🏍️ FastDAG2PAG-python

FastDAG2PAG-python is a Python tool for converting Directed Acyclic Graphs (DAGs) into Partial Ancestral Graphs (PAGs).
It is inspired by the dag2pag function from the R package pcalg 🧠, reimplemented in Python for improved accessibility and integration into modern workflows.

⚠️ Notes

• The skeleton learning implemented in this tool does not account for selection bias (I will address this in future updates), following the same theoretical foundation as the dag2pag in the R package pcalg.
  • Reference: Richardson, T. S., & Spirtes, P. (2002). Ancestral Graph Markov Models, Theorem 4.2.
• The core logic has been refactored and optimized for Python.

Additionally, I have discovered a potential issue in pcalg's dag2pag implementation. For details, please see compare_pcalg.ipynb in this repository.

🚀 Usage Example

See example_dag2pag.ipynb.

🛠️ Dependencies

• networkx==3.2.1
• scipy==1.11.4
• igraph==0.11.8
• pgmpy==0.1.19
• graphviz==0.20.3
• pydot==3.0.2
• causallearn==0.1.4.1(for DAGtoPAGPython part only)

🤝 Contributing & Contact

If you have questions or suggestions, feel free to open an issue or contact me directly.

Email: [zhengli0060(at)gmail(dot)com]