def test__generate_all_triplets_multiple_negative(self):
"""
Case where just one positive example and 1 negative sample
"""
# Arrange
input_target = torch.tensor([0, 0, 1, 2])
expected_triplet_indices = torch.tensor([[0, 1, 2], [0, 1, 3]])
sut = OnlineTripletLoss(.5)
# Act
actual = sut._generate_all_triplets(input_target)
# Assert
self.assertSequenceEqual(
expected_triplet_indices.cpu().numpy().tolist(),
actual.cpu().numpy().tolist())
def test__generate_all_triplets_three_clases_missing_target(self):
"""
Case where not all target classes are presented in the target ( can happen within a batch)
"""
# Arrange
input_target = torch.tensor([0, 0, 1, 4])
expected_triplet_indices = torch.tensor([[0, 1, 2], [0, 1, 3]])
sut = OnlineTripletLoss(.5)
# Act
actual = sut._generate_all_triplets(input_target)
# Assert
self.assertSequenceEqual(
expected_triplet_indices.cpu().numpy().tolist(),
actual.cpu().numpy().tolist())