Overview of Retraining Models
You can retrain models to respond to data and market changes. Retraining models enables
you to update out-of-date models and improve model performance. When you edit a model retrain definition, you can select multiple models to be retrained at the same time. The retrain definition
for a model includes the destination version and training data source. The destination version is an existing version or new version that is associated
with the selected project and stores the retrained model information.
The training data source contains new data for retraining the selected models. You
can also specify a location
to store the comparison reports and retrain results. When you select the models to include in the comparison report, you can use the training data source or select a different data source to compare the performance of the new models. You
can also specify the report options, including the name, format, and style of the comparison report. Email notifications can also be specified as
part of a model retrain definition and are sent after you execute a model retrain definition.
By default, the champion model for the selected project is selected for retrain. If the Register
new trained model option was selected after you execute a model retrain definition, the new models
are registered to the destination version. The comparison report is available on the
Results tab
of the Retrain page. The definition is executed
on the SAS Application Server that is specified. The report folder
is stored on the SAS Content Server.
Note: Only R models and those that
were created with SAS Enterprise Miner, SAS/STAT, and SAS/ETS can
be retrained. SAS Factory Miner models and SAS Viya models that were
created with SAS Visual Data Mining and Machine Learning modeling
procedures cannot be retrained in SAS Model Manager.
Also note this: if a SAS model package (SPK) file was created with
the Model Comparison node in SAS Enterprise
Miner, the SPK file contains the score code for the best model and
the training code for all of the models that were part of the model
comparison. Therefore, when you are retraining the model in SAS Model
Manager, be aware that the algorithm that is used for the retrained
model could be different.
Copyright © SAS Institute Inc. All Rights Reserved.
Last updated: June 12, 2017