DAG-NoCurl

Code for DAG-NoCurl work

Getting Started

Prerequisites

Python 3.7
PyTorch >1.0

How to Run

Synthetic linear data experiments. Please download the dataset at

https://drive.google.com/file/d/1O52SlAHPRw_iFW_sAfm_vR3oMnoEb8am/view?usp=sharing

For the synthetic nonlinear data experiments, codes will be shared at

DAG-GNN https://github.com/fishmoon1234/DAG-GNN

Synthetic linear data Experiments

CHOICE = nocurl, corresponding to the linear experiments, NoCurl-2 case in the paper

CHOICE = notear, corresponding to the linear experiments, NOTEARS case in the paper

LAMBDA1 = 10, corresponding to the parameter lambda_1 in the paper.

LAMBDA2 = 1000, corresponding to the parameter lambda_2 in the paper.

python main_efficient.py --data_variable_size=10 --graph_type="erdos-renyi" --repeat=100 --methods=<CHOICE> --h_tol=1e-8 --graph_degree=4 --lambda1=<LAMBDA1> --lambda2=<LAMBDA2> --data_type="synthetic"

Cite

If you make use of this code in your own work, please cite our paper:

@inproceedings{yu2021dag,
  title={DAGs with No Curl: An Efficient DAG Structure Learning Approach},
  author={Yue Yu, Tian Gao, Naiyu Yin and Qiang Ji},
  booktitle={Proceedings of the 38th International Conference on Machine Learning},
  year={2021}
}

Acknowledgments

Our work and code benefit from two existing works, which we are very grateful.

DAG NOTEAR https://github.com/xunzheng/notears
DAG NOFEAR https://github.com/skypea/DAG_No_Fear

skypea / DAG-NoCurl