Installation¶

From PyPI (recommended)¶

pip install cdans

This pulls in the runtime dependencies — NumPy, SciPy, pandas, NetworkX, scikit-learn, tqdm. No causal-discovery libraries (no tigramite, no causal-learn) are required at runtime; PCMCI and KCI are both implemented in-house.

From source¶

git clone https://github.com/hferdous/CDANs.git
cd CDANs
pip install -e .

Use -e . for an editable install if you plan to modify the library; omit -e for a regular install.

Optional extras¶

Extra	Adds
`gp`	GPy for future GP-based kernel-width learning
`viz`	matplotlib + pydot for graph drawing
`dev`	pytest, ruff, mypy, build (for development)
`docs`	mkdocs + mkdocs-material + mkdocstrings (this site)
`all`	everything above

pip install "cdans[dev]"        # for development
pip install "cdans[docs]"       # to build this docs site
pip install "cdans[all]"        # everything

Verify the install¶

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

Run the test suite¶

pip install "cdans[dev]"
pytest

Around 90 tests, finishes in a few seconds. They cover the algorithm steps end-to-end on synthetic data, the KCI port (validated against the causal-learn reference on a battery of test cases), and the MATLAB-faithful direction inference.