Descriptions of Project Tables

A project control table is a data set that contains the projects, models, and segments that are used to create the structure of the projects within a portfolio. The project control table must at least contain a project variable with the name of project_name. If you want to monitor the performance of the champion models within a portfolio, then the project control table must also contain a segment ID variable. The segment ID variable must also be in the performance tables that are used to monitor performance. If you want to include the models for each project when creating a portfolio, then the control table must also contain the model variable.

A project input table is an optional SAS data set that contains the champion model input variables and their attributes. It is a prototype table that can be used to define the project input variables and the variable attributes such as data type and length. A project can have numerous candidate models that use different predictor variables as input. Because the project input table must contain all champion model input variables, the variables in the project input table are a super set of all input variables that any candidate model in the project might use.

A project input table can have one or more observations. Data that is in a project input table is not used by SAS Model Manager.

If you use a prototype table to define the project input variables, either create the table and register the table to the SAS Metadata Repository in the Data category view or by using SAS Management Console. Tables that are registered using the SAS Management Console must be made available to SAS Model Manager using the Data category view.

The project input variables must be available to SAS Model Manager either by specifying a project input table or by defining individual variables before you set a champion model. You can view input variables for a project on the Input tab of the project’s Variables page, or in the Data category view.

A project output table is an optional SAS data set or database table that defines project output variables and variable attributes such as data type and length. It is a prototype table that contains a subset of the output variables that any model in the project might create.

A project output table can have one or more observations. Data that is in a project output table is not used by SAS Model Manager.

If you use a prototype table to define the project output variables, either create the table and register the table to the SAS Metadata Repository in the Data category view or by using SAS Management Console. Tables that are registered using the SAS Management Console must be made available to SAS Model Manager using the Data category view. For more information, seeManaging Data Tables.

The project output variables must be available to SAS Model Manager either by specifying a project output table or by defining individual variables before you set a champion model. You can view output variables for a project on the Output tab of the project’s Variables page or in the Data category view.

A scoring input table is a SAS data set that contains the input data that is used in a scoring test.

Before you can create a scoring test, you must create a scoring input table and register it in the SAS Metadata Repository in the Data category view or by using SAS Management Console. Tables that are registered using the SAS Management Console must be made available to SAS Model Manager using the Data category view. In SAS Model Manager, you can view scoring input tables in the Data category view.

A scoring output table is used by a scoring test to define the variables for the scoring results table.

Depending on the mode in which a scoring test is run, the scoring output table can be a prototype table or a physical data table. A scoring test can run in test mode, which is the default mode, or it can run in production mode. In both test mode and production mode, a scoring test output table is used by the scoring test to define the structure of the scoring results table. When the scoring test runs, it creates a scoring results table. In test mode, the scoring results table is stored in the SAS Model Manager model repository or on a local or network drive. You can view the scoring results table on the Results tab of the Scoring page for a project. The scoring output table in the SAS Metadata Repository or on a local or network drive is not updated in test mode. In production mode, the contents of the scoring output table in the SAS Metadata Repository or the local or network drive are replaced by the contents of the scoring results table. The scoring results table is not stored in the SAS Model Manager model repository or on a local or network drive.

Before you can create a scoring test, the scoring output table must be added and accessible from the Data category view. To add the scoring output table to SAS Model Manager, perform one the following actions:

You can view scoring output tables in the Data category view.

A train table is used to build predictive models. Whether your predictive models are created using SAS Enterprise Miner or you created SAS code models, you used a train table to build your predictive model. SAS Model Manager uses this same train table. The train table must be registered in the SAS Metadata Repository and accessible to SAS Model Manager in the Data category view.

You specify a train table as a version-level property. When you define the train table at the version level, the table can be used to build all predictive models that are defined on the Models page for a project.

In SAS Model Manager, train tables are used for information purposes only with one exception. SAS Model Manager uses train tables to validate scoring results immediately after you publish a scoring function or model scoring files, and if the Validate scoring results box is selected when you publish scoring functions or model scoring files to a database.

For information about registering a train table using the Data category view, see Managing Data Tables.

A test table is used to create the Dynamic Lift report and the Interval Target Variable report that can be used to identify the champion model. Test tables are typically a subset of a train table, and they are identical in table structure to the corresponding train table. Update test tables by creating a new subset of the corresponding train table.

To view test tables in SAS Model Manager, the tables must be registered in the SAS Metadata Repository. In SAS Model Manager, you can view test tables in the Data category view.

After a test table is added to SAS Model Manager, you can specify the table in the Default test table field in the project properties.

For information about registering test tables using the Data category view, see Managing Data Tables.

A performance table is a SAS data set that is used as the input table for each SAS Model Manager performance definition. A performance definition is used to monitor a champion model's performance by comparing the observed target variable values with the predicted target variable values. A performance table is a sampling of operational data that is taken at a single point in time. Each time you run a performance definition, you use a new performance table to take a new sampling of the operational data. For example, a champion model is deployed to a production environment for the first time in March 2013. You might want to take a new sampling of the operational data in June 2013, September 2013, and January 2014. These new tables are performance tables in the context of SAS Model Manager.

To view a performance table in SAS Model Manager, you must register the tables in the SAS Metadata Repository using the Data category view or by using SAS Management Console. You can view performance tables in the Data category view. After a performance table is registered, you can specify the table in the Default performance table field in the project properties. The default performance table value at the project level is the default value for the Performance data source field in the Edit Performance Definition wizard.