Multiple labels in the file for multi-class classification task

General API discussion
1

Hi,

I would like to know how to enlist multiple labels in the uploaded file for the multi-class classification?
Is this a correct approach?

{"text": "It is a great day today", "label":  ["positive", "clear_statement"]}
{"text": "It is a bad xxsks", "label": ["negative", "unclear_statement"]}

Thanks for your time to help me resolve this :slight_smile:

1 Like
3

We’ve generally been a little skittish on multi-label classification specifically because it doesn’t really work with the autoregressive nature of our models. The first label generated influences the second label generated and usually users want the labels considered independently of each other.

Instead, we usually suggest folks upload a file with the same texts and different sets of labels. in your case, you might have a file with just the positive/negative label and another with unclear/clear.

2 Likes
4

It looks like you’re doing sentiment analysis with dialog acts. That should be pretty easy. Let me do some tests and get back to you.

UPDATE: This is the approach I’ve used extensively for multiple labels. GPT-3 can even explain why. I use some REGEX to interpret the response.

image
image1911×696 61.8 KB

image
image1196×572 50.5 KB

1 Like
5

@daveshapautomator - I think you misinterpreted my question. I am not looking for this. :slight_smile:

6

Hi,

2 years later, I have the same issue than sbharati and I hope that the time or the team solved the problem :slight_smile:

@sbharati Did you find an other way to do what you wanted ?
@hallacy Do we still need to create severals fine-tuning models for multi label classification ?

1 Like