[torchmetrics] Difference between dice_score and f1_score

[digital-idiot]

In the torchmetrics what is the difference between f1_score and dice_score? Aren't they theoretically the same metric ^[1]? Looking at source of dice_score it appears to be using the formula directly whereas f1_score calls fbeta_score with beta=1.

Also, it is possible to pass beta value other than 1 to f1_score which is a bit confusing and redundant.

[1] https://chenriang.me/f1-equal-dice-coefficient.html

[SkafteNicki]

Hi @digital-idiot,
You are right that f1_score and dice_score calculates the same thing.
You are also right that beta can be set to something else than 1 in f1_score, which is a mistake.
We are going to do a major refactor of the classification package in the next release where we will probably remove dice_score (as f1_score is more general) and rectify the other mistake.
Thanks for bringing this to our attention :]