Stopping criteria for classification in prodigy

Lately, i wrote a binary classifier recipe, and make it avaiable to prodigy. According to a previous topic, the update function should return the loss value. Does prodigy take the returned loss value as a parameter to judge whether the whole task should be terminated or not?

Sort of. The loss value is supplied to the progress callback, which you can customize in the components dictionary returned from your recipe. The progress tracker is displayed in the web app.

The default progress function tries to predict how many updates remain until the loss will be at 0.