I created a dataset with multiple annotators using the textcat.manual recipe with --exclusive parameter set to false. Now when I use the prodigy review recipe to resolve conflicts, the prodigy review recipe forces me to annotate with mutual exclusive labels. Is this a bug that I can't choose to use or not use mutually exclusive labels or is this expected behavior?
Kind Regards.
Hi! The choice_style config setting is a recipe setting, so it's not stored in the data. If you're using the review recipe, an easy solution is to just put a prodigy.json in your project working directory that sets "choice_style": "multiple". This will set the choice style for the choice interface.
v1.10 will allow overwriting config settings via the "config" of individual tasks – so we could consider having the choice interface preserve the choice style on the task, so it can be picked up again when you reannotate the data.
Thank you Ines, for the workaround. It worked for me. I agree that your v1.10 solution will be an improvement.
Just released Prodigy v1.10, which now supports config options on individual annotation tasks. The choice interface now stores the "choice_style" with each task, so the style will be set automatically when you run review (and the information will be preserved with the annotation).
1 Like