RobustnessOptions

Options & Configuration · AiDotNet.Evaluation.Options

Configuration options for robustness evaluation.

For Beginners

Robustness tests answer questions like:

  • What happens if there’s noise in the input? (Noise robustness)
  • What if someone intentionally tries to fool the model? (Adversarial robustness)
  • What if the input has small errors or missing values? (Perturbation robustness)

This is important for real-world deployments where inputs aren’t always clean.

How It Works

Robustness evaluation measures how well a model performs under input perturbations, noise, and adversarial attacks. A robust model maintains performance despite variations.

Constructors

Constructor Summary
RobustnessOptions()

Properties

Property Summary
AdversarialAttackMethod? AttackMethod { get; set; } Adversarial attack method.
MissingValueStrategy? MissingStrategy { get; set; } Missing value handling strategy.
bool? ComputeFlipRate { get; set; } Whether to compute prediction flip rate under perturbation.
bool? ComputeInputGradients { get; set; } Whether to compute input gradient norms.
bool? EstimateLipschitzConstant { get; set; } Whether to estimate local Lipschitz constant.
bool? ParallelExecution { get; set; } Whether to run tests in parallel.
bool? TestAdversarialRobustness { get; set; } Whether to test adversarial robustness.
bool? TestFeaturePerturbation { get; set; } Whether to test feature perturbation impact.
bool? TestGaussianNoise { get; set; } Whether to test Gaussian noise robustness.
bool? TestMissingValues { get; set; } Whether to test missing value robustness.
bool? TestOutlierRobustness { get; set; } Whether to test outlier robustness.
bool? TestUniformNoise { get; set; } Whether to test uniform noise.
double? AdversarialEpsilon { get; set; } Epsilon for adversarial attacks.
double? PerturbationMagnitude { get; set; } Perturbation magnitude (relative).
double? SampleRatio { get; set; } Subset of samples to test (ratio).
double[] DropoutRates { get; set; } Feature dropout rates to test (random feature removal).
double[] MissingValueRates { get; set; } Missing value rates to test.
double[] NoiseLevels { get; set; } Noise levels (standard deviations) to test.
double[] OutlierRates { get; set; } Outlier contamination rates to test.
int? AttackIterations { get; set; } Number of attack iterations (for iterative methods).
int? LipschitzSamples { get; set; } Number of samples for Lipschitz estimation.
int? MaxDegreeOfParallelism { get; set; } Maximum parallelism.
int? MaxSamples { get; set; } Maximum samples to test.
int? NoiseSamplesPerInput { get; set; } Number of noise samples per input.
int? RandomSeed { get; set; } Random seed for reproducibility.
int[] FeaturesToPerturb { get; set; } Specific features to perturb.