Skip to content

Do not promote argmin VNNLIB properties to top-1 robustness; - #122

Merged
yuleisui merged 1 commit into
SVF-tools:mainfrom
guanqin-123:vnnlib-argmin-fix
Oct 24, 2026
Merged

yuleisui merged 1 commit into
SVF-tools:mainfrom
guanqin-123:vnnlib-argmin-fix

Conversation

@guanqin-123

Copy link
Copy Markdown
Contributor

Rows with -1 on candidate class C ("C scores lowest") were promoted to TOP1_ROBUST, which gave wrong FALSIFIED verdicts on ACAS Xu prop_5, prop_9 and prop_10 (networks 1_1, 3_3 and 4_5);
keep them UNSAFE_LINEAR. CIFAR-100 keeps one top-1 query: both VNNLIB orientations canonicalise to Y_true - Y_j <= 0.

Fix with A missing file in an existing category raises FileNotFoundError naming the path instead of downloading.

…leNotFoundError for missing instance files

Rows with -1 on candidate class C ("C scores lowest") were promoted to
TOP1_ROBUST, which gave wrong FALSIFIED verdicts on ACAS Xu prop_5,
prop_9 and prop_10 (networks 1_1, 3_3 and 4_5); keep them UNSAFE_LINEAR.
CIFAR-100 keeps one top-1 query: both VNNLIB orientations canonicalise
to Y_true - Y_j <= 0. A missing file in an existing category raises
FileNotFoundError naming the path instead of downloading.
@codecov

codecov Bot commented Oct 24, 2026 •

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 77.62%. Comparing base (a4b19aa) to head (ee48c1b).

Additional details and impacted files
@@           Coverage Diff           @@
##             main     #122   +/-   ##
=======================================
  Coverage   77.62%   77.62%           
=======================================
  Files          95       95           
  Lines       23009    23007    -2     
=======================================
- Hits        17860    17859    -1     
+ Misses       5149     5148    -1     
Flag Coverage Δ
bab 47.09% <100.00%> (-0.01%) ⬇️
backend-float32 50.27% <0.00%> (+0.13%) ⬆️
backend-float64 51.99% <0.00%> (+<0.01%) ⬆️
frontend 30.91% <33.33%> (+<0.01%) ⬆️
pipeline-fuzz 20.31% <100.00%> (-0.01%) ⬇️
pipeline-verify 43.23% <100.00%> (-0.01%) ⬇️

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

Files with missing lines Coverage Δ
act/front_end/vnnlib_loader/data_model_loader.py 77.54% <100.00%> (ø)
act/front_end/vnnlib_loader/vnnlib_parser.py 80.80% <100.00%> (-0.06%) ⬇️

... and 3 files with indirect coverage changes


Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a4b19aa...ee48c1b. Read the comment docs.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@yuleisui
yuleisui merged commit 4a7d3db into SVF-tools:main Oct 24, 2026
8 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants