Add Hinge metric

[ethanwharris]

Before submitting

• Was this discussed/approved 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?

What does this PR do?

Fixes #89 - added the multiclass_mode parameter to enable computation in a one-vs-all fashion (e.g. for multi-class SVMs)

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 🙃

[codecov]

Codecov Report

Merging #120 (bdae664) into master (26eae39) will decrease coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #120      +/-   ##
==========================================
- Coverage   96.38%   96.35%   -0.03%     
==========================================
  Files         140       72      -68     
  Lines        4420     2280    -2140     
==========================================
- Hits         4260     2197    -2063     
+ Misses        160       83      -77     

Flag	Coverage Δ
Linux	79.07% <88.57%> (+0.30%)	⬆️
Windows	79.07% <88.57%> (+0.30%)	⬆️
cpu	96.35% <100.00%> (+0.11%)	⬆️
gpu	?
macOS	96.35% <100.00%> (+0.11%)	⬆️
pytest	96.35% <100.00%> (-0.03%)	⬇️
python3.6	96.35% <100.00%> (+0.11%)	⬆️
python3.8	96.35% <100.00%> (+0.11%)	⬆️
python3.9	96.35% <100.00%> (+0.11%)	⬆️
torch1.3.1	95.12% <100.00%> (+0.15%)	⬆️
torch1.4.0	95.26% <100.00%> (+0.15%)	⬆️
torch1.8.1	96.35% <100.00%> (+0.11%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
torchmetrics/__init__.py	100.00% <ø> (ø)
torchmetrics/classification/__init__.py	100.00% <100.00%> (ø)
torchmetrics/classification/hinge.py	100.00% <100.00%> (ø)
torchmetrics/functional/__init__.py	100.00% <100.00%> (ø)
torchmetrics/functional/classification/__init__.py	100.00% <100.00%> (ø)
torchmetrics/functional/classification/hinge.py	100.00% <100.00%> (ø)
__w/1/s/torchmetrics/utilities/enums.py
__w/1/s/torchmetrics/info.py
...etrics/functional/regression/mean_squared_error.py
.../1/s/torchmetrics/functional/retrieval/__init__.py
... and 68 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 26eae39...bdae664. Read the comment docs.

[Borda]

I am not very sure about adding anything named as Loss, could we have it consistent with others?

[ethanwharris]

@Borda Happy to change if needed 😃 although not sure what else to call it. I guess the options are:

• hinge_loss (consistent with sklearn)
• hinge (avoids "loss", looks like Keras does this)
• it's sometimes also called the max-margin objective, so maybe that?

Any other suggestions @SkafteNicki @justusschock ?

[ethanwharris]

@Borda @SkafteNicki Thanks for the comments 😃 I've added check for multiclass_mode and switched hinge loss to just hinge so should all be ready for review now.

[Borda]

very nice, almost there, just last comments from me :]

[pep8speaks]

Hello @ethanwharris! Thanks for updating this PR.

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2021-03-25 19:51:13 UTC

[SkafteNicki]

smaller comments else LGTM

[ethanwharris]

@PyTorchLightning/core-metrics I was having a weird issue here where the test was hanging forever rather than reporting an error with ddp and dist_sync_on_step both True. Managed to fix the bug so all works now, but ideally it would report the test failure rather than hang. Is this a known issue?