geomstats.varifold package

Subpackages

• geomstats.varifold.keops package
  • Submodules
  • geomstats.varifold.keops.genred module
    • BinetKernel()
    • CauchyKernel()
    • GaussianBinetPairing
      • GaussianBinetPairing.kernel_prod()
    • GaussianKernel()
    • LinearKernel()
    • OrientedGaussianKernel()
    • UnorientedGaussianKernel()
  • geomstats.varifold.keops.lazy module
    • BinetKernel()
    • CauchyKernel()
    • GaussianBinetKernel()
    • GaussianKernel()
    • LinearKernel()
    • RestrictedGaussianKernel()
    • SurfaceKernelPairing
  • Module contents

Submodules

geomstats.varifold.base module

Base objects for varifold implementation.

class geomstats.varifold.base.Pairing

Bases: ABC

Kernel pairing.

geomstats.varifold.kernel module

Kernel pairings.

geomstats.varifold.kernel.GaussianBinetPairing(sigma, backend='auto')

Instantiate a Gaussian-Binet kernel pairing.

This pairing is defined by

\[K(x, y, u, v) = exp(-||x - y||^2 / sigma^2) * <u, v>^2\]

Parameters:

• sigma (float) – Positive bandwidth parameter of the Gaussian kernel.

• backend ({“auto”, “torch”, “keops”, “keops_genred”, “keops_lazy”}) – Implementation backend.

  • “auto”: Select an implementation automatically (typically prefers a KeOps-based implementation when available, otherwise falls back to a Torch/NumPy implementation).

  • “backend”: Dense implementation using the current geomstats backend.

  • “keops”: Alias for “keops_genred”.

  • “keops_genred”: KeOps implementation using Genred reductions.

  • “keops_lazy”: KeOps LazyTensor-based implementation.

Returns:

Pairing – An object implementing the kernel pairing.

Notes

The dense (“backend”) implementation materializes pairwise matrices and is memory-bound for large inputs. KeOps-based implementations avoid forming the full kernel matrix and are more efficient for large-scale problems.

The “auto” backend does not guarantee optimal performance in all cases, but provides a reasonable default based on available dependencies.

Module contents