For certain systems, we offer pre-built packages of Shogun. This is the easiest way to start using it. In other cases, we describe how to build Shogun from source code.
The base shogun library and its Python interface are available through the conda package manager, via conda-forge. To install both:
conda install -c conda-forge shogun
or to get just the library:
conda install -c conda-forge shogun-cpp
These packages include most of the optional dependencies and are currently available for Linux, macOS, and Windows.
We are working on integrating Shogun with Debian/Ubuntu. In the meantime, we offer a prepackaged PPA. These currently do contain the C++ library and Python bindings. Add this to your system as
sudo add-apt-repository ppa:shogun-toolbox/stable
sudo apt-get update
Then, install as
sudo apt-get install libshogun18
The Python (2) bindings can be installed as
sudo apt-get install python-shogun
In addition to the latest stable release, we offer nightly builds of our development branch.
The latest packages for Debian jessie
and stretch
are available in our own repository at http://apt.shogun.ml.
We provide both the stable and nightly packages, currently only for amd64 architecture.
In order to add the stable packages to your system, simply run the following commands
sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 3DD2174ACAB30365
echo "deb http://apt.shogun.ml/ stretch main" | sudo tee /etc/apt/sources.list.d/shogun-toolbox.list > /dev/null
sudo apt-get update
After this just simply install the shogun library
sudo apt-get install libshogun18
The nightly packages are available in the nightly
component, i.e. deb http://apt.shogun.ml/ stretch nightly
Shogun is part of Fedora 25. Install as
sudo dnf install shogun
Shogun is part of Homebrew. Install the latest stable version as
brew install shogun
Shogun natively compiles under Windows using MSVC, see the Windows section. We currently only support binary packages via conda. If you are interested in packaging, documenting, or contributing otherwise, please contact us.
You can set up Shogun using our Docker images as:
sudo docker pull shogun/shogun:master
sudo docker run -it shogun/shogun:master bash
The docker image follows both the master
and the develop
branch of the repository, just specify the desired branch name as a tag for the image. For example, to use the develop version of shogun simply pull the shogun/shogun:develop
docker image.
There's an SDK docker image for shogun development as well, which we use to run our Travis CI jobs.
Sometimes mounting a local folder into the docker image is useful. You can do this via passing an additional option
-v /your/local/folder:/same/folder/in/docker
See the Docker documentation for further details.
You can download and install shogun using the vcpkg dependency manager:
git clone https://github.com/Microsoft/vcpkg.git
cd vcpkg
./bootstrap-vcpkg.sh
./vcpkg integrate install
./vcpkg install shogun
The shogun port in vcpkg is kept up to date by Microsoft team members and community contributors. If the version is out of date, please create an issue or pull request on the vcpkg repository.
Shogun is can be automatically built from the following language's source.
You can install it from PyPI. There is limited control over options and it might take a long time as everything is done from scratch.
pip install shogun-ml
We do not recommend this option and suggest to rather compile by hand as described below.
In case none of the binary packages listed on our website work for your system, or you want to modify Shogun, you will need to build it from the source.
The standard GNU/Linux tools and Python are minimal requirements to compile Shogun.
By default, building the meta examples is enabled, requiring python-ply
and ctags
. Disable using -DBUILD_META_EXAMPLES=Off
if those requirements are a problem, also see generating meta examples.
To compile the interfaces, in addition to swig itself, you will need language-specific development packages installed, see interfaces below.
There is a larger number of optional requirements. The output of cmake output lists optional dependencies that were found and not found. If a particular Shogun class is unavailable, this is likely due to an unmet dependency. See our docker configuration file for an example configuration used in our test builds.
You need at least 1GB free disk space. If you compile any interface, roughly 4 GB RAM needed (we are working on reducing this). CCache will massively speed up the compilation process and is enabled by default if installed.
Shogun uses CMake for its build. The general workflow is now explained. For further details on testing etc, see DEVELOPING.md.
Download the latest stable release source code, or (as demonstrated here) clone the latest development code. Potentially update submodules
git clone https://github.com/shogun-toolbox/shogun.git
cd shogun
git submodule update --init
Create the build directory in the source tree root
mkdir build
Configure cmake, from the build directory, passing the Shogun source root as an argument.
It is recommended to use any of CMake GUIs (e.g. replace cmake ..
with ccmake ..
), in particular, if you feel unsure about possible parameters and configurations.
Note that all cmake options read as -DOPTION=VALUE
.
cd build
cmake [options] ..
Compile
make
Install (prepend sudo
if installing system-wide), and you are done.
make install
Sometimes you might need to clean up your build (e.g. in case of some major changes). First, try
make clean
If that does not help, try removing the build directory and starting from scratch afterward
rm -rf build
If you prefer to not run the sudo make install
command system-wide, you can either install Shogun to a custom location (-DCMAKE_INSTALL_PREFIX=/custom/path
, defaults to /usr/local
), or even skip make install
at all.
In both cases, it is necessary to set a number of system libraries for using Shogun, see INTERFACES.md.
The native C++ interface is always included.
The cmake options for building interfaces are -DINTERFACE_PYTHON=ON -DINTERFACE_R ..
etc.
For example, replace the cmake step above by
cmake -DINTERFACE_PYTHON=ON [potentially more options] ..
The required packages (here debian/Ubuntu package names) for each interface are
- Python
python-dev python-numpy
- For dealing with customized Python environments, see here
- Octave
octave liboctave-dev
- R
r-base-core
- Java
oracle-java8-installer
, non-standard, e.g.https://launchpad.net/~webupd8team/+archive/ubuntu/java
jblas
, a standard third party library,https://mikiobraun.github.io/jblas/
- Ruby
ruby ruby-dev
, andnarray
a non-standard third party library,http://masa16.github.io/narray/
, install withgem install narray
- Lua
lua5.1 liblua5.1-0-dev
- C-Sharp
mono-devel mono-gmcs cli-common-dev
To use the interfaces, in particular, if not installing to the default system-wide location, see INTERFACES.md. See examples below for how to create the examples from the website locally.
All Shogun examples at our website are automatically generated code. You can
generate them (plus additional ones) locally (needs cmake switch -DBUILD_META_EXAMPLES=ON
)
make meta_examples
This requires PLY for Python, package python-ply
, and ctags, package ctags
.
Both source code and potential executables (C++, Java, C-Sharp) are created in build/examples/meta/
when running make
.
See INTERFACES.md to run the generated examples and see EXAMPLES.md for more details on their mechanics. See DEVELOPING.md for how the examples are used as tests.
In case you have a problem building Shogun, please open an issue on GitHub with your system details, exact commands used, and logs posted as a gist.
CMake is a beast. Make sure to read the docs and CMake_Useful_Variables. Make sure to understand the concept of out of source builds. Here are some tips on common options that are useful
Specify a different swig executable:
cmake -DSWIG_EXECUTABLE=/usr/bin/swig_custom
To specify a different compiler, see CMake FAQ, "How do I use a different compiler?". You might have to delete the build directory or clear the cmake cache otherwise for this to work.
CC=/path/to/gcc CXX=/path/to/g++ cmake ..
In case header files or libraries are not at standard locations one needs
to manually adjust the library and include paths, -DCMAKE_INCLUDE_PATH=/my/include/path
and -DCMAKE_LIBRARY_PATH=/my/library/path
.
Often, there are multiple Python versions installed on the system. There are various reasons for this, i.e. Linux without root access, MacOS + homebrew, using Anaconda or virtualenv. If Shogun is executed using a version of Python different from the one it was built, there will be import crashes. If this is your setup, you need to make sure that Shogun is both built and executed against the Python environment of your choice. For that, you need to do something similar to
cmake -DPYTHON_INCLUDE_DIR=/path/to/python/include/dir -DPYTHON_LIBRARY=path/to/python/libpythonVERSION.{so|dynlib} -DPYTHON_EXECUTABLE=/path/to/python/executable -DPYTHON_PACKAGES_PATH=/path/to/python/dist-packages ..
For example, for brew
installed Python under MacOS, use something like:
cmake -DPYTHON_INCLUDE_DIR=/usr/local/Cellar/python/2.7.5/Frameworks/Python.framework/Headers -DPYTHON_LIBRARY=/usr/local/Cellar/python/2.7.5/Frameworks/Python.framework/Versions/2.7/lib/libpython2.7.dylib -DINTERFACE_PYTHON=ON ..
Under Linux, where you want to use Python 3, which is not the system's default:
cmake -DPYTHON_INCLUDE_DIR=/usr/include/python3.3 -DPYTHON_EXECUTABLE:FILEPATH=/usr/bin/python3 -DPYTHON_PACKAGES_PATH=/usr/local/lib/python3.3/dist-packages -DINTERFACE_PYTHON=ON ..
On a Linux cluster without root access, using Anaconda (note you will need to activate your environment every time you want to run Shogun):
source path/to/anaconda/bin/activate
cmake -DCMAKE_INSTALL_PREFIX=path/to/shogun/install/dir -DPYTHON_INCLUDE_DIR=path/to/anaconda/include/python2.7/ -DPYTHON_LIBRARY=path/to/anaconda/lib/libpython2.7.so -DPYTHON_EXECUTABLE=path/to/anaconda/bin/python -DINTERFACE_PYTHON=ON ..
Please see any of our Windows py3X Azure Pipelines build to get any other information on the build process.
It is recommended to use Visual Studio 16 2019" or "MSBuild".
-
Install miniconda
-
Open the start menu, and run Anaconda Prompt (Miniconda3)
-
In the Windows Start menu, find the path of the Native Tools Command Prompt for VS shortcut that is relative to your system architecture and Visual Studio version (x32/x64, 2017/2019/..). For example, for x64 Native Tools Command Prompt for VS 2019, it looks so:
%comspec% /k "X:\Path\To\Microsoft Visual Studio\2019\Community\VC\Auxiliary\Build\vcvars64.bat"
. For detailed information check out link 1 and link 2 -
Execute this path in the Anaconda Prompt to run Native Tools there
-
Put these lines into the Anaconda Prompt:
rem - Records comments (remarks); @rem - do not print command just its result (the rem result is empty)
Be careful cmake can search out of the build directory to load libraries thus remove other directories that contain the shogun sources
@rem [FILL THIS SECTION WITH THE VALUES YOU WANT
@rem For example: SET MAIN_DIR=D:\Build
SET MAIN_DIR=X:\Path\To\Build
@rem Basically your number of cores
SET MAX_CPU_COUNT=8
@rem Take a look at https://cmake.org/cmake/help/latest/manual/cmake-generators.7.html#visual-studio-generators
SET PLATFORM=-G "Visual Studio 16 2019" -A x64
@rem FILL THIS SECTION WITH THE VALUES YOU WANT]
SET REPO_DIR=%MAIN_DIR%\shogun
SET VENV_DIR=%MAIN_DIR%\envs\shogun
SET CLCACHE_DIR=%MAIN_DIR%\clcache
SET SourcesDirectory=%REPO_DIR%\build
SET BinariesDirectory=%MAIN_DIR%\binaries
SET targetPrefix=%BinariesDirectory%\opt
SET clcacheArtifactName=clcache-vs17
SET buildConfiguration=Release
SET SourceBranchName=develop
git clone https://github.com/shogun-toolbox/shogun %REPO_DIR%
CHDIR /d %REPO_DIR%
git submodule update --init --force --depth=5
conda create --quiet --prefix %VENV_DIR% --mkdir --yes python=3.6.* setuptools numpy scipy eigen snappy zlib ctags ply jinja2 gtest mkl-devel swig -c conda-forge
activate %VENV_DIR%
%REPO_DIR%\.ci\setup_clcache.cmd
MKDIR %BinariesDirectory%
CHDIR /d %BinariesDirectory%
%REPO_DIR%\.ci\get_latest_artifact.py %SourceBranchName% %clcacheArtifactName%
@rem If there is no tar program in your Windows do:
@rem conda install -c haasad eidl7zip --yes
@rem FOR /F "delims=" %i IN ('where .\%clcacheArtifactName%:*.tar*') DO 7za x %i -so | 7za x -aoa -si -ttar -o%CLCACHE_DIR%
@rem Or simply extract the file whose name is the output of the 'where .\%clcacheArtifactName%:*.tar*' command
FOR /F "delims=" %i IN ('where .\%clcacheArtifactName%:*.tar*') DO tar --extract --file=%i --directory %CLCACHE_DIR% --gzip
MKDIR %targetPrefix% %SourcesDirectory%
CHDIR /d %SourcesDirectory%
@rem It is necessary to have '/' not '\' in the DBLAS_LIBRARIES and DLAPACK_LIBRARIES attributes
cmake %PLATFORM% -DCMAKE_BUILD_TYPE=%buildConfiguration% -DCMAKE_PREFIX_PATH=%VENV_DIR%\Library -DENABLE_TESTING=ON -DCMAKE_INSTALL_PREFIX=%targetPrefix% -DBUILD_META_EXAMPLES=OFF -DBLAS_LIBRARIES=%VENV_DIR%/Library/lib/mkl_core_dll.lib -DLAPACK_LIBRARIES=%VENV_DIR%/Library/lib/mkl_core_dll.lib ..
cmake --build . --config %buildConfiguration% --target INSTALL -- -p:TrackFileAccess=false -p:CLToolExe=clcache.exe -maxcpucount:%MAX_CPU_COUNT%
- Now you can use the Conda environment (located in VENV_DIR) containing shogun library in your Python projects!