def test_ls_celoss(self):
num_batches = 2
num_classes = 4
# Common verification
self._test_loss_function('ls_cross_entropy', 0.1 / num_classes * (num_classes - 1) * 9)
x = torch.rand(num_batches, num_classes, 20, 20)
target = (num_classes * torch.rand(num_batches, 20, 20)).to(torch.long)
# Value check
self.assertAlmostEqual(F.ls_cross_entropy(x, target, eps=0).item(),
nn.functional.cross_entropy(x, target).item(), places=5)
self.assertAlmostEqual(F.ls_cross_entropy(x, target, eps=1).item(),
-1 / num_classes * nn.functional.log_softmax(x, dim=1).sum(dim=1).mean().item(),
places=5)
def test_ls_celoss():
num_batches = 2
num_classes = 4
# Common verification
_test_loss_function(F.ls_cross_entropy,
0.1 / num_classes * (num_classes - 1) * 9)
x = torch.rand(num_batches, num_classes, 20, 20)
target = (num_classes * torch.rand(num_batches, 20, 20)).to(torch.long)
# Value check
assert torch.allclose(F.ls_cross_entropy(x, target, eps=0),
cross_entropy(x, target),
atol=1e-5)
assert torch.allclose(F.ls_cross_entropy(x, target, eps=1),
-1 / num_classes *
log_softmax(x, dim=1).sum(dim=1).mean(),
atol=1e-5)
assert repr(nn.LabelSmoothingCrossEntropy()
) == "LabelSmoothingCrossEntropy(eps=0.1, reduction='mean')"