Fix min_epsilon of BoundaryAttack to receive a non-negative value

[kaitokishi]

Signed-off-by: kaitokishi kishi.kaito@fujitsu.com

Description

Passing a positive float value into min_epsilon of BoundaryAttack raises Exception. This is because _check_params() raises ValueError even if self.min_epsilon is positive float. This PR is to fix this bug.

Also, this PR makes it simple to use min_epsilon by avoiding to use None.

Fixes #1258

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

• Below code

attack = BoundaryAttack(model, batch_size=100, targeted=False, max_iter=20, epsilon=0.1, min_epsilon=0.02)
x_adv = attack.generate(x=x)

works.

Test Configuration:

• OS: Ubuntu 18.04.5 LTS
• Python version: 3.6.9
• ART version or commit number: 1.7.0
• TensorFlow version: 2.3.0

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

[kaitokishi]

I would like to make a pull request to dev_1.7.2 but I cannot select the branch because there is no dev_1.7.2 branch. Thus, I selected dev_1.8.0.

[codecov-commenter]

Codecov Report

Merging #1262 (f55dc71) into dev_1.7.2 (4b2745e) will increase coverage by 0.00%.
The diff coverage is 0.00%.

@@            Coverage Diff             @@
##           dev_1.7.2    #1262   +/-   ##
==========================================
  Coverage      43.42%   43.42%           
==========================================
  Files            231      231           
  Lines          21201    21202    +1     
  Branches        3956     3956           
==========================================
+ Hits            9207     9208    +1     
  Misses         10852    10852           
  Partials        1142     1142           

Impacted Files	Coverage Δ
art/attacks/evasion/boundary.py	76.00% <0.00%> (ø)
art/estimators/classification/pytorch.py	72.43% <0.00%> (+0.06%)	⬆️

[beat-buesser]

Suggested change

	min_epsilon: float = 0,
	min_epsilon: Union[float, int] = 0,

[kaitokishi]

In PEP 484,

when an argument is annotated as having type float, an argument of type int is acceptable

so I think only float is sufficient as in epsilon and delta.

[beat-buesser]

Hi @kaitokishi Thank you very much for your first pull request to ART! This is really great! I have created branch dev_1.7.2 now and updated the PR accordingly. I like removing the None, but I would like to suggest to apply typing and testing for int and float explicitly for good testing. What do you think?

[kaitokishi]

Thank you for your suggestion. I think using Union[float, int] is redundent as I commented. Nevertheless, if you would like to write it explicitly, I accept your suggestion.

[beat-buesser]

Hi @kaitokishi Thank you very much for you first pull request to ART!