vraven.darkside · function
analyse_causal_synergy
analyse_causal_synergy(model: 'torch.nn.Module', inputs: 'Any', *, target: 'int | None', groups: 'Sequence[UnitGroup]', baseline: 'str' = 'zero', maximum_pairs: 'int' = 256, config: 'VRavenConfig | None' = None) -> 'AnalysisResult'Estimate pairwise interaction using single and joint ablations.
For target score ``s`` and score drops ``d(A)``, ``d(B)`` and ``d(A,B)``, the
reported interaction is ``d(A,B) - d(A) - d(B)``. Positive values are
super-additive under this intervention, while negative values indicate
redundancy or compensation. The quantity is baseline and context specific.
Usage
from vraven.darkside import analyse_causal_synergy
result = analyse_causal_synergy(... )Source of truthThis page is generated from the checked-out public symbol, signature, and docstring. Pin the package version for reproducible work.