Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?


Failed to load latest commit information.
Latest commit message
Commit time

KiDS Cosmology Analysis Pipeline

This repository contains the cosmology inference pipeline that was used in the KiDS-1000 analyses:

The pipeline is built on CosmoSIS, albeit a modified version that is pip-installable and doesn't rely on environmental variables.

A MontePython likelihood that wraps the kcap functionality can be found at here. Note that the standard version of MontePython does not support non-flat priors yet, which is a problem for samplers that distiguish between likelihood and prior (such as MultiNest and PolyChord). A version that supports Gaussian priors with MultiNest can be found here.

The different modules (CosmoSIS standard library, etc) are included as git subtree. Users don't have to worry about this detail but if you make changes to any of the modules it helps to structure your commits such that they only touch on one module at a time, such that these changes can be easily backported to the individual repositories.

For a fiducial KV450 setup, have a look at runs/config/KV450_fiducial.ini.


Clone the repository:

git clone
cd kcap

It's strongly recommended to use some kind of encapsulated environment to install kcap, e.g., using conda. Here we assume that there is a anaconda installation available, that we need MPI support, and that we're on a machine with up-to-date GCC compilers. Notes on installations on macOS and details on how to set up things manually are here.

On machines with module support (e.g., cuillin), load the anaconda and openmpi modules first:

module load anaconda
module load openmpi

If there's no automated way to load these modules, make sure conda and MPI executables (mpif90) are on your PATH. For instructions on how to set up your own conda installation, see Install conda. If you're using your own anaconda installation, don't load the module as well, as this just causes conflicts.

Now set up the conda environment using the provided conda_env.yaml file:

conda env create -f conda_env.yaml

This creates a kcap_env environment that should have all the necessary dependencies. Activate the environment with source activate kcap_env.

We need to install CAMB because we use the new python interface for it. If kcap is to be used on a local machine, pip install camb is all there is to do. On a heterogenous cluster like cuillin, we need to build CAMB ourselves, however. To do so, run

git clone --recursive
python build_cluster
python install

We can now build kcap (which installs a standalone version of CosmoSIS):


To uninstall CosmoSIS (for example if you need to get the newest version), run pip uninstall cosmosis_standalone. To make a fresh installation of kcap, run python --clean.

Installation on macOS and other details

The default macOS compilers are supported now but gfortran still needs to be installed. This can be done with homebrew by running brew install gcc. Note that gcc 9.2 seems to be incompatible with the PolyChord samplers included in cosmosis, so use a different version (e.g., 9.1).

If no MPI support is required, run python --no-mpi.

Install conda

Get miniconda. For example on a Linux machine:

chmod +x

The installation will ask whether you want to add this conda installation to your PATH by adding some lines to your .bashrc. If you choose not to (for example because you don't want it to be the default installation), make sure the installation is accessible when building and running kcap.


Make sure conda and MPI are accessible (e.g., by running module load anaconda and module load openmpi) and that kcap_env is activated (source activate kcap_env). To test that everything is working, run the tests (todo...) and some of the configs in runs/config:

mkdir runs/output
cosmosis runs/config/KV450_no_sys.ini

For MPI:

mpirun -n 4 cosmosis --mpi runs/config/KV450_no_sys.ini

Repository structure


The different modules are organised as git subtrees.

Pull updates from a specific module repository

Using git subtree:

git subtree pull --prefix=cosebis --squash cosebis-remote kcap

where cosebis is the module to be updated, cosebis-remote is the remote for the module, and kcap the remote branch. The option --squash collapses the histories of the modules. Especially for the CosmoSIS standard library this is useful, since we don't want its whole history in kcap.

Using the subtree merge strategy:

git merge -s subtree --squash --allow-unrelated-histories cosebis-remote/kcap

The --allow-unrelated-histories seems to be necessary (probably because of the --squash option used earlier).

Push changes of modules in the kcap repository to the module repository

Using git subtree:

git subtree push --prefix=cosebis cosebis-remote remote_branch

where cosebis is again the module to has been updated, cosebis-remote is the remote for the module, and remote_branch the remote branch that the updates will get push to.

Using the subtree merge strategy:

git checkout -b backport cosebis-remote/kcap
git cherry-pick -x --strategy=subtree commits_to_push
git push

This is a bit more involved but allows for more control. First create a branch (backport) tracking the remote branch that is being targeted (cosebis-remote/kcap). Then cherry pick the commits (commits_to_push) that should be pushed to the module repository.


KiDS cosmology analysis pipeline






No releases published


No packages published