hypll.manifolds.poincare_ball.manifold

Classes

PoincareBall

Class representing the Poincare ball model of hyperbolic space.

class hypll.manifolds.poincare_ball.manifold.PoincareBall

Class representing the Poincare ball model of hyperbolic space.

Implementation based on the geoopt implementation, but changed to use hyperbolic torch functions.

c

Curvature of the manifold.

__init__(c: Curvature)

Initializes an instance of PoincareBall manifold.

Examples

>>> from hypll.manifolds.poincare_ball import PoincareBall, Curvature
>>> curvature = Curvature(value=1.0)
>>> manifold = Manifold(c=curvature)

mobius_add(x: ManifoldTensor, y: ManifoldTensor) → ManifoldTensor

project(x: ManifoldTensor, eps: float = -1.0) → ManifoldTensor

expmap(v: TangentTensor) → ManifoldTensor

logmap(x: ManifoldTensor | None, y: ManifoldTensor)

gyration(u: ManifoldTensor, v: ManifoldTensor, w: ManifoldTensor) → ManifoldTensor

transp(v: TangentTensor, y: ManifoldTensor) → TangentTensor

dist(x: ManifoldTensor, y: ManifoldTensor) → Tensor

inner(u: TangentTensor, v: TangentTensor, keepdim: bool = False, safe_mode: bool = True) → Tensor

euc_to_tangent(x: ManifoldTensor, u: ManifoldTensor) → TangentTensor

hyperplane_dists(x: ManifoldTensor, z: ManifoldTensor, r: Tensor | None) → Tensor

fully_connected(x: ManifoldTensor, z: ManifoldTensor, bias: Tensor | None) → ManifoldTensor

frechet_mean(x: ManifoldTensor, batch_dim: int | list[int] = 0, keepdim: bool = False) → ManifoldTensor

midpoint(x: ManifoldTensor, batch_dim: int = 0, w: Tensor | None = None, keepdim: bool = False) → ManifoldTensor

frechet_variance(x: ManifoldTensor, mu: ManifoldTensor | None = None, batch_dim: int | list[int] = -1, keepdim: bool = False) → Tensor

construct_dl_parameters(in_features: int, out_features: int, bias: bool = True) → tuple[ManifoldParameter, Parameter | None]

reset_parameters(weight: ManifoldParameter, bias: Parameter | None) → None

unfold(input: ManifoldTensor, kernel_size: int | Tuple[int, int], dilation: int | Tuple[int, int] = 1, padding: int | Tuple[int, int] = 0, stride: int | Tuple[int, int] = 1) → ManifoldTensor

flatten(x: ManifoldTensor, start_dim: int = 1, end_dim: int = -1) → ManifoldTensor

Flattens a manifold tensor by reshaping it. If start_dim or end_dim are passed, only dimensions starting with start_dim and ending with end_dim are flattend.

If the manifold dimension of the input tensor is among the dimensions which are flattened, applies beta-concatenation to the points on the manifold. Otherwise simply flattens the tensor using torch.flatten.

Updates the manifold dimension if necessary.

cdist(x: ManifoldTensor, y: ManifoldTensor) → Tensor

cat(manifold_tensors: Tuple[ManifoldTensor, ...] | List[ManifoldTensor], dim: int = 0) → ManifoldTensor