Hey everyone.
So I have a bunch of prompts unlabelled, wanted to know if there’s a way of using my NLU classifier to batch predict all of these so I could label, and review the results. Is this a good idea? I think it will only create bias but some people in my team think that’s not the case! Any ideas how to do/ is this a good idea?
Mm, it’s a little hacky but I would say it’s not necessarily a bad idea by I agree that bias could be a problem. Since it sounds like you’re planning on hand reviewing the output by hand though, I think it’s probably not too bad and may end up saving you some time.
The easiest way to do it would probably be to set up an NLU only server and use the /model/parse endpoint for your predictions.
Cool thanks for the response! I’ll try it !