Genentech/dcdfg

Differentiable Causal Discovery with Factor Graphs

This repository contains an implementation of the structure learning method described in "Large-Scale Differentiable Causal Discovery of Factor Graphs".

If you find it useful, please consider citing:

@inproceedings{Lopez2022largescale,
  author = {Lopez, Romain and Hütter, Jan-Christian and Pritchard, Jonathan K. and Regev, Aviv}, 
  title = {Large-Scale Differentiable Causal Discovery of Factor Graphs},
  booktitle = {Advances in Neural Information Processing Systems},
  year = {2022},
}

Requirements

Python 3.9+ is required. To install the requirements:

pip install -r requirements.txt

wandb is required for now (a PR to make remove this requirement is welcome). Follow the steps here.

Running DCD-FG

SEMs simulations (full usage in files)

1. 'python make_lowrank_dataset.py'
2. 'python run_gaussian.py'

Biological dataset

1. 'perturb-cite-seq/0-data-download.ipynb'
2. 'perturb-cite-seq/1-assignments-vs-variability.ipynb'
3. 'python run_perturbseq_linear.py'

Acknowledgments

• This repository was originally forked from DCDI. Please refer to the license file for more information.
• Most files in this codebase have been rewritten for:

1. vectorization and scaling to large graphs
2. incorporating the semantic of factor graphs
3. refactoring and implementation in pytorch lightning
4. implementation of DCD-FG, NOTEARS, NOTEARS-LR and NOBEARS

• We are grateful to the authors of the baseline methods for releasing their code.