Refactor Evasion Detectors

[f4str]

Description

Refactor the art.defences.detector.evasion module by creating a base abstract class EvasionDetector which all of the evasion detectors now inherit from. This base class has a .detect() method which will return a report and binary numpy array for which samples are detected adversarial. The BinaryInputDetector, BinaryActivationDetector, and SubsetScanningDetector classes have been refactored accordingly to inherit from EvasionDetector and use the .detect() method.

The unit tests have been updated accordingly and now use the pytest framework, so they have been removed from legacy tests.

This is technically a "breaking change" since the module has been refactored, but existing functionality is unchanged. Documentation has been updated along with updating notebooks/detection_adversarial_samples_cifar10.ipynb to use the new evasion detector structure.

Fixes #1913

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.

• Unit tests for BinaryInputDetector using Keras, TensorFlow, and PyTorch
• Unit tests for BinaryActivationDetector using Keras, TensorFlow, and PyTorch
• Unit tests for SubsetScanningDetector using Keras, TensorFlow, and PyTorch

Test Configuration:

• OS
• Python version
• ART version or commit number
• TensorFlow / Keras / PyTorch / MXNet version

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 #1993 (a2b2ceb) into dev_1.14.0 (5ddd8ef) will decrease coverage by 0.01%.
The diff coverage is 80.66%.

📣 This organization is not using Codecov’s GitHub App Integration. We recommend you install it so Codecov can continue to function properly for your repositories. Learn more

@@              Coverage Diff               @@
##           dev_1.14.0    #1993      +/-   ##
==============================================
- Coverage       85.41%   85.41%   -0.01%     
==============================================
  Files             290      291       +1     
  Lines           25437    25453      +16     
  Branches         4573     4587      +14     
==============================================
+ Hits            21728    21740      +12     
+ Misses           2541     2539       -2     
- Partials         1168     1174       +6     

Impacted Files	Coverage Δ
...tector/evasion/subsetscanning/scoring_functions.py	66.66% <ø> (ø)
...ces/detector/evasion/binary_activation_detector.py	68.42% <68.42%> (ø)
art/defences/detector/evasion/evasion_detector.py	69.56% <69.56%> (ø)
...fences/detector/evasion/subsetscanning/detector.py	83.19% <84.84%> (-0.62%)	⬇️
art/defences/detector/evasion/__init__.py	100.00% <100.00%> (ø)
...defences/detector/evasion/binary_input_detector.py	100.00% <100.00%> (ø)
art/attacks/evasion/boundary.py	92.72% <0.00%> (-1.22%)	⬇️

[beat-buesser]

Hi @f4str Thank you very much for refactoring the evasion detectors, this should be very useful for many users! I also like how you made the detector tests framework independent!