i'm trying to retrain existing model of Textcat but this time i want to remove exclusive option ?!
Thanks for your question!
Can you provide a little of context on your use case?
Yes, I think it is possible.
Let's say you first started with mutually exclusive labels (
Then you decide you want to move to non-mutually exclusive labels so you then try
If you combine both datasets (either both used same data or combine them with
db-merge), you can use
prodigy train --textcat-multilabel on all of the labels. The only problem is that it'll assume for your exclusive labels. I don't think the opposite can be done -- combine both of their labels and try to train a mutually exclusive model, i.e.,
prodigy train --textcat.
Then you could use your multilabel model for new labels (e.g.,
textcat.correct). If you do this, perhaps you could pass through your original dataset without the exclusive labels, try to correct that so you add in additional labels that you missed.
Let me know if this works or if you have other questions!