Structural Hawkes Processes for Learning Causal Structure from Discrete-Time Event Sequences

The python implementation of paper Structural Hawkes Processes for Learning Causal Structure from Discrete-Time Event Sequences. (IJCAI 2023)

Usage

The running example of SHP is given below.

from SHP import SHP_exp
from utils import get_performance

likelihood, fited_alpha, fited_mu, real_edge_mat, real_alpha, real_mu = SHP_exp(n=20, sample_size=20000,
                                                                                out_degree_rate=1.5,
                                                                                mu_range_str="0.00005,0.0001",
                                                                                alpha_range_str="0.5,0.7",
                                                                                decay=5, model_decay=0.35, seed=0,
                                                                                time_interval=5, penalty='BIC',
                                                                                hill_climb=True, reg=0.85)
res = get_performance(fited_alpha, real_edge_mat)

Create environment

The environment config is given in environment.yml and your can create the environment using:

conda env create -f environment.yml
conda activate shp

Real World Experiment

The real world experiment is implemented in real_world_data_exp.py. The dataset used for this experiment, 18V_55N_Wireless.tar.gz, can be found on: PCIC Causal Discovery Competition 2021.

To run the real experiment, make sure to download the dataset and place Alarm.csv and DAG.npy in the appropriate directory.

Citation

If you find this useful for your research, we would be appreciated if you cite the following papers:

@inproceedings{qiao2023shp,
  author       = {Jie Qiao and
                  Ruichu Cai and
                  Siyu Wu and
                  Yu Xiang and
                  Keli Zhang and
                  Zhifeng Hao},
  title        = {Structural Hawkes Processes for Learning Causal Structure from Discrete-Time
                  Event Sequences},
  booktitle    = {Proceedings of the Thirty-Second International Joint Conference on
                  Artificial Intelligence, {IJCAI} 2023, 19th-25th August 2023, Macao,
                  SAR, China},
  pages        = {5702--5710},
  publisher    = {ijcai.org},
  year         = {2023},
  url          = {https://doi.org/10.24963/ijcai.2023/633}
}