ExplanationFaithfulnessReport<T>
Models & Types · AiDotNet.Interpretability
The auto-audit of how faithful a model’s explanations are — whether the features they highlight actually drive the model’s output. Surfaced on the built result when a model explainer is configured.
How It Works
Mainstream explainability libraries produce attributions but do not tell you whether to trust them. This runs deletion/insertion and comprehensiveness/sufficiency perturbation tests against the trained model, so a low faithfulness score is a warning that the explanation does not reflect the model.
Constructors
| Constructor | Summary |
|---|---|
ExplanationFaithfulnessReport() |
Properties
| Property | Summary |
|---|---|
FaithfulnessReport<T> ExplainerFaithfulness { get; set; } |
Faithfulness of the configured explainer’s global attributions, or null when the explainer cannot produce a global attribution vector (does not implement IGlobalAttributionExplainer). |
FaithfulnessReport<T> ModelImportanceFaithfulness { get; set; } |
Faithfulness of the model’s own built-in feature importances, or null when they could not be aligned to the input features. |
string ExplainerName { get; set; } |
The configured explainer’s type name, for reference. |