Installation¶

ZADU requires Python 3.10 or newer.

Base package¶

Install the default NumPy/SciPy execution path from PyPI:

python -m pip install zadu

The base package is sufficient for all 22 measures. Optional frameworks are not imported or installed unless you request them.

Optional visualization¶

Install Matplotlib and ZADUVis helpers with:

python -m pip install "zadu[vis]"

See Visualization for an example.

Optional accelerator backends¶

Apple Silicon users can install MLX separately:

python -m pip install "zadu[mlx]"

Install PyTorch support with:

python -m pip install "zadu[torch]"

An installed framework is never selected implicitly. Configure a backend with ExecutionConfig, and benchmark the actual data and metric specification before assuming it will be faster. The backend capability table documents supported devices, dtypes, fallbacks, and validation status.

Development installation¶

Clone the repository and install an editable development environment:

git clone https://github.com/hj-n/zadu.git
cd zadu
python -m pip install -e ".[dev]"

Documentation dependencies are separate:

python -m pip install -e ".[docs]"
python -m mkdocs serve