Coupled Counterfactual Generative Adversarial Model (C²GAM)

Source code for A Generative Approach for Treatment Effect Estimation under Collider Bias: From an Out-of-Distribution Perspective.

Get Started

1. To install the necessary packages, run the following command-line code.

pip install -r requirements.txt

2. Run the demo (experiments on IHDP) in main.py.

Useful Links

• wasserstein_distance.py is adapted from SinkhornAutoDiff.

Citation

@InProceedings{pmlr-v235-li24al,
  title={A Generative Approach for Treatment Effect Estimation under Collider Bias: From an Out-of-Distribution Perspective},
  author={Li, Baohong and Li, Haoxuan and Wu, Anpeng and Zhu, Minqin and Peng, Shiyuan and Cao, Qingyu and Kuang, Kun},
  booktitle={Proceedings of the 41st International Conference on Machine Learning},
  pages={28132--28145},
  year={2024},
  organization={PMLR}
}