[REVIEW] Add cython hinge_loss

[Nanthini10]

Implements hinge_loss with tests. API matches and code is adapted from sklearn https://scikit-learn.org/stable/modules/generated/sklearn.metrics.hinge_loss.html

NOTE: We're not using the C++ hinge implementation in order to comply with sklearn API. The C++ implementation is used in the intermediate step in SGD and accept different set of parameters than what is used to calculate hinge_loss.

[wphicks]

Looks like this was hit by the IVFPQ error on CUDA 11 and network errors on 10.1 and 10.2. Going to rerun.

[wphicks]

rerun tests

[wphicks]

@Nanthini10 Could you merge branch-0.18 into your branch? We just got the xfail on IVFPQ in, and I think that's the only thing keeping your PR from passing. The other failures were timeouts on the clone.

[wphicks]

One small tweak for a redundant check, and one question (which I'm honestly not sure the answer to), but otherwise this looks great!

I know you mentioned it to me offline, but would you mind also adding a comment (either on the PR or in the code) that explains why we're not using the hinge_loss implemented in the CUDA/CPP layer? I figure having that there may help avoid confusion down the line about why this doesn't match the pattern of other metrics.

[mdemoret-nv]

Everything looks good to me. Well done.

[beckernick]

Should this be targeting 0.19?

[Nanthini10]

Should this be targeting 0.19?

Yes, this is for 0.19 @beckernick

[mdemoret-nv]

@Nanthini10 I changed the base to branch-0.19. Tests are running.

[codecov-io]

Codecov Report

Merging #3409 (57ea041) into branch-0.19 (39c7262) will increase coverage by 7.57%.
The diff coverage is 75.00%.

@@               Coverage Diff               @@
##           branch-0.19    #3409      +/-   ##
===============================================
+ Coverage        71.77%   79.35%   +7.57%     
===============================================
  Files              212      226      +14     
  Lines            17075    18007     +932     
===============================================
+ Hits             12256    14289    +2033     
+ Misses            4819     3718    -1101     

Flag	Coverage Δ
dask	44.01% <21.95%> (?)
non-dask	71.66% <73.10%> (?)

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

Impacted Files	Coverage Δ
...on/cuml/_thirdparty/sklearn/preprocessing/_data.py	63.61% <ø> (+0.07%)	⬆️
...l/_thirdparty/sklearn/preprocessing/_imputation.py	62.40% <ø> (-0.11%)	⬇️
python/cuml/experimental/explainer/common.py	88.05% <42.85%> (-4.01%)	⬇️
python/cuml/neighbors/nearest_neighbors.pyx	92.43% <50.00%> (-0.29%)	⬇️
python/cuml/common/import_utils.py	59.43% <66.66%> (+3.43%)	⬆️
python/cuml/experimental/explainer/base.pyx	67.06% <67.06%> (ø)
python/cuml/metrics/hinge_loss.pyx	73.33% <73.33%> (ø)
python/cuml/dask/common/utils.py	43.68% <83.33%> (+16.13%)	⬆️
python/cuml/experimental/explainer/kernel_shap.pyx	97.75% <100.00%> (+0.48%)	⬆️
...n/cuml/experimental/explainer/permutation_shap.pyx	98.82% <100.00%> (+0.84%)	⬆️
... and 74 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 befdf98...57ea041. Read the comment docs.

[JohnZed]

@gpucibot merge