a relational database management system in response to one or more pure SQL queries contained in one or more driver scripts, a model table in the relational database that contains a weight for each unique feature associated with one or more of a plurality of examples included in a training data table included in the relational database;
• generating, by the relational database management system in response to one or more pure SQL queries contained in the one or more driver scripts, a score table in the relational database that contains a score for each of the plurality of examples;
• generating, by the relational database management system in response to one or more pure SQL queries contained in the one or more driver scripts, a single combined table from the training data table, the model table, and the score table;
• and
• determining, by the relational database management system in response to one or more pure SQL queries contained in the one or more driver scripts, an updated weight for each unique feature in the model table, wherein the updated weight for each unique feature is based at least in part on application of an update function to each row of the single combined table that includes such unique feature.

One or more non-transitory computer-readable media that aggregately store one or more driver scripts that contain a set of pure SQL queries that, when executed by one or more processors of a computer system, cause the computer system to perform operations, the operations comprising:

• accessing a current model table in a relational database, the current model table containing a current weight for each of a plurality of unique features;
• updating a scores table in the relational database, the scores table containing a score for each of a plurality of examples;
• joining the scores table and the current model table with a single combined table in the relational database;
• and
• generating a new model table in the relational database with an updated weight for each of the plurality of unique features, the updated weight for each unique feature computed by application of an update function to each row of the single combined table that contains such unique feature.