torch_geometric.explain.metric.fidelity_curve_auc

fidelity_curve_auc(pos_fidelity: Tensor, neg_fidelity: Tensor, x: Tensor) → Tensor[source]

返回保真度曲线的AUC，如“GraphFramEx: Towards Systematic Evaluation of Explainability Methods for Graph Neural Networks”论文中所述。

更准确地说，返回的AUC是

\[f(x) = \frac{\textrm{fid}_{+}}{1 - \textrm{fid}_{-}}\]
Parameters:

  • pos_fidelity (torch.Tensor) – The positive fidelity \(\textrm{fid}_{+}\).

  • neg_fidelity (torch.Tensor) – The negative fidelity \(\textrm{fid}_{-}\).

  • x (torch.Tensor) – 包含\(x\)轴上点的张量。 需要按升序排序。

Return type:

Tensor