Code for the paper ACCEPTED to ICLR 2023 "Diffusion Models for Causal Discovery via Topological Ordering".
See example of how to use DiffAN class in demo_DiffAN.py. Once an instance diffan_obj of DiffAN is created, diffan_obj.fit() will
- train the diffusion model,
- run topological ordering and,
- perform pruning of spurious edges.
The output is the adjacency matrix of the causal graph.
Most hyperaparameters are hard-coded into constructor of the DiffAN class and we verified they work across a wide set of datasets.
However, as decribed in the paper, there is an approximation of DiffAN referred to as "DiffAN Masking" that has better scaling laws while maintaining similar performance. Setting the residue to False in
diffan = DiffAN(n_nodes, residue = False)
will use the approximation.
The repo use python and R code. To install the python dependencies, run:
conda env create -f environment.yml
The most important python dependencies are numpy, torch and functorch which can be easily installed manually if one prefer not to use conda. cdt is needed for using the R pruning code.
If the R dependencies are not installed, some of the pruning functions won't work. In this case, we advise to open an R session and install the specific packages from R directly.