def test_simple_run(self):
x_np = np.random.rand(1023).astype(np.float32)
x_torch = torch.Tensor(x_np)
clip_op = clip.Clip(torch.tensor(0.3), torch.tensor(0.7))
clip_x_np = np.clip(x_np, 0.3, 0.7)
clip_x_torch = clip_op(x_torch)
np.testing.assert_array_equal(clip_x_torch.cpu().numpy(), clip_x_np)
def test_backward(self):
x = torch.randn(3, 7, requires_grad=True)
x.retain_grad()
min_value = 0.3
max_value = 0.7
clip_op = clip.Clip(min_value, max_value, learn_min=True, learn_max=True)
clip_x = clip_op(x)
clip_x.retain_grad()
labels = torch.randint(6, (3,)).type(torch.LongTensor).cuda()
criterion = torch.nn.CrossEntropyLoss()
loss = criterion(clip_x, labels)
loss.backward()
assert x.grad.cpu()[x < min_value].sum() == 0
assert x.grad.cpu()[x > max_value].sum() == 0
assert torch.equal(clip_x.grad[(x > min_value) & (x < max_value)], x.grad[(x > min_value) & (x < max_value)])
def test_raise(self):
with pytest.raises(ValueError, match="must be scalar"):
clip_op = clip.Clip(torch.tensor(0.3), torch.tensor(0.7), learn_min=True)