Prodigy NER model active learning

Hi Ines,
This is Yan, I work for Amazon, I took a look at your video about Active Learning with Prodigy for NER model which could be very helpful for our project.
is there any way I can use the prodigy interface but don't use Spacy's ner model, but have my own more complex NER model?
if that is doable, how does prodigy charge the money if I want to use it on business?
Thank you so much your help.

Sure! Prodigy is fully scriptable, so if you can load your custom model in Python, you can use it in Prodigy. Check out the docs on plugging in custom models for NER: https://prodi.gy/docs/named-entity-recognition#custom-model

If you want to update the model in the loop, you can use the recipe's update callback, which receives batches of answers as they're sent back to the server. You can read more about custom recipes here: https://prodi.gy/docs/custom-recipes

The only thing to keep in mind is that if you want to update your model in the loop, it should be sensitive enough to small single-batch updates (which isn't always what model implementations are optimised for). The active learning workflows only make sense if you can see the updates reflected in the predictions quickly.

Prodigy is a downloadable tool and you pay for a lifetime license, including 12 months of free upgrades. A company pack includes 5 floating seats, so you and up to 4 other team members could run the app and back-end at the same time. See here for more info and pricing: https://prodi.gy/buy The easiest way to order is directly from our online store, but of course we also accept purchase orders via email.

hey Ines, that is great information! Thank you so much! I will do experiment to see if this tool fits my use cases.

1 Like