Edit a Model Retrain Definition

To define a model retrain definition:
  1. In the Definition tab on the Retrain page, click Edit Definition and select one or more models to retrain. By default, the champion model is selected if it can be retrained.
    Retrain Definition Step 1
  2. Select a data processing method.
    • To run a standard environment, select Standard configuration.
    • To run the performance monitoring definition in a High-Performance Analytics environment, select High-performance configuration.
  3. (Optional) Select Register new trained model to register the new models in the destination version on the SAS Content Server.
  4. (Optional) If Register new trained model is checked, select a destination version for new models. Select New version from the drop-down menu to create a new version for the models.
  5. Click Browse to select a training data source from a library. Click OK.
  6. Click the SAS Application Server list and select a server.
  7. Click Browse to select a report folder in which to store the comparison report.
  8. Click Browse to select a retrain results folder to store the model training results.
  9. (Optional) Select Trace on to print trace information to the SAS log file.
  10. (Optional) Select Retrain when the dashboard project status is Alert or Warning. If the dashboard project status is Alert, the model is automatically retrained. If the dashboard project status is Warning, select whether to retrain the model or ignore the task. If the dashboard project status is Normal, the model will not be retrained.
  11. Click Next.
    Retrain Definition Step 2
  12. Select the models to compare.
    Note: If you do not select a model, the champion model is used to perform the comparison.
  13. Specify the data source options:
    • Select Use training data source to use the whole training data source to compare or partition it into two parts, based on partition percent and random seed. The percent that is specified is the percentage of data that is used for model comparison; the other part of the data is used for training. The random seed value is used to generate the training data based on the random sampling method.
    • Click Browse to select a performance data set as the comparison data source.
  14. Specify the report options:
    • Enter a report name.
    • Select a format for the report output. The standard formats that are available are RTF, PDF, HTML, and EXCEL. The default is RTF.
    • Select a style for the report. The available styles are SAS default, Seaside, Meadow, and Harvest. The default is SAS default.
  15. Click Next.
  16. (Optional) To send the retrain results by e-mail, click Add and enter an e-mail address.
  17. Click Save.
Copyright © SAS Institute Inc. All rights reserved.