Gower Express

Fast Gower distance for mixed data in Python

Gower Express provides Gower distance utilities for mixed numerical and categorical data, with optional sklearn integration and accelerator paths for larger workloads.

Install

pip install gower_exp

Optional extras:

pip install gower_exp[gpu]
pip install gower_exp[sklearn]
pip install gower_exp[gpu,sklearn]

Quick Start

import pandas as pd
import gower_exp as gower

data = pd.DataFrame(
    {
        "age": [25, 30, 35, 40],
        "category": ["A", "B", "A", "C"],
        "salary": [50_000, 60_000, 55_000, 65_000],
        "city": ["NYC", "LA", "NYC", "Chicago"],
    }
)

distances = gower.gower_matrix(data)
similar = gower.gower_topn(data.iloc[0:1], data, n=3)

Features

• Pairwise Gower distance for mixed-type datasets
• Top-N similarity search helpers
• Optional sklearn-compatible interfaces
• Optional accelerator paths using numba, vectorized code, and GPU support when available

Project Layout

• gower_exp/: package source
• tests/: pytest suite
• benchmarks/: performance scripts
• docs/: user and contributor documentation
• examples/: notebooks and examples

Development

Set up a contributor environment with uv:

uv sync --all-extras --dev
pre-commit install

Run the standard checks before opening a pull request:

uv run ruff check .
uv run ruff format --check .
uv run bandit -r gower_exp/ -c pyproject.toml
uv run pytest tests/ --cov=gower_exp --cov-report=term-missing

Short imperative commit messages work best, for example fix sklearn edge case or bump test coverage. If a change affects performance-sensitive code, include benchmark notes in the pull request summary.

See docs/development.md for the full contributor workflow.

Releases

PyPI publishing is triggered by publishing a GitHub Release from a vX.Y.Z tag. Pushing a tag alone does not publish a package.

The expected maintainer flow is:

1. Update version in pyproject.toml.
2. Run just check and uv build.
3. Commit the version bump.
4. Create and push tag vX.Y.Z.
5. Publish a GitHub Release from that tag to trigger .github/workflows/publish.yml.

Documentation

• docs/README.md
• docs/development.md
• docs/sklearn-integration.md
• docs/benchmarks.md

License

MIT. See LICENSE.