textcat.manual binary annotation without labels


I have a pre-annotated dataset. Each document in the .jsonl file has a different label. I have been able to set up a system where the pre-annotated label is in the banner above the sentence/document, but it required me to have a list of dropdown or checkbox labels that I need to declare as an argument on the command line. Is there a way to just have the banner with the pre-annotated label (which is different per document) and the accept/reject button without the --labels requirement on the command line?

I tried to use mark with a list of labels, as I just want the label as a banner with the sentence, and the accept and reject buttons. Each document is pre-annotated and has its own label taken from a list of ~100 labels. Example from .jsonl

{"id":"1234","deployment":"xxxx","date":"2021-10-20 12:46:00","text":"random text here","label":"label1","orig_accept":["label1"]}
{"id":"1234","deployment":"xxxx","date":"2021-10-20 12:46:00","text":"random text here","label":"label2","orig_accept":["label2"]}

Thank you!

I may have solved this by using the following command:

PRODIGY_ALLOWED_SESSIONS=max,sam prodigy mark testset_name /<path_to_file>/file.jsonl --view-id classification

1 Like

Yes, that's a good solution. If you already have pre-annotated labels, all you need to do is render the exact data that comes in.

1 Like