|
pulsatrix 1.1.0
|
ExplanationSensitivity()'s settings. More...
#include <explanation_robustness.hpp>
Public Attributes | |
| double | radius = 0.1 |
| The perturbations' L∞ radius, as a fraction of the input's standard deviation (its largest absolute value when the input is constant). Quantus's default is an absolute 0.2. | |
| int64_t | samples = 20 |
| Perturbations to sample; Quantus uses 200, which is slow for sampling explainers. | |
| uint64_t | seed = 0 |
| int64_t | max_halvings = 4 |
| The adversarial point: the FGSM step is halved until the prediction stays, at most this many times. | |
ExplanationSensitivity()'s settings.
| int64_t pulsatrix::SensitivityOptions::max_halvings = 4 |
The adversarial point: the FGSM step is halved until the prediction stays, at most this many times.
| double pulsatrix::SensitivityOptions::radius = 0.1 |
The perturbations' L∞ radius, as a fraction of the input's standard deviation (its largest absolute value when the input is constant). Quantus's default is an absolute 0.2.
| int64_t pulsatrix::SensitivityOptions::samples = 20 |
Perturbations to sample; Quantus uses 200, which is slow for sampling explainers.
| uint64_t pulsatrix::SensitivityOptions::seed = 0 |