Fix top_k for multiclass-f1score

[rittik9]

What does this PR do?

Fixes #1653

Before submitting

• Was this discussed/agreed via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

---

📚 Documentation preview 📚: https://torchmetrics--2839.org.readthedocs.build/en/2839/

[eneserdo]

Thanks for the effort. Imho, you could have handled the "refining" on directly preds tensor using something like this:

preds_topk = torch.argsort(preds, dim=-1, descending=True)[:, :top_k]
preds_top1 = preds_topk[:, 0]
preds=torch.where((target.view(-1, 1) == preds_topk).sum(dim=-1).bool(), target, preds_top1)

Which is more compact way of doing the same job. (Cloning and reshaping are omitted here)

Also, these changes will break the current tests for all top_k related classes/functions e.g. for recall, accuracy, f1, so on so forth. I think it is important to re-write these tests. Additionally, maybe for the topk accuracy you can take the scikit learn's top_k_accuracy_score as a reference.

[rittik9]

Thanks for your suggestions.I've noticed some of the tests have failed. I'm working on them. I am also comparing them with other library implementations. I'll keep updating here.

[lantiga]

Thanks @rittik9!