Retraining a classification model

Retraining a classification model takes two personas to complete. The administrator must create a classification model for the data expert to prepare feedback. The administrator starts the retraining process and the data expert runs the enhancement to improve its accuracy.

Before you begin

With the assistance of IBM software support, modify the data-learning settings.
  1. Click Project > Project Options.
  2. In the Project Options window, click Advanced.
  3. Select Maintain learning data set with knowledge base. Accept all other default values.
  4. Click OK.

The data-learning settings can also be modified in the Create, Analyze, and Learn dialog box by clicking Settings. For more information, see http://www-01.ibm.com/support/knowledgecenter/SSBRAM_8.8.0/com.ibm.classify.workbench.doc/c_WBG_Saving_Learning_Data.htm.

Procedure

  1. Click Auto Classification > Create Model.
    You must set an attribute when you create a classification model.
  2. Enter the information to complete the fields and save it.
  3. At this point, the data expert needs to complete the following actions from IBM® StoredIQ® Data Workbench.
    • Create a user infoset.
    • Run the Step-up Snippet action.
    • Apply an enhancement.
    • Apply a filter.
    • View the objects in the Data Object Viewer.
    • Provide feedback for this data object.
    For detailed descriptions of these actions, see Retraining a classification model.
  4. In IBM StoredIQ Administrator, click Auto-Classification and select the classification model. The <model name> Details window shows that it received feedback. Click Retrain to start the retrain process for this model.
    If the model was uploaded without a learning archive (SARC file) or if it received no feedback, then the Retrain button is disabled.

    Next, the data expert runs the enhancement against the user infoset again from IBM StoredIQ Data Workbench to see the new scores within the Data Object Viewer. The improved scores indicate greater validity and accuracy.