src.evaluation_metric.matching_accuracy¶
- src.evaluation_metric.matching_accuracy(pmat_pred: torch.Tensor, pmat_gt: torch.Tensor, ns: torch.Tensor, idx: int) torch.Tensor[source]¶
Matching Accuracy between predicted permutation matrix and ground truth permutation matrix.
\[\text{matching recall} = \frac{tr(\mathbf{X}\cdot {\mathbf{X}^{gt}}^\top)}{\sum \mathbf{X}^{gt}}\]This function is a wrapper of
matching_recall.- Parameters
pmat_pred – \((b\times n_1 \times n_2)\) predicted permutation matrix \((\mathbf{X})\)
pmat_gt – \((b\times n_1 \times n_2)\) ground truth permutation matrix \((\mathbf{X}^{gt})\)
ns – \((b\times g)\) number of nodes in graphs, where \(g=2\) for 2GM, and \(g>2\) for MGM. We support batched instances with different number of nodes, and
nsis required to specify the exact number of nodes of each instance in the batch.idx – \((int)\) index of source graph in the graph pair.
- Returns
\((b)\) matching accuracy
Note
If the graph matching problem has no outliers, it is proper to use this metric and papers call it “matching accuracy”. If there are outliers, it is better to use
matching_precisionandmatching_recall.