Analysis Model Configurations

  • Article

An analysis model configuration contains information used to build Prediction and Segment models. The Predictor resource uses analysis model configurations to map data from the Commerce Server Data Warehouse database to cases. A case is the set of information known about a specific user and consists of a set of attribute/value pairs. Cases are used to build analysis models, which in turn, are used by the Predictor Client Object to make predictions about the future behavior of users.

An analysis model configuration specifies:

  • A collection of tables and/or views in the Data Warehouse that contains the data to be modeled.

  • Attributes for the model. An attribute is a useful distinction about a user — for example, the age of a user, whether a user has purchased a particular product, or whether a user has clicked on a particular ad.

Commerce Server 2000 provides a Transactions model configuration you can use to build Prediction and Segment models to analyze transaction data from the Data Warehouse. If you want to build Prediction and Segment models based on other types of data, you must create new model configurations with the Predictor schema. For information about how to build a new analysis model configuration, see Predictor Schema.

You can build your own analysis model configuration manually for any of the following reasons:

  • The field names in your tables (or views) differ from those in the built-in tables.

  • You do not use the built-in data tables included with Commerce Server.

  • The model configurations provided do not meet your business needs.

The Transactions model configuration provided by Commerce Server is maintained in the Data Warehouse database in three main tables and in an optional fourth table:

  • PredictorModelCfgs. Lists the existing model configuration. This model configuration is included under Model Configurations in Commerce Server Manager.

  • PredictorDataTables. Lists and describes the tables containing the data to be modeled.

  • PredictorModels. Stores the actual analysis models and the parameters that were used to build them.

  • PredictorAttributes <configuration name>. Indicates which properties are to be used to build the model or which properties you want to be predicted. This table is optional and can be used to override the automatic assignment of attribute distributions.

For information about viewing the model configuration tables using Microsoft SQL Server, see Viewing Analysis Model Configuration Tables.


All rights reserved.