Causal discovery with a Differentiable Adjacency Test

The repo includes code implementing DAT and DAT-Graph from Scalable and Flexible Causal Discovery with an Efficient Test for Adjacency appearing at ICML 2024. Authors are Alan Nawzad Amin and Andrew Gordon Wilson.

The notebook run_DAT_graph.ipynb includes code to generate synthetic data and analyze it using DAT-Graph. To run the notebook you will need a GPU and you will need to install dependencies. To do so, you can run the following code

conda create --name dat_graph python==3.10 -y
conda activate dat_graph
conda install pip -y
pip install .
python -m ipykernel install --user --name dat_graph