vraven · class
IntervalBoundVerifier
IntervalBoundVerifier(model: 'nn.Module', *, model_name: 'str | None' = None) -> 'None'Certify local L-infinity robustness for a supported PyTorch FX graph.
Usage
from vraven import IntervalBoundVerifier
value = IntervalBoundVerifier(...)Methods
bound_graph(self, x: 'Tensor', *, epsilon: 'float', clip_min: 'float | None' = None, clip_max: 'float | None' = None, capture_nodes: 'Sequence[str]' = (), bound_transforms: 'Mapping[str, BoundTransform] | None' = None) -> 'tuple[Bound, list[LayerIntervalRecord], dict[str, Bound]]'
Propagate one input box and optionally capture or transform graph nodes.
bound_graph_difference(self, x: 'Tensor', *, epsilon: 'float', clip_min: 'float | None' = None, clip_max: 'float | None' = None, paired_bound_transforms: 'Mapping[str, PairedBoundTransform] | None' = None, output_linear_coefficients: 'Tensor | None' = None) -> 'tuple[Bound, Bound, Bound]'
Propagate original/intervened boxes with a shared-input difference bound.
certify(self, x: 'Tensor', *, target: 'int | None' = None, epsilon_grid: 'tuple[float, ...]' = (0.0, 0.001, 0.005, 0.01, 0.05, 0.1), clip_min: 'float | None' = None, clip_max: 'float | None' = None, cohort_x: 'Tensor | None' = None, cohort_y: 'Tensor | list[int] | None' = None, selected_indices: 'list[int] | None' = None, radius_search_steps: 'int' = 24, input_spec: 'InputDomainSpec | None' = None) -> 'FormalRobustnessCertificate'
Build a formal certificate curve for a sample and optional cohort.
verify_sample(self, x: 'Tensor', *, epsilon: 'float', target: 'int | None' = None, clip_min: 'float | None' = None, clip_max: 'float | None' = None) -> 'FormalVerificationPoint'
Verify one fixed-epsilon local classification property.
Source of truthThis page is generated from the checked-out public symbol, signature, and docstring. Pin the package version for reproducible work.