Installation

Standard Installation

You can install polygraph using pip:

pip install polygraph-benchmark

Check that it's installed correctly by running:

import polygraph

Installation with Mamba to include `graph_tool`

Since graph_tool (which can be optionally used by polygraph to do SBM validity checking, as it uses community detection algorithms that are commonly used in the graph generation literature) requires C++ dependencies, you need to install it using conda or mamba.

mamba create -n <your_model_env> python=3.10
mamba activate <your_model_env>
mamba install -c conda-forge graph-tool

Check that it's installed correctly by running:

import graph_tool.all as gt

Note that graph_tool is GPL licensed, so any code that uses it must also be GPL licensed. If you don't want to include it in your project, you can install polygraph without it using the pip command above. We provide BSD-licensed implementations of the SBM graph validity checking functions in is_valid_alt.

Installation with Pixi (Conda + PyPI)

Pixi can manage conda and PyPI packages together and ensures graph_tool is installed via conda-forge first.

# One-time: install pixi (see docs at https://pixi.sh)

# From repo root with pixi.toml present
pixi install

# Enter the environment
pixi shell

# Run tests or docs
pixi run test
pixi run docs

The provided pixi.toml pins Python 3.10, installs graph-tool from conda-forge, and installs this package in editable mode with [dev] extras.

🛠️ Development Installation

We recommend to install the full gamut of dependencies including optional development dependencies for development purposes to ensure all tests can be run. Therefore we provide here the steps to create a mamba environment and install the dependencies there.

# Clone the repository
git clone https://github.com/polygraph-eval/polygraph-benchmark.git
cd polygraph

# Create a new mamba environment
mamba create -n polygraph python=3.10
mamba activate polygraph

# Install graph_tool first to avoid issues down the line
mamba install -c conda-forge graph-tool

# Install the rest of the package and its pip-dependencies.
pip install -e .[dev]

### Running the tests You can then run the tests to ensure everything is installed correctly. To run the tests in parallel, you can use:

pytest -n 10 -sv --skip-slow ./tests/

To run the tests sequentially, you can use:

pytest -svx --skip-slow ./tests/

Installation with uv

uv is a fast pip/venv manager. Since graph_tool is not available on PyPI for all platforms, install it with conda first if you need SBM validation via graph-tool; otherwise skip it.

Option A: With conda for graph_tool first, then use uv for Python deps:

mamba create -n polygraph python=3.10 graph-tool -c conda-forge
mamba activate polygraph
pip install uv
uv pip install -e .[dev]

Option B: Pure uv (no graph_tool support):

pip install uv
uv venv
uv pip install -e .[dev]

Installation with pip (no graph_tool)

If you don't need graph_tool, a standard virtualenv works:

python -m venv .venv
source .venv/bin/activate
pip install -e .[dev]