Installation and Deployment¶
SModelS is a Python library that requires Python version 2.6 or later, including version 3, which is the default. It depends on the following external Python libraries:
- pyhf>=0.4.3 (>=0.5.2 recommended!)
For speed reasons, we moreover recommend pytorch>=1.6.0 as backend for pyhf. This is, however, optional: if pytorch is not available, SModelS will use the default backend.
These tools need not be installed separately, as the SModelS build system takes care of that. The current default is that both Pythia6 and Pythia8 are installed together with NLLfast. Finally, the database browser provided by smodelsTools.py requires IPython, while the interactive plotter requires plotly and pandas.
The first installation method installs SModelS in the source directory. After downloading the source from the SModelS releases page and extracting it, run:make smodels
in the top-level directory. The installation will remove redundant folders, install the required dependencies (using pip install) and compile Pythia and NLL-fast. If the cross section computer is not needed, one can replace smodels with smodels_noexternaltools in the above command. In case the Python libraries can not be successfully installed, the user can install them separately using his/her preferred method. Pythia and NLL-fast can also be compiled separately running make externaltools.
If Python’s setuptools is installed in your machine, SModelS and its dependencies can also be installed without the use of pip. After downloading the source from the SModelS releases page and extracting it, run:setup.py install
within the main smodels directory. If the python libraries are installed in a system folder (as is the default behavior), it will be necessary to run the install command with superuser privilege. Alternatively, one can run setup.py with the “–user” flag:setup.py install --user
If setuptools is not installed, you can try to install the external libraries manually and then rerun setup.py. For Ubuntu, SL6 machines and other platforms, a recipe is given below.
Note that this installation method will install smodels into the default system or user directory (e.g. ~/.local/lib/python3/site-packages/). Depending on your platform, the environment variables $PATH, $PYTHONPATH, $LD_LIBRARY_PATH (or $DYLD_LIBRARY_PATH) might have to be set appropriately.
Finally, if pip3 (or pip) is installed in your machine, it is also possible to install SModelS directly without the need for downloading the source code:pip3 install smodels
in case of system-wide installs or :pip3 install --user smodels
for user-specific installations.
Note that this installation method will install smodels into the default system or user directory (e.g. ~/.local/lib/python3/site-packages/). Depending on your platform, the environment variables $PATH, $PYTHONPATH, $LD_LIBRARY_PATH (or $DYLD_LIBRARY_PATH) might have to be set appropriately. Be aware that the example files and the parameters file discussed in the manual will also be located in your default system or user directory. Furthermore the database folder is not included (see database installation below).
There is also a diagnostic tool available:
should list and check all internal tools (Pythia and NLL-fast) and external (numpy, scipy, unum, … ) dependencies.
In case everything fails, please contact firstname.lastname@example.org
Installing the SModelS Database¶
The installation methods explained above (except for pip install) also install SModelS’ database of experimental results in the smodels-database subdirectory. The first time SModelS is run, a binary file will be built using this text database folder, which can then be used in all subsequent runs. However, from v1.1.3 onwards it is recommended to provide the URL of the official database as the database path when running SModelS (see path in parameters file). In this case the corresponding database version binary file will be automatically downloaded and used. The available database URLs can be found in the SModelS Database releases page .
The complete list of analyses and results included in the database can be consulted at https://smodels.github.io/wiki/ListOfAnalyses. We note that all the results in the official database release have been carefully validated and the validation material can be found at https://smodels.github.io/wiki/Validation.
The database can conveniently be updated independently from SModelS code updates. It suffices to unpack any new database tarball and replace the database directory or provide the path to the new folder, binary or URL address. In the same fashion, one can easily add additional results as explained below.
Adding FastLim data¶
The official SModelS database can be augmented with data from the fastlim results. For using SModelS with the text database, a tarball with the properly converted fastlim-1.0 efficiency maps can be found in the smodels-database folder. The tarball then needs to be exploded in the top level directory of the database:
cd <smodels-database folder> tar -xzvf smodels-v1.1-fastlim-1.0.tgz rm smodels-v1.1-fastlim-1.0.tgz
Once the fastlim folders have been added to the database, SModelS auto-detects fastlim results and issues an acknowledgement.
As discussed above, from v1.1.3 onwards it is also possible to directly download the database binary file using the URLs provided in the SModelS Database releases page . Separate URLs are provided for the database including the Fastlim maps, so the user can choose which database to use.
When using the Fastlim results, please properly cite the fastlim paper; for convenience, a bibtex file is provided in the smodels-fastlim tarball.
Finally we point out that when converting the Fastlim efficiency maps efficiencies with a relative statistical uncertainty greater than 25% were set to zero. Also, per default we discard zeroes-only results.
Adding one’s own results¶
The Database of Experimental Results is organized as files in an ordinary directory hierarchy. Therefore, adding additional experimental results is a matter of copying and editing text files. Once the new folders and files have been added following the database structure format, SModelS automatically rebuilds the binary (Pickle) database file. The added results will then be available for using with the the SModelS tools.
System-specific Installation Instructions¶
Installation on Ubuntu >= 16.04¶
Installation on Ubuntu machines should be straightforward with superuser privileges (if you do not have superuser privileges see instructions below):
- sudo apt install gfortran python-setuptools python-scipy python-numpy python-docutils python-argparse
- setup.py install
Note that the last command can be run as superuser, or with the “–user” flag.
Installation on SL7¶
Installation on an SL7 or CentOS7 is straightforward:
- yum install gcc-c++ scipy numpy
- pip install unum pyslha argparse
Installation on SL6¶
Installation on an SL6 (Scientific Linux 6 or Scientific Linux CERN 6) machine is tricky, because SModelS requires a more recent version of scipy than is provided by SL6. We succeeded to install SModelS on SL6 by doing:
- yum install gcc-c++ libstdc++-devel libevent-devel python-devel lapack lapack-devel blas blas-devel libgfortran python-distutils-extra
- pip install nose unum argparse numpy pyslha scipy
Note, that these steps can safely be done within a Python
Pip can also be called with the “–user” flag.
Installation on SL5 and similar distributions¶
In some distributions like SL5, the Python default version may be smaller than
2.6. In these cases,
virtualenv has to be set up for a Python version >= 2.6. E.g. for Python 2.6, do
virtualenv --python=python2.6 <envname>, and modify by hand the first line in the executable from
Then perform the steps listed under
Installation on SL6.
Installation on other platforms or without superuser privileges using Anaconda¶
Another easy and platform independent way of installing SModelS without superuser priviledges is via Anaconda (https://www.continuum.io/downloads). Anaconda provides a local installation of pip as well as several additional python packages. Here we assume a version of gfortran is already installed in your system.
download and install Anaconda for Python 3.6 (https://www.continuum.io/downloads)
make sure Anaconda’s bin and lib folders are added to your system and Python pathsPATH="<anaconda-folder>/bin:$PATH" PYTHONPATH=$PYTHONPATH:"<anaconda-folder>/lib/python3.6/site-packages"
and then install SModelS as a user:
setup.py install --user
In order to make sure all libraries have been correctly installed, you can run:
Installation of the C++ interface¶
SModelS v1.1.1 comes with a simple C++ interface, see the cpp directory. Obviously, a C++ compiler is need, alongside with the python developers (header) files (libpython-dev on ubuntu, python-devel on rpm-based distros).