SPAMS

Downloads

For any question regarding SPAMS, you can contact us at "spams.dev'AT'inria.fr" (replace 'AT' by @).

Version 2.6.2: R and Matlab versions remain at release 2.6 for the moment. SPAMS-python version 2.6.2 corresponds to a library reorganization and installation pipeline improvement. The Python version of SPAMS now has a dedicated Git forge.

Version 2.6.1: R and Matlab versions remain at release 2.6 for the moment. SPAMS-python version 2.6.1 corresponds to a reunification of the sources for python2 and python3 (a single package for both python versions).

Version 2.6: The toolbox was successfully tested with gcc-4.7, gcc-5.4, and the BLAS/LAPACK library atlas. Windows pre-compiled packages are not available yet.

Version <=2.5: The toolbox was successfully tested with gcc-4.6, the intel compiler version 12 and visual studio compiler 10, and the BLAS/LAPACK library atlas, Intel MKL and AMD ACML.
The best performance was obtained using the Intel compiler + Intel MKL under Linux. Since the toolbox was mostly tested with Matlab under Linux, please use the matlab Linux 64bits version if you want to report speed results.
To use SPAMS with matlab R2011a and later, do not forget to run the command setenv('MKL_NUM_THREADS','1') before using SPAMS.

Starting from version 2.6, the project files are hosted on INRIA GitLab with public git access. This repository is now considered the default SPAMS repository.
Starting from version 2.2 and until version 2.6, the project files are hosted on INRIA Gforge. If you are interested in more information, svn access, click here. It is recommended to use the Git repository for now on (especially for version >2.6). The svn repository is kept for legacy reason.

Version: 2.6.2

/!\ Matlab and R packages are still in version 2.6 (c.f. below).

/!\ SPAMS-python version 2.6.2 was only tested on GNU/Linux (Ubuntu 18.04) at the moment. Compatibility with MacOS should be OK, however Windows is not supported at the moment.



Regarding the SPAMS-python package:

• SPAMS is available on PyPI: you can run `pip install spams` to install it.
• The library reorganization and installation pipeline improvement was done by François Rheault and Samuel Saint-Jean.
• Please check the new dedicated Git forge for more details, especially regarding installation requirements. In particular, to improve performance, we recommend to install the MKL Intel library (available in the Anaconda Python distribution) before installing SPAMS.
• Tar-gzipped sources of python interface (python3.x compatible, MKL compatible).

Version: 2.6.1

We have fixed a few reported bugs during the past months and updates the sources accordingly. The following links point to the most recent version of SPAMS.

/!\ Matlab and R packages are still in version 2.6.

/!\ SPAMS-python version 2.6.1 was only tested on GNU/Linux (Ubuntu 16.04) at the moment. Compatibility with MacOS and Windows is still under testing. Recommendations for installation on these two operating systems will soon be uploaded. A specific version is available to be used in Anaconda Python distribution (tested with Anaconda2-5.0.1 and Anaconda3-5.0.1) to benefit from the MKL Intel library available in Anaconda.


• For Python, SPAMS is available on PyPI: you can do `pip install spams` or `pip install spams_mkl` to use the specific version compatible with the MKL Intel library if you have it.
• Tar-gzipped, sources, matlab interface,
• Tar-gzipped, precompiled matlab package for Linux,
• Tar-gzipped, precompiled matlab package for MacOS X (>=10.9.5),
• Tar-gzipped, sources, R interface (compatible with R-3.x),
• Tar-gzipped, sources, python interface to be used with Anaconda Python distribution, to benefit from the MKL Intel libray (python2.x and python3.x compatible, 64bits system),
• Tar-gzipped, sources, python interface (python2.x and python3.x compatible, 64bits system),

+ Matlab precompiled package for GNU/Linux and MacOS X
+ R interface compatible with R-3.x
+ Python interface compatible with python2.x and python3.x (same sources)
+ Disclaimer: the multi-threading is not available with the Matlab precompiled package for MacOS X (because OpenMP is not supported by the compiler clang at the moment)



Regarding the SPAMS-python package:

• In addition, a version of the SPAMS Python library (available here) is maintained by John Kirkham on the conda-forge (an open source community-led packaging effort supplying release quality binary packages for use on the platforms Mac and GNU/Linux with the conda package manager).


• Regarding installation on Windows, you can have a look at the Scipy webpage dedicated to compilation and use of BLAS on Windows for help or you can contact us.


• Regarding the installation on 32bits systems, you can edit the 'setup.py' file (basically you just have to replace the compilation option '-m64' by '-m32') or you can contact us.


• The SPAMS Python library will also be soon available on the PyPI website for simple installation with the 'pip' package manager (with 'pip install spams').
  A testing version is already available on the PyPI testing website. It can be installed with the following command:

  pip install --index-url https://test.pypi.org/simple/ spams

  (tested on GNU/Linux with python2.7 and python3.x).

Version: 2.6

We have fixed a few reported bugs during the past months and updates the sources accordingly. The following links point to the most recent version of SPAMS.


• Tar-gzipped, sources, matlab interface,
• Tar-gzipped, precompiled matlab package for Linux,
• Tar-gzipped, precompiled matlab package for MacOS X (>=10.9.5),
• Tar-gzipped, sources, R interface (compatible with R-3.x),
• Tar-gzipped, sources, python interface (python2.x compatible version),
• Tar-gzipped, sources, python3 interface (python3.x compatible version),

+ Matlab precompiled package for Linux and MacOS X
+ R interface compatible with R-3.X
+ Python interface compatible with python2.x or python3.x (different sources)
+ Disclaimer: the multi-threading is not available with the Matlab precompiled package for MacOS X (because OpenMP is not supported by the compiler clang at the moment)

Version: 2.5

• Tar-gzipped, sources, matlab interface,
• Tar-gzipped, sources, R interface,
• Tar-gzipped, sources, python interface,

+ implementation of archetypal analysis (by Yuansi Chen)
+ solver for regularized Poisson regression
+ first functions of the Image toolbox
+ fix compatibility with Matlab 2014a + MacOS Maverick + visual studio 2013

Version: 2.4

• Tar-gzipped, sources, matlab interface,
• Tar-gzipped, sources, R interface,
• Tar-gzipped, sources, python interface,
• exe, binary windows 64 bits package for R, with instructions,
• exe, binary windows 23 bits package for R, with instructions,
• exe, binary windows 64bits package for Python, with instructions,
• exe, binary windows 32bits package for Python, with instructions,
• python graphical interface for dictionary learning.

+ online dictionary learning with structured regularization (coded by Jean-Paul Chieze)
+ python graphical interface for dictionary learning (coded by Jean-Paul Chieze)
+ incremental and stochastic proximal gradient algorithms for Matlab (mexIncrementalProx and mexStochasticProx)
+ a few bug corrections.

Version: 2.3

• Tar-gzipped, sources, matlab interface,
• Tar-gzipped, sources, R interface,
• Tar-gzipped, sources, python interface,
• exe, binary windows 64 bits package for R, with instructions,
• exe, binary windows 32bits package for Python, with instructions,
• exe, binary windows 64bits package for Python, with instructions,

+ new documentation.
+ new functionalities for omp, ompmask and trainDL.
+ feature selection tools for graphs (in the matlab version).
+ a few bug corrections.

Version: 2.2

• Tar-gzipped, sources, matlab interface,
• Tar-gzipped, sources, R interface,
• Tar-gzipped, sources, python interface,

+ R and Python interface developed by Jean-Paul Chieze (INRIA).
+ Windows 64bits compatibility thanks to some advices from Alex Bronstein and Min-Hyuk Sung.
+ easier compilation scripts for Linux/Mac OS/Windows 32 and 64 bits, compatible with gcc/icc/vcc, and different BLAS/LAPACK libraries.
+ Trace Norm regularization has been added to proximal toolbox.
+ New options have been added to the algorithm FISTA.
+ A few bug corrections.

Version: 2.1

• Tar-gzipped, sources,
• Tar-gzipped, Linux 64bits, Intel Compiler, MKL,

+ Bug corrected in mexLasso, when reaching the end of the regularization path. Numerical Stability has been improved.

Version: 2.0

• Tar-gzipped, Linux 64bits, Intel Compiler, MKL,
• Tar-gzipped, Linux 64bits, GNU Compiler, ATLAS,
• Tar-gzipped, Linux 32bits, Intel Compiler, MKL,
• Tar-gzipped, Linux 32bits, GNU Compiler, ATLAS,
• Tar-gzipped, Mac OS 32bits, Intel Compiler, MKL,
• Tar-gzipped, Mac OS 64bits, Intel Compiler, MKL,

+ Proximal Toolbox

Version: 1.02

• Tar-gzipped, Linux 64bits, Intel Compiler, MKL,
• Tar-gzipped, Linux 64bits, GNU Compiler, ATLAS,
• Tar-gzipped, Linux 32bits, Intel Compiler, MKL,
• Tar-gzipped, Linux 32bits, GNU Compiler, ATLAS,
• Tar-gzipped, Mac OS 32bits, Intel Compiler, MKL,
• Tar-gzipped, Mac OS 64bits, Intel Compiler, MKL,
• Tar-gzipped, Windows 32bits, Intel Compiler, MKL,

+ Elastic-Net, a few heuristics are added.

Warning: the file test_TrainDL in the win32 version was incorrect. This has been corrected.

Warning: It seems that Matlab >= R2009b is required for the win32 version.

Version: 1.01

• Tar-gzipped, Linux 64bits, Intel Compiler, MKL,
• Tar-gzipped, Linux 64bits, GNU Compiler, ATLAS,
• Tar-gzipped, Linux 32bits, Intel Compiler, MKL,
• Tar-gzipped, Linux 32bits, GNU Compiler, ATLAS,
• Tar-gzipped, Mac OS 32bits, Intel Compiler, MKL,
• Tar-gzipped, Mac OS 64bits, Intel Compiler, MKL,

+ functions are robust to bad input parameters.

Version: 1.00

• Tar-gzipped, Linux 64bits, Intel Compiler, MKL,
• Tar-gzipped, Linux 64bits, GNU Compiler, ATLAS,
• Tar-gzipped, Linux 32bits, Intel Compiler, MKL,
• Tar-gzipped, Linux 32bits, GNU Compiler, ATLAS,
• Tar-gzipped, Mac OS 32bits, Intel Compiler, MKL,