About Partitioned Models

Partitioned models allow you to divide your data set into multiple partitions based on specific attributes and build a model for each partition. The system automates the creation and management of these models, reducing manual effort.

When you build a model on a data set, a single generic model may not perform well on new or evolving data. To address this, you can specify partition columns to create separate models for each partition based on some characteristics. For example, if your data set includes a "REGION" attribute with four values, four models will be created automatically, each corresponding to a region. The system manages these sub-models as part of a single partitioned model.

The partitioned model resides as a first-class, persistent database object, with all sub-models stored on disk. This structure improves performance, especially for models with a large number of partitions, and allows for efficient management, such as dropping individual sub-models when needed.

When building or scoring partitioned models, not all sub-models need to be loaded into memory simultaneously. This approach optimizes memory usage and enhances processing efficiency. The system provides a single model for scoring, while users can still access individual component models if needed.

See Also:

Oracle Machine Learning for SQL User’s Guide to understand more about partitioned models.