Normal Frame

Normal frame manifolds.

class Manifolds.NormalFrame(dim, nb_pts, gd=None, tan=None, cotan=None, device=None)

add_cotan(cotan)

Same as add_gd(), for cotangents.

add_gd(gd)

Adds gd to the manifold geometrical descriptors.

Parameters

gd (Iterable or torch.Tensor) – Tensors that will be added to the manifold geometrical descriptors.

add_tan(tan)

Same as add_gd(), for tangents.

clone(requires_grad=None)

Returns a copy of the manifold. Detaches the computation graph.

Parameters

requires_grad (bool, default=None) – Set to True to record futher operations on the manifold tensors. Set to None to not change the requires_grad setting.

cot_to_vs(sigma, backend=None)

TODO: Add documentation.

property cotan

cotan_requires_grad_(requires_grad=True, index=- 1)

property device

property dim

property dtype

fill(manifold, copy=False, requires_grad=True)

TODO: write documentation

fill_cotan(cotan, copy=False, requires_grad=True)

Same as fill_cotan(), for cotangents.

fill_cotan_randn(requires_grad=True)

Same as fill_gd_randn(), for cotangents.

fill_cotan_zeros(requires_grad=True)

Same as fill_gd_zeros(), for cotangents.

fill_gd(gd, copy=False, requires_grad=True)

Fill geometrical descriptors to the manifold.

Parameters

• copy (bool, default=False) – If True, copy the geometrical descriptors into the manifold, detaching the computation graph. If False, the tensors are passed by references.

• requires_grad (bool, default=True) – If copy**=True, set the **requires_grad flag of the copied geometrical descriptors tensors to the given boolean value.

fill_gd_randn(requires_grad=True)

Fill geometrical descriptors with a standard normal distribution.

Parameters

requires_grad (bool, default=True) – Value given to the requires_grad flag of the random tensors.

fill_gd_zeros(requires_grad=True)

Fill geometrical descriptors with zero tensors.

Parameters

requires_grad (bool, default=True) – Value given to the requires_grad flag of the zero tensors.

fill_tan(tan, copy=False, requires_grad=True)

Same as fill_gd(), for tangents.

fill_tan_randn(requires_grad=True)

Same as fill_gd_randn(), for tangents.

fill_tan_zeros(requires_grad=True)

Same as fill_gd_zeros(), for tangents.

property gd

gd_requires_grad_(requires_grad=True, index=- 1)

infinitesimal_action(field)

Applies the vector field generated by the module on the landmark.

inner_prod_field(field)

TODO: Add documentation.

property len_gd

Returns the dimensions of each tensors for a manifold tensor of the manifold.

property nb_pts

Returns the number of points of the manifold.

negate_cotan()

Negate cotangents.

negate_gd()

Negate geometrical descriptors.

negate_tan()

Negate tangents.

property numel_gd

Returns the total number of elements for a manifold tensor of the manifold.

roll_cotan(l)

roll_gd(l)

roll_tan(l)

property shape_gd

Returns the shape of each tensors for a manifold tensor of the manifold.

property tan

tan_requires_grad_(requires_grad=True, index=- 1)

to_(*argv, **kwargs)

Performs manifold dtype or/and device conversion. A torch.dtype and torch.device are inferred from the arguments.

unroll_cotan()

unroll_gd()

unroll_tan()