SDCD: Stable Differentiable Causal Discovery

SDCD is a method for inferring causal graphs from labeled interventional data.
You can read the associated preprint, "Stable Differentiable Causal Discovery", on arXiv.

If you find this work useful, please consider citing our work:

@article{nazaret2023stable,
  title={Stable Differentiable Causal Discovery},
  author={Achille Nazaret and Justin Hong and Elham Azizi and David Blei},
  journal={arXiv preprint arXiv:2311.10263},
  year={2023}
}

---

Quick Start

You can install the package via pip install sdcd.

For the main implementation of the method, see the SDCD class.

For a tutorial on the basic usage of SDCD, see this notebook.

Code used to generate paper figures can be found in this folder.