nemopy

A column-vector-first NumPy wrapper. Vectors are (n, 1) by default; matrices are constructed column-by-column; arithmetic raises a clear ShapeError when shapes disagree instead of silently broadcasting.

Install

nemopy is not on PyPI yet — install it from the git repository. It requires Python >=3.10 (the repo pins 3.14 via .python-version) and is developed with uv. The only hard dependency is NumPy.

There are two modes. Mode A (pure-Python) is the default and needs no toolchain beyond Python; Mode B (Rust-enabled) additionally compiles the optional nemopy._rust_core extension to unlock the Tier-3 surface.

Mode A — pure-Python

uv add "git+https://github.com/nster101/nemopy"                  # core (numpy only)
uv add "nemopy[pandas] @ git+https://github.com/nster101/nemopy" # + pandas interop
uv add "nemopy[polars] @ git+https://github.com/nster101/nemopy" # + polars interop

This gives you the core types (ColVec, Mat), shape-guarded arithmetic, the Tier-2 decompositions (svd, qr, lu, cholesky, eigh) and the stats helpers — everything that wraps NumPy directly. Tier-3 methods (advanced decompositions, elimination, and future LP / network / Markov features) are not available in this mode: calling one raises a clear ImportError pointing at the build step, rather than silently falling back.

Mode B — Rust-enabled

To unlock the Tier-3 surface, build the optional Rust extension after cloning the repository. scripts/build_rust.sh is the canonical build command (it wraps maturin):

git clone https://github.com/nster101/nemopy
cd nemopy
uv sync                             # install Python deps (incl. dev toolchain)
scripts/build_rust.sh               # compile nemopy._rust_core via maturin

Verify which mode is active:

python -c "import nemopy._core as c; print('rust active:', c._RUST is not None)"

Prints rust active: True once the extension is built (Mode B) and rust active: False in pure-Python mode (Mode A).

For development (pytest + sphinx and the full toolchain), use the dev extra:

uv sync --extra dev

Quick start

import nemopy as nm

u = nm._c[1, 2, 3]                  # ColVec, shape (3, 1)
v = nm._c[4, 5, 6]                  # ColVec, shape (3, 1)

A = nm.mat(u, v)                    # Mat, shape (3, 2), columns are u and v
I = nm.eye(3)                       # 3x3 identity Mat

# MATLAB-style string syntax — columns separated by ';', elements by ',' or whitespace
B = nm._m["1, 2, 3; 4, 5, 6; 7, 8, 9"]   # Mat (3,3), columns [1,2,3] [4,5,6] [7,8,9]
B_rows = nm._m["1, 2, 3; 4, 5, 6; 7, 8, 9"].T   # rows [1,2,3] [4,5,6] [7,8,9]

# Inbound converters
c = nm.as_col([10, 20, 30])         # ColVec from list / Series / 1D array
M = nm.as_mat([[1, 2], [3, 4]])     # Mat from row-first nested list / DataFrame

# Outbound converters
u.to_list()                         # [1.0, 2.0, 3.0]
A.to_numpy()                        # plain ndarray, shape (3, 2)

# Matrix properties
A_sq = nm.mat(nm._c[1, 2], nm._c[3, 4])
A_sq.det                            # determinant
A_sq.inv                            # inverse Mat
A_sq.is_singular                    # bool

Why column-first?

Linear algebra is column-first: a matrix-vector product A @ x reads naturally when x is a column. nemopy enforces that convention end-to-end so column extraction (A[:, j]) returns a ColVec you can plug straight into @ without reshaping.

Documentation

The full behavioural specification lives in .github/DESIGN.md and .github/DESIGN_APPENDICES.md. Sphinx-built API docs are configured in docs/.

License

BSD 3-Clause. See LICENSE.