skshapes.convolutions package

Submodules

skshapes.convolutions.constant_kernel module

Constant kernel class.

class skshapes.convolutions.constant_kernel.Constant1Kernel(points, target_points=None)

Bases: object

A summation operator that stands for torch.ones(N, N).

Parameters:

  • points – The points in the source space.

  • target_points – The points in the target space.

property T

Returns the transpose of the kernel.

skshapes.convolutions.constant_kernel.constant_1_kernel(*, points, target_points=None, **kwargs)

(N, N) matrix of squared distances between points.

For geodesic kernels, we may want to stick to connected components?

Return type:

Constant1Kernel

skshapes.convolutions.linear_operator module

Linear operators with scaling.

class skshapes.convolutions.linear_operator.LinearOperator(matrix, input_scaling=None, output_scaling=None)

Bases: object

A simple wrapper for scaled linear operators.

Parameters:

  • matrix – The matrix to wrap.

  • input_scaling – _description_

  • output_scaling – _description_

property T

Transpose the kernel.

property shape

Shape of the kernel.

skshapes.convolutions.mesh_convolution module

Mesh convolution kernel.

skshapes.convolutions.point_convolution module

Point convolutions kernel.

skshapes.convolutions.squared_distances module

Squared distances between points.

class skshapes.convolutions.squared_distances.KeOpsSquaredDistances(*, points, cutoff=None, kernel=None, target_points=None)

Bases: object

Squared distances between points using KeOps.

Parameters:

  • points – The points in the source space.

  • cutoff (int | float | None) – The cutoff value for the window.

  • kernel (Callable | None) – The kernel to use.

  • target_points – The points in the target space.

property T

Transpose of the kernel.

sum(*args, **kwargs)

Sum of the kernel.

skshapes.convolutions.squared_distances.squared_distances(*, points, window=None, cutoff=None, geodesic=False, kernel=None, target_points=None)

Matrix of squared distances between points.

If source_points is not None, then the (N, M) matrix of squared distances between points (in rows) and source_points (in columns) is returned.

Else, the (N, N) matrix of squared distances between points is returned.

Parameters:

  • points – The points in the target space.

  • window (Literal[None, 'ball', 'knn', 'spectral']) – The type of window to use.

  • cutoff (int | float | None) – The cutoff value for the window.

  • geodesic (bool) – Whether to use geodesic distances.

  • kernel (Callable | None) – The kernel to use.

  • target_points – The points in the source space. If None, then the points in the target space are used.