Repository files navigation

LensKit is a set of Python tools for experimenting with and studying recommendersystems. It provides support for training, running, and evaluating recommenderalgorithms in a flexible fashion suitable for research and education.

LensKit for Python (LKPY) is the successor to the Java-based LensKit project.

To install the current release withuv (recommended):

$uv pip install lenskit

Or, to add it to your project's dependencies and virtual environment:

$uv add lenskit

Classicpip also works:

$python -m pip install lenskit

Then seeGetting Started

You can also install LensKit fromconda-forge withpixi:

$pixi add lenskit

Orconda:

$conda install -c conda-forge lenskit

To use the latest development version, you have two options. You can installdirectly from GitHub:

$uv pip install -U git+https://github.com/lenskit/lkpy

Or you can use our PyPI index, by adding topyproject.toml:

[[tool.uv.index]]name ="lenskit"url ="https://pypi.lenskit.org/lenskit-dev/"

Binary wheels of LensKit development (and release) versions are automaticallypushed to this index, although they are not guaranteed to be permanentlyavailable. Reproducible code should generally depend on released versionspublished to PyPI.

We also provide mirrors of the PyTorch package repositories that are filtered toonly include PyTorch and directly supporting dependencies, without otherpackages that conflict with or mask packages from PyPI, and with fallbacks forother platforms (i.e., our CUDA indices include CPU-only MacOS packages). Thismakes it easier to install specific versions of PyTorch in your project withthe index priority and fallthrough logic implemented byuv. To make yourproject only use CPU-based PyTorch, you can add topyproject.toml:

[[tool.uv.index]]name ="torch-cpu"url ="https://pypi.lenskit.org/torch/cpu/"

Or CUDA 12.8:

[[tool.uv.index]]name ="torch-gpu"url ="https://pypi.lenskit.org/torch/cu128/"

These indices provide the same package distributions as the official PyTorchrepositories (in fact, they link directly to the PyTorch packages). They arejust an alternate index view that reduces some package conflicts.

To contribute to LensKit, clone or fork the repository, get to work, and submita pull request. We welcome contributions from anyone; if you are looking for aplace to get started, see theissue tracker.

Our development workflow is documented inthe wiki; the wiki alsocontains other information ondeveloping LensKit. User-facing documentation isathttps://lkpy.lenskit.org.

We useuv for developing LensKit and managingdevelopment environments. Ourpyproject.toml file contains the Pythondevelopment dependencies; you also need a working Rust compiler (typically viarustup). Before setting up to work on LensKit, youtherefore need:

• Git
• uv
• rustup and a working Rust compiler (rustup install stable)
• A working C compiler compatible with Python
  • On Windows, this is either Visual Studio (with C++ development) or theVisual C++ Build Tools. See theRustup Windows installinstructions for details.
  • On Mac, install Xcode.
  • On Linux, see your system package manager instructions.

>winget install Git.Git astral-sh.uv Rustlang.Rustup>rustup install stable-msvc

$brew install git uv rustup

Once you have the dependencies installed, set up your LensKit developmentenvironment:

$uv venv -p 3.12$uv sync

If you want all extras (may not work on Windows), do:

$uv sync --all-extras

You can then activate the virtual environment to have the tools available andrun tools likepytest:

$. ./.venv/bin/activate

You should always test your changes by running the LensKit test suite:

pytest tests

If you want to use your changes in a LensKit experiment, you can locally installyour modified LensKit into your experiment's environment. We recommend usingseparate environments for LensKit development and for each experiment; you willneed to install the modified LensKit into your experiment's repository:

uv pip install -e /path/to/lkpy

• Documentation
• Discussion and Announcements

This material is based upon work supported by the National Science Foundationunder Grant No. IIS 17-51278. Any opinions, findings, and conclusions orrecommendations expressed in this material are those of the author(s) and do notnecessarily reflect the views of the National Science Foundation.