Logits and multilabel: Add support for threshold <= 0?

[leonweber]

🚀 Feature

Add support for logit input in multilabel classification

Motivation

The docs state that F1 accepts logits and supports multilabel. However, if one uses logits as input in multilabel classification with a NLL loss, the threshold cannot be set to its' natural value 0:

f1 = F1(num_classes=3, threshold=0.0)
ValueError: The `threshold` should be a float in the (0,1) interval, got 0.0

This makes logits input for F1 basically unusable in a multilabel task.

Pitch

It might be a good idea to (a) prominently point out in the docs that the threshold has to be adapted for logits input and (b) allow thresholds smaller than 0.

[github-actions]

Hi! thanks for your contribution!, great first issue!

[Borda]

@SkafteNicki I think we forgot this after #200

[SkafteNicki]

Yes, most definitely, going to send a fix soon :]