scitex-linalg 0.1.4

Small linear-algebra helpers (distance, geometric median, cosine, nannorm) — standalone module from the SciTeX ecosystem

scitex-linalg

Small linear-algebra helpers — distances, NaN-aware norms, geometric median, vector projections.

Full Documentation · uv pip install scitex-linalg[all]

---

Installation

pip install scitex-linalg            # core (numpy/scipy/sympy)
pip install "scitex-linalg[torch]"   # + geometric_median (torch + geom-median)

Architecture

scitex_linalg/
├── _distance.py             ← euclidean_distance, cdist, edist, cosine
├── _misc.py                 ← nannorm, rebase_a_vec, three_line_lengths_to_coords
├── _geometric_median.py     ← torch geometric median (optional [torch] extra)
├── _vendor_decorators/      ← vendored numpy_fn / torch_fn / wrap (no scitex.* runtime dep)
└── _skills/                 ← agent-facing skill pages

Tiny single-purpose helpers. Pure numpy/scipy core; the geometric-median path opts into torch only when the [torch] extra is installed.

1 Interfaces

Python API

import scitex_linalg as sxl

sxl.euclidean_distance(u, v, axis=0)      # element-wise Euclidean distance
sxl.cdist(u, v)                           # pairwise distances
sxl.edist(u, v)                           # alias for cdist
sxl.cosine(v1, v2)                        # cosine similarity (NaN-safe)
sxl.nannorm(v, axis=-1)                   # NaN-aware vector norm
sxl.rebase_a_vec(v, v_base)               # project v onto v_base basis
sxl.three_line_lengths_to_coords(a, b, c) # triangle side lengths -> 2-D coords
sxl.geometric_median(xx, dim=-1)          # torch geometric median (requires [torch] extra)

Demo

flowchart LR
    A["u, v (np.ndarray)"] --> B["scitex_linalg.cdist"]
    B --> C["pairwise distance matrix"]
    A2["v with NaNs"] --> D["scitex_linalg.nannorm"]
    D --> E["NaN-safe vector norm"]
    A3["v, v_base"] --> F["scitex_linalg.rebase_a_vec"]
    F --> G["projected coords"]
    A4["xx (torch.Tensor)"] --> H["scitex_linalg.geometric_median"]
    H --> I["robust median point"]

>>> import numpy as np, scitex_linalg as sxl
>>> sxl.cosine(np.array([1, 0]), np.array([1, 1]))
0.7071...
>>> sxl.nannorm(np.array([3.0, np.nan, 4.0]))
5.0

Quick Start

import scitex_linalg as sxl

sxl.cdist(u, v)                # pairwise distances
sxl.cosine(v1, v2)             # cosine similarity (NaN-safe)
sxl.nannorm(v, axis=-1)        # NaN-aware norm
sxl.rebase_a_vec(v, v_base)    # project v onto v_base basis

Status

Standalone fork of scitex.linalg — intended to remain importable as scitex.linalg via the SciTeX umbrella package's bridge module. Decorators (numpy_fn, torch_fn, wrap) are vendored under _vendor_decorators/ to keep the package free of scitex.* runtime deps; when scitex-decorators is split out, those will be replaced with a direct dependency.

Part of SciTeX

scitex-linalg is part of SciTeX. Install via the umbrella with pip install scitex[linalg] to use as scitex.linalg (Python).

Four Freedoms for Research

0. The freedom to run your research anywhere — your machine, your terms.
1. The freedom to study how every step works — from raw data to final manuscript.
2. The freedom to redistribute your workflows, not just your papers.
3. The freedom to modify any module and share improvements with the community.

AGPL-3.0 — because we believe research infrastructure deserves the same freedoms as the software it runs on.

License

AGPL-3.0-only (see LICENSE).

---