Editing or retraining a model

Edit an existing model's Matrix (edit feature columns, or change the training and test dataset), edit a model's name or description, or retrain a model using a new version of an engine or database.

  • To edit an untrained model's name or description:
    1. Click the model in the list to select it, then click Rename model or right-click and select Rename.
    2. In the dialog, edit the model Name and Description.
    3. Click Save.
  • To view a trained model, click a model in the list to select it, then click Edit model or right-click and select Edit.
    Note: The trained model's information is displayed on the Define Matrix, Populate Matrix and Train and Test tabs with the message "[icon] This model is read-only".

  • To edit an untrained model, click a model in the list to select it, then click Edit model or right-click and select Edit.

    You are taken to the Define Matrix, Populate Matrix, and Train and Test tabs. Any changes you make on these tabs are saved automatically.

    Note:
    • Once a model is trained, you cannot edit or rename it. Clone the original model to create an untrained copy you can edit or retrain. A model which is sent for training and fails is considered untrained, and can be edited.
    • Once a primary table is selected, it cannot be changed. You must create a new model.
  • To edit feature matrix columns, add and remove linked tables and attributes from the Feature Set on the Define Matrix tab.

    Detailed instructions can be found in Adding tables and attributes to a Matrix.

    As long as they are from the primary table, you can add or remove records from the list on the Populate Matrix tab to edit or expand the training and test datasets. Detailed instructions can be found in Adding records to a Matrix.

    If you want to train your model using records from a table other than the primary table, you will need to either:
    • Create a new model where the table containing those records is the primary.
    • Add or move those records to the primary table in Granta MI.
    Ansys do not recommend retraining failed models without editing. Failure to train is almost always caused by issues with the training data or Matrix definition (for example, the wrong transform or feature type has been selected by accident).
  • To retrain a model using an updated database or engine, you will need to Clone the original then Train and test the clone.
    Note: Engine version is saved as part of the model, but currently database and record version information is not. Include the database version in the model description for reference.