oineus.diff.cube_filtration

oineus.diff.cube_filtration(data, negate=False, wrap=False, max_dim=None, values_on='vertices', n_threads=1)[source]

Differentiable cubical filtration from tensor-valued grid data.

Mirrors oineus.cube_filtration, but returns a DiffFiltration whose filtration values are gathered back onto the autograd-tracked input.

Parameters:
  • data – 1D/2D/3D tensor-like (PyTorch/JAX/NumPy) of grid values.

  • negate (bool) – Compute upper-star (superlevel) instead of lower-star.

  • wrap (bool) – Not supported; raises if True.

  • max_dim (int | None) – Maximal cube dimension; defaults to data.ndim.

  • values_on (str) – "vertices" (default) or "cells". Critical indices are flat indices into the original data array in either case.

  • n_threads (int) – Threads for C++ filtration construction.

Return type:

DiffFiltration