The goal of a modeling
project is to identify a champion model that a scoring application
uses to predict an outcome. SAS Model Manager provides tools to evaluate
candidate models, declare champion models, and inform your scoring
officer that a predictive model is ready for validation or production.
To deploy a model from
SAS Model Manager, you might use the following scenario:
-
Identify the model that
outperforms other candidate models and declare this model to be a
champion model. You can also flag challenger models for the champion
model.
-
Test and validate the
model before you declare the model ready for production.
-
Freeze the model version
to prevent changes to the model.
-
Update your life cycle
milestones.
-
Publish the champion
model and challenger models (optional) so that you can deploy them
to a production environment.
The following figure
illustrates activities that might occur before a model is deployed.
Note: Prior to SAS Model Manager
12.1, all versions of a project could contain a champion model, and
one of the tasks in deploying models was to set the version that contained
the champion model as the default version for a project. Beginning
in Model Manager 12.1, when you set a champion model, SAS Model Manager
sets the version that contains the champion model as the default version.
A project can have only one champion model. Therefore, the task of
setting a default version has been eliminated.