Function robustness_eval 

pub fn robustness_eval(
    model: &dyn AttackModel,
    inputs: &[Vec<f64>],
    labels: &[Vec<f64>],
    config: &AttackConfig,
    seed: u64,
) -> Result<f64, AdversarialError>

Evaluate the model’s adversarial robustness on a set of samples.

For each sample the PGD attack is run; a sample is considered “robust” if the argmax prediction does not change after the attack (for classification), or equivalently if the adversarial loss is not greater than the clean loss (for regression).

Returns the fraction of samples that remain correctly classified (robust), in the range [0, 1].