more efficient implementation

Author: lucamarini22

src/pytorch_metric_learning/losses/generic_pair_loss.py

@@ -42,8 +42,4 @@ def pair_based_loss(self, mat, indices_tuple):
 
     @staticmethod
     def _assert_either_pos_or_neg(pos_mask, neg_mask):
-        pos_indices = set(pos_mask.flatten().nonzero().flatten().tolist())
-        neg_indices = set(neg_mask.flatten().nonzero().flatten().tolist())
-        assert (
-            pos_indices.isdisjoint(neg_indices)
-        ), "Each pair should be either be positive or negative"
+        assert not torch.any((pos_mask != 0) & (neg_mask != 0)), "Each pair should be either be positive or negative"