-
Notifications
You must be signed in to change notification settings - Fork 9
Updating models
The incremental learning aspect of SAIL is utilized when you can update an existing model trained on older data on new labeled data of yours. The process is very easy and GUI based. The update process is as follows:
- Get the predictions from an existing model.
- Change the labels which were predicted incorrectly.
- Update the model using the new labels or an already exiting labeled data.
In order to change the label after doing a prediction do the following:
- Double click the label and it will show the list of possible labels.
- Choose the correct label.
- If you want to go back to annotating more data in future just click on Save
- If you want to update the model with your new data click Next
In order to update the model with the new data just click the Retrain button Once the model is updated it will be saved into the updates folder in your output folder. The GUI will also show how the model improved using 10 batches of the updated data. NOTE: In order to get significant improvement by model update, you should provide at least 100+ new labeled examples.
For making predictions using updated model just select the custom model file from your output folder on the model selection option and follow the other steps as usual.