oineus.alpha_filtration

oineus.alpha_filtration(points, weights=None, exact=False, periodic=False, compute_bounding_box=True, bbox_min=None, bbox_max=None, packed=True, n_threads=1)[source]

Build an alpha-shape filtration from a 2D or 3D point cloud.

Combinatorics come from diode (CGAL Delaunay); filtration values are the alpha values returned by diode. For one-shot diagrams use compute_diagrams_alpha(); the differentiable Cech-Delaunay path reuses the same combinatorics with autograd-attached values.

Parameters:

  • points (ndarray) – NumPy array of shape (n, 2) or (n, 3).

  • weights (ndarray | None) – Optional 1D array of length n. If provided, computes weighted (regular) alpha-shapes; currently 3D only.

  • exact (bool) – Use CGAL’s exact kernel. Slower but robust against numerical pathologies.

  • periodic (bool) – Use periodic alpha-shapes on a torus.

  • compute_bounding_box (bool) – If True, use the bounding box of the data.

  • bbox_min – Bounding box if compute_bounding_box=False.

  • bbox_max – Bounding box if compute_bounding_box=False.

  • packed (bool) – Use the compact bit-packed cell encoding (the default) on the fast unweighted/non-periodic array path when the vertex ids fit a 64/128-bit word. Pass packed=False to force the fat encoding.

  • n_threads (int) – Threads used inside the Filtration constructor.

Returns:

A Filtration over alpha-shape simplices.