def create_lazy_tensor(self):
a = torch.tensor([[4, 0, 2], [0, 3, -1], [2, -1, 3]],
dtype=torch.float)
b = torch.tensor([[2, 1], [1, 2]], dtype=torch.float)
c = torch.tensor(
[[4, 0.5, 1, 0], [0.5, 4, -1, 0], [1, -1, 3, 0], [0, 0, 0, 4]],
dtype=torch.float)
d = torch.tensor([2], dtype=torch.float)
e = torch.tensor([5], dtype=torch.float)
f = torch.tensor([2.5], dtype=torch.float)
a.requires_grad_(True)
b.requires_grad_(True)
c.requires_grad_(True)
d.requires_grad_(True)
e.requires_grad_(True)
f.requires_grad_(True)
kp_lazy_tensor = KroneckerProductLazyTensor(NonLazyTensor(a),
NonLazyTensor(b),
NonLazyTensor(c))
diag_lazy_tensor = KroneckerProductDiagLazyTensor(
ConstantDiagLazyTensor(d, diag_shape=3),
ConstantDiagLazyTensor(e, diag_shape=2),
ConstantDiagLazyTensor(f, diag_shape=4),
)
return KroneckerProductAddedDiagLazyTensor(kp_lazy_tensor,
diag_lazy_tensor)
def create_lazy_tensor(self):
a = torch.tensor([[4, 0, 2], [0, 3, -1], [2, -1, 3]], dtype=torch.float)
b = torch.tensor([[2, 1], [1, 2]], dtype=torch.float)
c = torch.tensor([[4, 0.5, 1, 0], [0.5, 4, -1, 0], [1, -1, 3, 0], [0, 0, 0, 4]], dtype=torch.float)
d = 0.5 * torch.rand(24, dtype=torch.float)
a.requires_grad_(True)
b.requires_grad_(True)
c.requires_grad_(True)
d.requires_grad_(True)
kp_lazy_tensor = KroneckerProductLazyTensor(NonLazyTensor(a), NonLazyTensor(b), NonLazyTensor(c))
diag_lazy_tensor = DiagLazyTensor(d)
return KroneckerProductAddedDiagLazyTensor(kp_lazy_tensor, diag_lazy_tensor)
def create_lazy_tensor(self):
a = torch.tensor([[4, 0, 2], [0, 3, -1], [2, -1, 3]], dtype=torch.float)
b = torch.tensor([[2, 1], [1, 2]], dtype=torch.float)
c = torch.tensor([[4, 0.5, 1, 0], [0.5, 4, -1, 0], [1, -1, 3, 0], [0, 0, 0, 4]], dtype=torch.float)
a.requires_grad_(True)
b.requires_grad_(True)
c.requires_grad_(True)
kp_lazy_tensor = KroneckerProductLazyTensor(NonLazyTensor(a), NonLazyTensor(b), NonLazyTensor(c))
diag_lazy_tensor = ConstantDiagLazyTensor(
torch.tensor([0.25], dtype=torch.float, requires_grad=True), kp_lazy_tensor.shape[-1],
)
return KroneckerProductAddedDiagLazyTensor(kp_lazy_tensor, diag_lazy_tensor)
def create_lazy_tensor(self):
a = torch.tensor([[4, 0, 2], [0, 3, -1], [2, -1, 3]],
dtype=torch.float)
b = torch.tensor([[2, 1], [1, 2]], dtype=torch.float)
c = torch.tensor(
[[4, 0.5, 1, 0], [0.5, 4, -1, 0], [1, -1, 3, 0], [0, 0, 0, 4]],
dtype=torch.float)
a.requires_grad_(True)
b.requires_grad_(True)
c.requires_grad_(True)
kp_lazy_tensor = KroneckerProductLazyTensor(NonLazyTensor(a),
NonLazyTensor(b),
NonLazyTensor(c))
return KroneckerProductAddedDiagLazyTensor(
kp_lazy_tensor,
DiagLazyTensor(0.1 * torch.ones(kp_lazy_tensor.shape[-1])))