cafpono's Introduction

Causal Discovery via Normalizing Flows for Post-Nonelinear model (CAF-PoNo)

This is the implementation for our paper: "Enabling Causal Discovery in Post-Nonlinear Models with Normalizing Flows", accepted at ECAI 2024.

Setup • Structure • Experiments • Citation

Setup

conda env create -n cafpono --file env.yml 
conda activate cafpono

Clone and install dodiscover package at https://github.com/francescomontagna/dodiscover

Structure

.
├── data                        # Data used in the paper
├── env.yml                     # For environment setup
├── experiments                 # Main dir for experiments
│   ├── bivariate
│   ├── multivariate_dim
│   ├── multivariate_real
│   └── multivariate_sample_size
├── README.md
└── src                         # Source code for the implementation of CAF-PoNo and other baseline methods
    ├── data
    ├── HSIC.py
    ├── methods
    ├── pruning.py
    └── utils.py

Experiments

Experiment for the bivariate setting

python experiments/bivariate/main.py

Experiment for multivariate settings

Run experiment with different sample sizes

python experiments/multivariate_sample_size/main.py

Run experiment with different data dimensions
```
python experiments/multivariate_dim/main.py
```

Run experiment with real data

python experiments/multivariate_real/main.py

Run experiment for running time comparison

python experiments/multivariate_dim/timing.py

Citation

@inproceedings{hoang2024enabling,
      title={Enabling Causal Discovery in Post-Nonlinear Models with Normalizing Flows}, 
      author={Hoang, Nu and Duong, Bao and Nguyen, Thin},
      year={2024},
      booktitle = {European Conference on Aritificial Intelligence (ECAI)},
}

Recommend Projects