Retraining and Redploying the AI Model

After some incoming records are merged with existing records, or several existing records are merged, the pair collection process collects new pairs from these merge events. See Training an AI Model for ways to trigger this process and interpret its results.

The newly retrained model is automatically approved under the following conditions:

  • The AutoDeploy property is set to true in the LearnConfig.xml file.

  • The Learn Collect Pairs Schedule job is configured.

  • The AI Matching system determines that the new model performs better than the currently deployed model

If the model is automatically approved, the State of the Active project instance is set to “Approved” after the pair collection process completes. For information about the states of projects, see States of a Project.

If the AutoDeploy property is set to false in the LearnConfig.xml file, you must perform the action based on the state of the Active project instance.

Before you begin See Using the AI Model.
    Procedure
  1. In ibi MDM, click Administration > AI Matching.
  2. On the Projects tab, select the project.
  3. Click the Manage Selected Project tab, perform the following actions based on the Active State of the selected project:
    • Approve: Click Approve. For further steps, see Managing the AI Model.
    • Do not approve: Wait for more pairs to be collected, do not deploy the model. The model performs the same or slightly worse than the deployed model, or too few pairs have been collected.
    • Resolve issues: The model performs significantly worse than the deployed model, or an error occurred while collecting pairs or retraining the model. This state might indicate a problem. After exporting the project for editing, examine the errors made by the model and the pair labels to investigate and resolve performance issues.
Resultibi MDM loads trained and deployed models.