Training the model incrementally

Use Incremental training to update saved models with the latest data increment while preserving the knowledge gained from the previous trainings.

Periodical model retraining is recommended in order to enhance the model accuracy with the latest data patterns. The incremental method lets you do this without the need to retrain the model from scratch on the entire dataset. This approach keeps the model up-to-date without the heavy resource and time costs.

Before you apply incremental training, ensure these prerequisites are met:

  • The initial model is trained using a custom algorithm.
  • The trained model is saved and accessible in the Models Library. Refer to the previous chapters for instructions on saving and managing trained models.
  • The latest data increment is imported as a separate dataset.
  1. Open the quest.
  2. Specify the incremental dataset in the Import Data activity.
  3. Add an Input Model activity from the catalog and connect it as the fourth input port of Train Model with Custom Algorithm activity.
    In the Train Model with Custom Algorithm activity, keep the Save Model option checked to overwrite the previously saved model with its latest version upon each activity execution. Unchecking the Save Model option stops saving newer versions of the model. Use this to preserve the last saved model when you are experimenting with the model settings.
  4. In the Input Model activity settings, select the previously saved model from the list.
  5. Click Save and run the quest.