Code title: "Riemannian gossip algorithms for decentralized matrix completion"

(c) 2017-2018 Bamdev Mishra <bamdevm@gmail.com>, Hiroyuki Kasai <kasai@is.uec.ac.jp>, Pratik Jawanpuria <pratik.iitb@gmail.com>, and Atul Saroop <asaroop@amazon.com>.

This package contains a MATLAB implementation of the algorithm presented in the report.

B. Mishra, H. Kasai, P. Jawanpuria, and A. Saroop
"A Riemannian gossip approach to decentralized subspace learning on Grassmann manifold,"
Technical report, arXiv preprint arXiv:1705.00467, 2017.

Installation:
-------------

- Set current directory as your current folder in Matlab or put it in your Matlab path.
- Run "run_me_first.m" to add folders to the working path. This needs to be done at the starting of each session.
- Run "install_mex.m" to compile mex files. This needs to be done only once.
- To check that everything works, run "test_online.m" at Matlab command prompt (you should see plots at the end).



Files:
------
- proposed/online_decentralized_MC.m:   the online Gossip algorithm.
- proposed/parallel_decentralized_MC.m: the parallel Gossip algorithm.
- proposed/Grassmann_update.m:          the standard updates on Grassmann.
- proposed/Grassmann_preconditioned_update: the preconditioned updates.
- propsed/mystepsizeguess: a guess for computing the initial stepsize by linearizing the cost function.



Disclaimer:
-----------

- All the files are written by the authors except
    - the mex files in the mex_files_rtrmc folder, which are written by Nicolas Boumal <nicolasboumal@gmail.com>.
    - D-LMaFit in /benchmarks/d-lmafit, which is written by A.-Y. Lin <inanya@mail.ustc.edu.cn>.

- Manopt is downloaded from http://manopt.org.
- Partitoc is downloaded from http://uk.mathworks.com/matlabcentral/fileexchange/27472-partictoc/content/Par.m.


    


