# Install
pip install pici
# Quickstart
import pici
print(pici.__version__)
Why pici?
-
Efficient
State-of-the-art algorithms that provide the best possible computing time for partial identifiability.
-
Lightweight
Pure Python and small footprint for easy installation and usage.
-
Practical
Helpers for common inference patterns; integrates into existing analysis stacks.
-
Open
Apache-2.0 licensed on GitHub — transparent development and community contributions.
Project updates & work done
- Initial v0 release published to PyPI.
- Complete toolkit for causal inference implemented.
- Unit tests and documentation added.
- New improvements to the main algorithm and features currently in development.
Useful links
Theoretical Background
The partial identifiable causal inference algorithm used in pici is based on the following work:
J. P. Arroyo, D. Mauá, J. G. B. Rodrigues, D. A. E. Lawand, J. Lee, R. Marinescu, A. G. Gray, E. R. Laurentino, and F. Cozman, “Multilinear and Linear Programs for Partially Identifiable Queries in Quasi-Markovian Structural Causal Models,” CAR @ UAI 2025 Workshop, 2025.📄 View on OpenReview
@inproceedings{arroyo2025multilinear,
title = {Multilinear and Linear Programs for Partially Identifiable Queries in Quasi-Markovian Structural Causal Models},
author = {Arroyo, João Pedro and Mauá, Denis and Rodrigues, João Gabriel Barreto and Lawand, Daniel A. E. and Lee, Junkyu and Marinescu, Radu and Gray, Alexander G. and Laurentino, Eduardo Rocha and Cozman, Fabio},
booktitle = {CAR @ UAI 2025 Workshop},
year = {2025},
url = {https://openreview.net/forum?id=aUPT1kEiwP},
note = {Poster}
}Acknowledgments
ICTI & C2D: We thank the Instituto de Ciência e Tecnologia Itaú (ICTI) for providing key funding for this work through the C2D – Centro de Ciência de Dados at the University of São Paulo.
Disclaimer: Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of Itaú Unibanco and the Instituto de Ciência e Tecnologia Itaú.
Data & Ethics: All data used in this study comply with Brazil’s General Data Protection Law (Lei nº 13.709/2018 – LGPD).
C4AI, FAPESP & IBM: The authors also thank the Center for Artificial Intelligence ( C4AI-USP ) and acknowledge the support from the São Paulo Research Foundation (FAPESP grant #2019/07665-4) and from the IBM Corporation.