Using Labeling Functions for multi-class classification problems

[stenpiren]

I have seen the tutorial on the spam classification where among other things, it is taught how to use labeling and transformation functions. But I see a limitation of the features snorkel provides when it comes to multi-class classification, as this example is strictly meant for a binary classification problem.

Isn't it better to include more features or perharps add an example where we can see the use of snorkel in order to label more than 3 classes.

For instance I have a problem involving 20 classes. For what I see as work around is to use labeling functions per each class, for instance treating each class one at a time as if it is a binary case.

The best would be to tackle all in one go, any suggestion besides the one I am thinking about?

[brahmaneya]

Hi @stenpiren, you can check out the visual_relation tutorial, which deals with a non-binary problem.

[stenpiren]

It would have been interesting to see one similar to spam/ham problem but with multiclass. The visual_relation want is a bit complicated as it involves text and image.

[github-actions[bot]]

This issue is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 7 days.