Installing CosmoSIS

We are deeply indebted to Matt Becker for getting the Conda installation method to work.

Pip (core package only, no cosmology content)

If you only need CosmoSIS for its core sampling tools and datablock, and don’t need the cosmological likelihoods, then you can install those with pip:

pip install cosmosis

If you get an error including the phrase --single-version-externally-managed not recognized then you should first upgrade your setuptools package like this:

pip install -U setuptools wheel

and then try again.

Conda-Forge (from scratch)

This is the easiest way to get the full CosmoSIS package. It downloads everything you need. If the conda command works on your computer already, use the instructions for an existing installation (next section) instead.

Otherwise first run one of these commands to download the installer:

On Linux:

wget -O Miniforge3.sh  https://github.com/conda-forge/miniforge/releases/download/24.3.0-0/Miniforge3-Linux-x86_64.sh

On Intel Macs:

wget -O Miniforge3.sh  https://github.com/conda-forge/miniforge/releases/download/24.3.0-0/Miniforge3-MacOSX-x86_64.sh

On Silicon (M1 or M2) Macs:

wget -O Miniforge3.sh  https://github.com/conda-forge/miniforge/releases/download/24.3.0-0/Miniforge3-MacOSX-arm64.sh

Then, whichever you do, run these commands to install everything and download the standard library (see below for CosmoPower instructions):

chmod +x Miniforge3.sh
./Miniforge3.sh -b -p ./env
source ./env/bin/activate
mamba install -y cosmosis cosmosis-build-standard-library
source cosmosis-configure
cosmosis-build-standard-library main

This will make a new directory cosmosis-standard-library with the cosmology packages in. Explore that directory to start using CosmoSIS.

If you want to use CosmoPower then change the fourth line above to:

mamba install -y cosmosis cosmosis-build-standard-library cosmopower

Whenever you start a fresh terminal shell you need to run these commands to get set up again:

source ./env/bin/activate
source cosmosis-configure

Conda-Forge (existing installation)

If you already have conda installed on your computer, then you can create a new environment and install cosmosis tools in it with:

conda create -p ./env -c conda-forge cosmosis cosmosis-build-standard-library
conda activate ./env
source cosmosis-configure
cosmosis-build-standard-library

This will make a new directory cosmosis-standard-library with the cosmology packages in.

If you find this takes a very long time, you can try instead (mamba is a faster version of conda):

conda create -p ./env -c conda-forge mamba
conda activate ./env
mamba install -y cosmosis cosmosis-build-standard-library
source cosmosis-configure
cosmosis-build-standard-library main

Whenever you start a fresh terminal shell you need to run these commands to get set up again:

conda activate ./env
source cosmosis-configure

NERSC

There is a globally-accessible CosmoSIS installation for the NERSC machine Perlmutter. You can access it by running:

source /dvs_ro/cfs/projectdirs/des/zuntz/cosmosis-global/setup-cosmosis3

This will set all the necessary environment variables; you can then clone the cosmosis-standard-library and the make command should work:

git clone https://github.com/cosmosis-developers/cosmosis-standard-library
cd cosmosis-standard-library
make

You need source the setup script each time you want to use the system, including in batch scripts and interactive jobs.

If you need your own python environment to install new dependencies, you can do that with:

conda create -p ./env --clone $CFS/des/zuntz/cosmosis-global/env-latest-${NERSC_HOST}

That will make a new environment in the ./env directory, which you can start using by doing:

source /dvs_ro/cfs/projectdirs/des/zuntz/cosmosis-global/setup-cosmosis3 ./env

You can then pip or conda install things in your new environment.

If your dependency may be generally useful please open an issue and we can install it centrally also: https://github.com/cosmosis-developers/cosmosis/issues/

Jupyter at NERSC

After setting up the conda environment as described above, you can create a Jupyter “kernel” which lets you run notebooks in that environment with:

python -m ipykernel install --user --name cosmosis-$(cosmosis --version)

Then you can select the kernel in the Jupyter notebook interface at jupyter.nersc.gov.

Installing manually on clusters and supercomputers

The conda-forge version above often works on larger machines, though not always. Give it a go first (and test to see if multi-node jobs work).

If not, you will need these dependencies, which many systems will have already:

First, export these environment variables:

  • GSL_INC the path to GSL header files

  • GSL_LIB the path to GSL library files

  • CFITSIO_INC the path to CFTSIO header files

  • CFITSIO_LIB the path to CFTSIO library files

  • FFTW_LIBRARY the path to FFTW header files

  • FFTW_INCLUDE_DIR the path to FFTW library files

  • LAPACK_LINK whatever command line you need to link to LAPACK

  • CXX Command for your C++ compiler

  • CC Command for your C compiler

  • FC Command for your Fortran compiler

  • MPIFC Command for your MPI Fortran compiler

  • COSMOSIS_ALT_COMPILERS=1

and run:

python -m venv ./env
source env/bin/activate
pip install cosmosis camb astropy fitsio cython scikit-learn fast-pt

Then clone and build the standard library repository:

source cosmosis-configure
git clone https://github.com/cosmosis-developers/cosmosis-standard-library
cd cosmosis-standard-library
make

Please open an issue if you have installation problems.

The cosmosis-configure command can also set you up to use HomeBrew to install things: source cosmosis-configure --brew