def test_ill_shape(self):
loss = TverskyLoss()
with self.assertRaisesRegex(AssertionError, ""):
loss.forward(torch.ones((2, 2, 3)), torch.ones((4, 5, 6)))
chn_input = torch.ones((1, 1, 3))
chn_target = torch.ones((1, 1, 3))
with self.assertRaisesRegex(ValueError, ""):
TverskyLoss(reduction="unknown")(chn_input, chn_target)
with self.assertRaisesRegex(ValueError, ""):
TverskyLoss(reduction=None)(chn_input, chn_target)
def test_shape(self, input_param, input_data, expected_val):
result = TverskyLoss(**input_param).forward(**input_data)
self.assertAlmostEqual(result.item(), expected_val, places=4)
def test_input_warnings(self):
chn_input = torch.ones((1, 1, 3))
chn_target = torch.ones((1, 1, 3))
with self.assertWarns(Warning):
loss = TverskyLoss(include_background=False)
loss.forward(chn_input, chn_target)
with self.assertWarns(Warning):
loss = TverskyLoss(softmax=True)
loss.forward(chn_input, chn_target)
with self.assertWarns(Warning):
loss = TverskyLoss(to_onehot_y=True)
loss.forward(chn_input, chn_target)
def test_shape(self, input_param, input_data, expected_val):
result = TverskyLoss(**input_param).forward(**input_data)
np.testing.assert_allclose(result.detach().cpu().numpy(),
expected_val,
rtol=1e-4)
def test_script(self):
loss = TverskyLoss()
test_input = torch.ones(2, 1, 8, 8)
test_script_save(loss, test_input, test_input)
