pyarmour 0.1.0

Zero-configuration adversarial robustness testing for ML models

PyArmour

Zero-configuration adversarial robustness testing for ML models using pytest.

Installation

pip install pyarmour

Quick Start

Decorator Usage

import pytest
from pyarmour import adversarial_test

@adversarial_test(model, attacks=["fgsm", "pgd"], epsilons=[0.03, 0.1])
def test_my_model(model, x, y):
    assert model(x).argmax() == y

CLI Usage

pyarmour run --model-path model.pth --data-path test_data/ --output report.html

Features

• Zero-configuration adversarial testing via pytest
• Pure NumPy implementation - no framework dependencies
• Built-in attacks: FGSM, PGD, DeepFool
• Visual diagnostics for vision models
• Text diff reports for NLP models

Documentation

Full documentation available at pyarmour.readthedocs.io