Variational Causal Networks

Pytorch implementation of Variational Causal Networks: Approximate Bayesian Inference over Causal Structures (Annadani et al. 2021).

Yashas Annadani, Jonas Rothfuss, Alexandre Lacoste, Nino Scherrer, Anirudh Goyal, Yoshua Bengio, Stefan Bauer

Installation

You can install the dependencies using pip install -r requirements.txt

Create Directory structure which looks as follows: [save_path]/er_1/

Examples

Run

python main.py --num_nodes [num_nodes] --data_seed [data_seed] --anneal --save_path [save_path]

In the paper we run the model on 20 different data seeds to obtain confidence intervals. If you would like to compare with factorised distribution, run:

python main.py --num_nodes [num_nodes] --data_seed [data_seed] --anneal --save_path [save_path] --no_autoreg_base

Contact

If you have any questions, please address them to: Yashas Annadani [email protected]

If you use this work, please cite:

@article{annadani2021variational,
title={Variational Causal Networks: Approximate Bayesian Inference over Causal Structures},
author={Annadani, Yashas and Rothfuss, Jonas and Lacoste, Alexandre and Scherrer, Nino and Goyal, Anirudh and Bengio, Yoshua and Bauer, Stefan},
journal={arXiv preprint arXiv:2106.07635},
year={2021}
}