Module core 

Backend-agnostic hyperbolic primitives.

This module deliberately avoids committing to a tensor/array backend. Everything operates on slices and returns Vec<T>.

The ndarray-based API remains available behind the ndarray feature.

References (for rationale and cross-checking formulas)

• Nickel & Kiela (2017): Poincaré Embeddings for Learning Hierarchical Representations.
• Ganea, Bécigneul, Hofmann (2018): Hyperbolic Neural Networks (Poincaré + Lorentz tooling).
• Gromov (1987): Hyperbolic groups (δ-hyperbolicity; four-point conditions).

Notes:

• The diagnostics submodule is intentionally “small n only” ((O(n^4)) exact δ) and exists to validate whether a dataset is plausibly tree-like before committing to heavier machinery.

Modules

conversions

Core conversions between Poincaré and Lorentz models.

diagnostics

Diagnostics for “tree-likeness”.

Structs

LorentzModelCore

Lorentz (hyperboloid) model operations on slice inputs.

PoincareBallCore

Poincaré ball operations on slice inputs.