def test_batch_matmul(self):
if not torch.cuda.is_available():
return
x1 = torch.randn(3, 2, 100, 3).cuda()
kern1 = MaternKernel(nu=2.5).cuda()
kern2 = GMaternKernel(nu=2.5).cuda()
rhs = torch.randn(3, 2, 100, 1).cuda()
res1 = kern1(x1, x1).matmul(rhs)
res2 = kern2(x1, x1).matmul(rhs)
self.assertLess(torch.norm(res1 - res2), 1e-4)
def forward_x1_eq_x2(self, nu):
if not torch.cuda.is_available():
return
x1 = torch.randn(100, 3).cuda()
kern1 = MaternKernel(nu=nu).cuda()
kern2 = GMaternKernel(nu=nu).cuda()
k1 = kern1(x1, x1).evaluate()
k2 = kern2(x1, x1).evaluate()
self.assertLess(torch.norm(k1 - k2), 1e-4)
def create_kernel_ard(self, num_dims, **kwargs):
return MaternKernel(nu=2.5, ard_num_dims=num_dims, **kwargs)
def create_kernel_no_ard(self, **kwargs):
return MaternKernel(nu=2.5, **kwargs)
