Champion Models :: SAS(R) Model Manager 13.1: User's Guide

About Champion Models

The champion model is the best predictive model that is chosen from a pool of candidate models. Before you identify the champion model, you can evaluate the structure, performance, and resilience of candidate models. When a champion model is ready for production scoring, you set the model as the champion model. The version that contains the champion model becomes the champion version for the project. A check mark Set champion

appears next to the version on the Versions page. You can publish the champion model to a database, the SAS Metadata Repository, and a SAS channel.

Requirements for a Champion Model

Before you identify a model as the champion, perform the following tasks:

Create a version for your project, and register at least one model.
Verify that the model is active. If the model expiration date has passed, you cannot set the model as a champion model.

Note: An authorized user can reset the expiration date to a later date so that it is possible to set the champion model. To reset the expiration date, select the Properties page for the model.

You might use the following criteria to identify a champion model:

model comparison reports that validate and assess the candidate models
business decision rules. For example, you might use a decision tree model because of difficulty interpreting results from a neural network model even when the neural network model outperforms the decision tree model
regulatory requirements, such as when the champion model should exclude certain specific attributes (age or race)

You can flag and publish a challenger model specifically for the purpose of comparison with the champion model. For example, your champion model for a production environment might omit restricted attributes during operational scoring because of regulatory requirements. You can use a challenger model that includes the restricted attributes in the development environment to evaluate its prediction power against the prediction of the champion model. Then you can determine the amount of predictive power that is lost because of the regulatory requirements.

Set a Champion Model

To set a champion model:

On the Models page of a project, select a model and click to set the model as the project champion model. The value in the Role column changes to Champion.

Note: Alternatively, you can right-click a model and select Set as champion.

Clear a Champion Model

To clear a champion model:

On the Models page of a project, select a model that is marked as Champion, and click to clear a flagged champion model.

Note: Alternatively, you can right-click a model that is marked as Champion and select Clear.