TMAC

Welcome to TMAC: A Toolbox of Modern Async-Parallel, Coordinate, Splitting, and Stochastic Methods

Features

A rich set of predefined operators.
A set of predefined operator splitting schemes.
An asynchronous coordinate update framework.
A synchronous parallel update driver.
Options to load datasets in various format.
A rich set of applications build on top of TMAC.
Support for Matlab, Python and Julia (coming soon)

Platforms

TMAC can be installed to run on the following platforms:

Linux
Mac OS X
Windows (MINGW32 (32-bit) guide, MINGW64 (64-bit) guide, Cygwin (32/64-bit) guide)

Building TMAC requires

GNU-compatible Make
GCC (version 4.7 or above)
Any BLAS library

Build

Building TMAC can be as simple as:

make

Use the Applications

We build a user-friendly interface to run TMAC for different applications through shell. You can run it through the following command:

./run_me.sh

Documentation

We use sphinx for documentation. To use it, you need to install sphinx and sphinx_rtd_theme. To compile the source code for the documentation, you can run the following commands::

cd doc
make html

Then the html files are in the build/html folder.

Contributing Code

We welcome patches. If you plan to contribute a patch, you need to write tests for any new code. Changes should be verified to not break existing tests before they are submitted for review. We use Google Test for unit testing. Our unit tests are in the test folder. If you are adding tests to a new file, you will need to modify the Makefile to compile the source code, otherwise, nothing needs to be changed in the Makefile. The following commands should build and run the unit tests.

The details for contributing to TMAC is here.

The contributors' list is here.

Getting the Data

So far, TMAC support the following data format:

You can download some example datasets from here You can also obtain the regression and classification datasets from the LIBSVM website.

Troubleshooting

You can report bugs through the issue tab
fatal error: <omp.h> file not found
- First make sure you are using the correct compiler, i.e., GCC >= 4.7 from the GNU Project. For OSX, please follow the instruction here to download and install gcc-4.9. For Ubuntu, you can follow this link.
install g++ without root privileges. This link is very helpful.

Acknowledgement

We would like to acknowledge the Eigen library, the sparse BLAS library, and Google Test.

Related Papers


@article{peng2015arock,
  title = {ARock: an Algorithmic Framework for Asynchronous Parallel Coordinate Updates},
  author = {Peng, Zhimin and Xu, Yangyang and Yan, Ming and Yin, Wotao},
  journal = {arXiv:1506.02396},
  year = {2015},
  publisher = {http://arxiv.org/abs/1506.02396}
}

@article{PengWuXuYanYin2016_coordinate,
  title = {Coordinate friendly structures, algorithms and applications},
  volume = {1},
  number = {1},
  journal = {Annals of Mathematical Sciences and Applications},
  author = {Peng, Zhimin and Wu, Tianyu and Xu, Yangyang and Yan, Ming and Yin, Wotao},
  month = jan,
  year = {2016},
  pages = {59--119}
}

FlorenceFeng/TMAC