High-Dimensional Causal Bayesian Optimization

The code for the paper "High-Dimensional Causal Bayesian Optimization" published in ECAI 2024. The main paper and appendix are contained in paper folder.

Preparations

Environment Preparation

Create the running environment with conda 23.9.0, python 3.8.18:

conda env create -f env.yaml
conda activate HCBOenv

Dataset Preparation

High dimensional dataset initialization:

python reproduce.py

Real-world dataset initialization:

python initialize_real.py CoralGraph
python initialize_real.py HealthGraph

Experiments

CID Validation

Remember to create ./result/EffDim folder before reproducing the result of CID validation experiments.

python run_eff_dim.py CoralGraph
python run_eff_dim.py HealthGraph
python run_eff_dim.py additive-50-124
python run_eff_dim.py additive-100-8
python run_eff_dim.py linear-100-124
python run_eff_dim.py non-additive-50-122
python run_eff_dim.py non-additive-100-124

python run_eff_dim_test.py linear-200-2

Performance Experiments

Run baseline experiments in this form:

python run.py <problem_name> --run_performance

For example:

python run.py additive-100-8 --run_performance

Ablation Study Experiment

python run_ablation.py additive-100-8

Hyperparameter Experiment

python run_hyperparameter.py linear-100-124

Visualization and statitical tests

Please refer to result_analysis_baseline.ipynb to visualize and conduct t-tests on baseline experiment results.

Please refer to result_analysis_others.ipynb to visualize the result of ablation study and hyper-parameter experiments.

Citations

@inproceedings{wuwang2024hcbo,
 author = {Yupeng Wu and Weiye Wang and Yangwenhui Zhang and Mingjia Li and Yuanhao Liu and Hong Qian and Aimin Zhou},
 booktitle = {Proceedings of the 27th European Conference on Artificial Intelligence (ECAI)},
 title = {High-Dimensional Causal Bayesian Optimization},
 year = {2024},
 address = {Santiago de Compostela, Spain}
}