spacy model predictions as training input on prodigy


I have a ner spacy model saved and I the test dataset that is used to measure the models performance. As a next step I want to save the the models annotations and use this data as the input for another ner model (its a long story byt I need to train the ner model using the models predicitions as the ground truth). I want to train this model using the prodigy cli but I am not sure how to save the predictions from my set so that prodigy can use it as training data.
How can I do this?

Hi! So just to make sure I understand the question correctly: you want to use one model to label your training data and then use the result to train another model? Do you want to review the annotations in Prodigy at all, or do you just want to train?

The ner.correct workflow lets you stream in the predictions by an existing model, correct them in the UI and save the results to a new dataset:

If you don't want to manually correct anything and just train, training via Prodigy wouldn't really make sense – instead, you can just write a script that extracts the model's predictions, save out the result and train with spaCy directly. (Prodigy's training workflows are really just a thin wrapper around spaCy.)


Thanks. Some very good points made. I'll use spacy directly then.