Attribute prediction

There are many different types of label attributes (string, multiple choice, etc.). Does the attribute prediction also work for the type string, or only for multiple choice?

Also, when using multiple choice, is it possible to add choices at a later stage in the interface (or now an option in the ‘edit project’ view?

Hi Melvin!

So there is currently no prediction service for string attributes. However, I would love to learn more about how you use strings and what you would like to predict. If you want to give me some examples of what you are thinking about, you can email me at

For your second question, I believe you are asking if you can add additional choices after starting the annotation process and how that would affect predictions. If this is true, the answers are:

  1. Yes, you can add new choices as you go
  2. It will take some time for the model to train on those new choices though. At the moment, we retrain the attribute model after a 20% increase in annotated data. Depending on the scale of your project, this might be after doing a couple of annotations, or after several thousand. If your project is quite large already, you can also email me directly and I’ll have a look at how we can help.

Best regards,
Alex Wennman