def test_focal_loss(self):
# Common verification
self._test_loss_function('focal_loss')
num_batches = 2
num_classes = 4
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.focal_loss(x, target, gamma=0).item(),
nn.functional.cross_entropy(x, target).item(), places=5)
# Equal probabilities
x = torch.ones(num_batches, num_classes, 20, 20)
self.assertAlmostEqual((1 - 1 / num_classes) * F.focal_loss(x, target, gamma=0).item(),
F.focal_loss(x, target, gamma=1).item(), places=5)
def test_focal_loss():
# Common verification
_test_loss_function(F.focal_loss)
num_batches = 2
num_classes = 4
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.focal_loss(x, target, gamma=0),
cross_entropy(x, target),
atol=1e-5)
# Equal probabilities
x = torch.ones(num_batches, num_classes, 20, 20)
assert torch.allclose(
(1 - 1 / num_classes) * F.focal_loss(x, target, gamma=0),
F.focal_loss(x, target, gamma=1),
atol=1e-5)
assert repr(nn.FocalLoss()) == "FocalLoss(gamma=2.0, reduction='mean')"