def test_batch_eval_neg_michalewicz(self, cuda=False): device = torch.device("cuda") if cuda else torch.device("cpu") for dtype in (torch.float, torch.double): X = torch.zeros(2, 10, device=device, dtype=dtype) res = neg_michalewicz(X) self.assertEqual(res.dtype, dtype) self.assertEqual(res.device.type, device.type) self.assertEqual(res.shape, torch.Size([2]))
def test_batch_eval_neg_michalewicz(self, cuda=False): device = torch.device("cuda") if cuda else torch.device("cpu") for dtype in (torch.float, torch.double): X = torch.zeros(2, 10, device=device, dtype=dtype) res = neg_michalewicz(X) self.assertEqual(res.dtype, dtype) self.assertEqual(res.device.type, device.type) self.assertEqual(res.shape, torch.Size([2]))
def test_neg_michalewicz_global_maximum(self, cuda=False): device = torch.device("cuda") if cuda else torch.device("cpu") for dtype in (torch.float, torch.double): X = torch.tensor( GLOBAL_MAXIMIZER, device=device, dtype=dtype, requires_grad=True ) res = neg_michalewicz(X) res.backward() self.assertAlmostEqual(res.item(), GLOBAL_MAXIMUM, places=4) self.assertLess(X.grad.abs().max().item(), 1e-3)
def test_neg_michalewicz_global_maximum(self, cuda=False): device = torch.device("cuda") if cuda else torch.device("cpu") for dtype in (torch.float, torch.double): X = torch.tensor(GLOBAL_MAXIMIZER, device=device, dtype=dtype, requires_grad=True) res = neg_michalewicz(X) self.assertAlmostEqual(res.item(), GLOBAL_MAXIMUM, places=4) grad = torch.autograd.grad(res, X)[0] self.assertLess(grad.abs().max().item(), 1e-3)