Gradient-Based Neural DAG Learning

This code was written by the authors of the ICLR 2020 submission: "Gradient-Based Neural DAG Learning". Our implementation is in PyTorch but some functions rely on the Causal Discovery Toolbox which relies partly on the R programming language.

Run the code

To use our implementation of GraN-DAG, simply install Singularity (instructions: and run the code in our container (download it here: Use (update the paths) to launch the differents methods (GraN-DAG, DAG-GNN, NOTEARS, CAM).