Demonstration of the Swept-AMP algorithm.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
matlab
python
.gitignore
readme.md
readme.pdf

readme.md

SwAMP Demo User's Manual

Artist's Rendition of SwAMP

Using this demo is supposed to be straightforward: one needs only to open Matlab, go to the current folder and run the command demo.

When the demo starts, a compilation will take place. SwAMP is written in C and must be compiled using Matlab's MEX API. If you have a C compiler on your computer, everything should (hopefully) go smoothly! We have tested the compilation using gcc in different platforms, but we'd expect it to work with other compilers as well. Make sure to run mex -setup if you have no previously used Matlab's MEX feature.

If you have problems, you can try the Python version which, in spite of being much slower, achieves the same results.

The SwAMP repository is hosted on GitHub..

Key Reference

A. Manoel, F. Krzakala, E. W. Tramel, L. Zdeborová, "Sparse Estimation with the Swept Approximated Message-Passing Algorithm," arXiv preprint 1406.4311.

Contributors to this Repository

  • Andre Manoel, original source author [andremanoel@gmail.com]
  • Eric W. Tramel, maintainer [eric.tramel@gmail.com]

A few details

  • The demo script calls functions from the the examples folder. By exploring these, one may get a better grasp of how to use SwAMP.

  • SwAMP's source code is located on the src folder; in particular, the bulk of the algorithm is contained in the src/solvers/amp.c file. This version follows exactly the listings in the paper, and is already optimized to work with sparse matrices. Additionally, 3 other versions are present in the same folder:

    • gamp.c, which implements G-SwAMP;
    • amp_dense.c, a version that isn't optimized for sparse matrices;
    • and amp_alt.c, a slight modification of the algorithm that, in spite of reaching the same results, sometimes converges faster.