update to Prodigy 1.8 and spaCy 2.1

I have used the version 1.7.1 with spaCy 2.0.18 for defining a model for customized NER. It works almost good as I have been asked I want to remove one tag and add this tag by regex and update model. I really like your new interface, do you think that should I update both spaCy and prodigy? I mean, the data that I provided, is consistent with these updates?

how can I update prodigy, (I do not have access to a new link) can I use the last link for updating in order to have the last update?

many many thanks

Hi! The download link never changes, so if you're still within your update period, you can just use the existing link and download the latest version from there :slightly_smiling_face:

I'd suggest to just try it out in a separate virtual environment. Both environments will read from the same prodigy.json and database, so you'll be able to run the same experiments. And if you need to test something with the old version, it's all still there. You can even back up your config and database if you want to be 100% safe.

You should still be able to use the same data in the latest Prodigy and spaCy and everything should be compatible. The only known problem that can happen in some specific cases is this one, which should be easy to fix – see thread for details: ner.batch-train after ner.maual results error (Value error : [E024]) Basically, in spaCy v2.1, it's now "illegal" for the named entity recognizer to predict whitespace-only/newline-only spans as entities. This is nice, because it should give you better accuracy overall. But it also means that you need to remove those spans from your training data if they're in there.

1 Like

Hello, I have been working with the prodigy 1.7.1 version and spacy==2.0.18 and I would like to migrated to spacy 2.1.x and prodigy 1.8.3 but for my use case the models display better results on the 2.0.18, could exist a solution to emulate the hyperparameters from architecture 2.0.18 on the 2.1?
Best regards