Installation¶

From PyPI¶

pip install scikit-route

2.0.0 is the first release of scikit-route published on PyPI. Binary wheels are provided for CPython 3.11, 3.12, 3.13 and 3.14 on:

Platform	Architectures	Wheel tag
Linux	x86_64, aarch64	`manylinux_2_28` (glibc 2.28 or newer: every distribution supported in 2026)
macOS	x86_64 (Intel), arm64 (Apple silicon)	`macosx`
Windows	AMD64	`win_amd64`

With a wheel you need no compiler. The runtime dependencies are numpy>=1.26, scipy>=1.11 and joblib>=1.3; numpy 2 is supported.

Check the installation:

python -c "import skroute; print(skroute.__version__)"

Extras¶

Extra	Installs	When you need it
`pandas`	`pandas>=2.2`	`fit` on a `DataFrame` cost matrix (labels from the index) and `load_*(as_frame=True)`
`google`	`googlemaps>=4.10`	`skroute.preprocessing.google.GoogleDistanceMatrix`, the optional client that fetches driving distances and times
`test`	pytest, pytest-cov, hypothesis, pandas	running the test suite against your installation
`dev`	`test` + ruff, mypy, cython-lint, pre-commit, build, cibuildwheel, Cython	contributing (see the contributing guide)
`docs`	mkdocs-material, mkdocstrings, mkdocs-gen-files, mkdocs-include-markdown-plugin	building this site

pip install "scikit-route[pandas]"
pip install "scikit-route[pandas,google]"

DataFrames are recognised by duck typing, so a plain pip install scikit-route never imports pandas.

From source¶

The core of scikit-route is a Cython 3 extension. Building it needs a C compiler and Cython 3.1 or newer; pip fetches Cython and setuptools itself in an isolated build environment, so you only provide the compiler:

Linux: gcc or clang (sudo apt install build-essential, sudo dnf groupinstall "Development Tools").
macOS: the Xcode command line tools (xcode-select --install).
Windows: the "Desktop development with C++" workload of the Visual Studio Build Tools.

# the released source distribution
pip install --no-binary scikit-route scikit-route

# the development version
git clone https://github.com/arubiales/scikit-route.git
cd scikit-route
pip install .

No numpy headers are needed at build time: the kernels work on typed memoryviews, so a wheel built against one numpy version runs with any supported numpy.

If import skroute fails with "scikit-route's compiled core is missing", the extension was not built: install a wheel, or install a compiler and run pip install . again from the source tree.

Supported Python versions¶

Python	Status
3.11, 3.12, 3.13, 3.14	supported; wheels published; tested on Linux, macOS and Windows in CI
3.15 pre-release	tested as an allowed failure; no wheel until the final release
3.10 and older	not supported (`requires-python = ">=3.11"`)

conda¶

There is no conda package yet. Install with pip inside your conda environment; a conda-forge feedstock is planned after the PyPI release.

Upgrading from 1.0¶

2.0 changes the fit signature, the return value and the objective. Read the migration guide before upgrading: the one-line summary is that 1.0's fit(route, time, cost) becomes fit(cost, time_matrix=time, depot=route[0]), fit returns the estimator, and the results are in route_, trips_ and cost_.