Adding label check to trades adversarial trainer

[GiulioZizzo]

Description

We add a dimensionality check to the labels in Trades adversarial training before applying argmax operations.

Fixes #2230

Type of change

Please check all relevant options.

• Improvement (non-breaking)
• Bug fix (non-breaking)
• New feature (non-breaking)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

Testing

Please describe the tests that you ran to verify your changes. Consider listing any relevant details of your test configuration.

• Updates to test_adversarial_trainer_trades_pytorch to check with both one hot and class index style of label.

Test Configuration:

• OS: MacOS
• Python version: 3.9
• ART version or commit number: ART 1.15
• TensorFlow / Keras / PyTorch / MXNet version: Torch 2.0.1

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[codecov-commenter]

Codecov Report

Merging #2231 (d43f473) into dev_1.15.1 (c63d5d5) will increase coverage by 1.00%.
The diff coverage is 100.00%.

❗ Your organization is not using the GitHub App Integration. As a result you may experience degraded service beginning May 15th. Please install the Github App Integration for your organization. Read more.

@@              Coverage Diff               @@
##           dev_1.15.1    #2231      +/-   ##
==============================================
+ Coverage       84.60%   85.61%   +1.00%     
==============================================
  Files             308      308              
  Lines           27464    27470       +6     
  Branches         5045     5046       +1     
==============================================
+ Hits            23237    23519     +282     
+ Misses           2961     2670     -291     
- Partials         1266     1281      +15     

Files Changed	Coverage Δ
...nces/trainer/adversarial_trainer_trades_pytorch.py	88.39% <100.00%> (+0.65%)	⬆️

... and 11 files with indirect coverage changes

[beat-buesser]

Hi @GiulioZizzo Thank you very much for you pull request! I have changed the target branch to dev_1.15.1 for the next patch release.

@Zaid-Hameed Because you have implemented the TRADES trainer, could you please add a review of the proposed changes?

[Zaid-Hameed]

Hi Giulio, Can you please check if adding from_logits=True makes any changes to tests outcomes i.e., changing line 37 to
classifier, _ = image_dl_estimator(from_logits=True)

This is just to make sure that softmax is not being applied twice in test files.

[Zaid-Hameed]

Thanks a lot Giulio for making the changes.

[beat-buesser]

Hi @GiulioZizzo I think you proposal for functionality is good. I'm proposing to follow the ART patterns in solving the issue, what do you think?

[beat-buesser]

This change is not needed because ART uses internally the one-hot-encoded format for classification labels.

[beat-buesser]

The idea of this change is good. Let's follow the ART pattern by using y = check_and_transform_label_format(y, nb_classes=self._classifier.nb_classes) for y and validation_data[1] imported with from art.utils import check_and_transform_label_format at the beginning of this method to ensure they are both one-hot-encoded.