Tree approximate message passing (Tree-AMP)

Implements gaussian expectation propagation for any tree-like probabilistic graphical model.

Documentation website: https://sphinxteam.github.io/tramp.docs

Requirements

• python>=3.6
• numpy/pandas/scipy/matplotlib
• networkx==1.11
• daft

Warning Currently the package does not support networkx 2.xx and will throw errors. We plan to upgrade to networkx 2.xx at some point.

Install

To install the package, go to the folder where setup.py is located and run:

pip install .

or if you want to install in development mode (changes to the repository will immediately affect the installed package without needing to re-install):

pip install --editable .

To install the package on a remote machine directly from the github repo:

pip install git+https://github.com/sphinxteam/tramp.git

See installing from sources for more details. In both cases, the necessary requirements should be automatically installed.

Citation

The package is presented in this article. To cite this work:

@article{baker2023tree,
  title={Tree-AMP: Compositional inference with tree approximate message passing},
  author={Baker, Antoine and Krzakala, Florent and Aubin, Benjamin and Zdeborová, Lenka},
  journal={Journal of Machine Learning Research},
  volume={24},
  number={57},
  pages={1--89},
  year={2023}
}

Examples

See the corresponding gallery in the documentation website.

Acknowledgments

Both the SPHINX team and the SMILE team acknowledge funding from: