The model details behind Prodigy


I wonder what kind of models are used behind the scene. For example, the "insults" example from the video. Are we using a machine learning model or a deep learning model or a frequency model to train it? How should I know about the model details?

Thank you in advance!



Hi! Prodigy itself doesn't include any model implementations for the tasks you can train, but it integrates with spaCy out-of-the-box, and that's what we're using in most of our examples. The text classification video uses spaCy's text classifier with the default ensemble architecture:

You can also use custom recipes to integrate any other model, or export the annotations you collect manually and then train your model in a separate process.

1 Like