Installing cosmological codes and data
To keep it light, maintainable and easily extensible, cobaya does not include code or data of many of the cosmological components used; instead, it provides interfaces and automatic installers for the external packages they require: the original code itself, a cosmological dataset, etc.
Installing a basic set of codes and likelihoods
To install a basic set of cosmology requisites (CAMB, CLASS, Planck, BAO, SN), in a
/path/to/packages folders of your choice:
$ cobaya-install cosmo -p /path/to/packages
This does not install the Planck 2018 clik likelihoods, just the more recent NPIPE (PR4) native python version. The original packages can then be installed using e.g.
$ cobaya-install planck_2018_highl_plik.TTTEEE
If you have your input .yaml file that you want to run set up, you can also install just the likelihoods you need using
$ cobaya-install MyFile.yaml
If this fails (see last printed message), keep on reading this section. Otherwise, you can go straight to Basic cosmology runs.
On top of the pre-requisites of cobaya, you will need some others, which are indicated in the documentation of each of the components. You may already fulfil them, so you may try to go ahead with the installation process and just take a look at the pre-requisites of the components whose installation fails.
You will need an internet connection with a decent bandwidth (don’t use your phone’s): you may need to download several gigabytes!
Using the automatic installer
The automatic installation script takes one or more input files that you intend to run, makes a list of the external packages that you will need, and downloads and installs them one by one.
You need to specify a folder where the resulting files will be placed, which for the purposes of these instructions will be called
/path/to/packages. This does not need to be the folder in which you will run your samples.
This folder will be accessed whenever you call cobaya, and may take several gigabytes; so, if you are in a cluster, make sure that said folder is placed in a scratch file system with rapid access from the nodes and a generous quota (this normally means that you should avoid your cluster’s
When you have prepared the relevant input files, call the automatic installation script as
$ cobaya-install input_1.yaml input_2.yaml [etc] --packages-path /path/to/packages
You can skip the
--packages-path option if a
packages_path field is already defined in one of the input files.
If you would like to skip the installation of the dependencies of some components, you can use the
--skip "word1 word2 [...]" argument, where
word[X] are sub-strings of the names of the corresponding components (case-insensitive), e.g.
If you would like to automatically skip installing external packages that are avaliable globally (e.g. if you can do
import classy from anywhere) add
--skip-global to the command above.
cobaya-install will save the packages installation path used into a global configuration file, so that you do not need to specify it in future calls to
cobaya-run, etc. To show the current default install path, run
To override the default path in a subsequent call to
cobaya-run, the alternatives are, in descending order of precedence:
--packages-path /override/path/to/packagescommand line argument.
packages_path: /override/path/to/packagessomewhere in your input file.
define an environment variable
COBAYA_PACKAGES_PATH=/override/path/to/packages(declare it with
You can run the
cobaya-install script as many times as you want and it won’t download or re-install already installed packages, unless the option
-f) is used.
/path/to/packages, the following file structure will be created, containing only the packages that you requested:
│ ├── planck
│ ├── CAMB
│ ├── classy
│ ├── PolyChordLite
│ └── [...]
To run the installer from a Python script or notebook:
from cobaya.install import install
install(info1, info2, [etc], path='/path/to/packages')
info[X] are input dictionaries.
path is not passed, it will be extracted from the given infos (it will fail if more than one have been defined).
Installing requisites manually
The automatic installation process above installs all the requisites for the components used in the simplest way possible, preferring the system folders when possible (e.g. code that can be installed as a Python package).
If you want to modify one of the external packages (e.g. one of the theory codes) you will probably prefer to install them manually. Each component’s documentation has a section on manual installation of its requisites, and on how to specify your installation folder at run time. Check the relevant section of the documentation of each component.
When an installation path for a particular component is given in its corresponding input block, it takes precedence over automatic installation folder described above, so that if you already installed a version automatically, it will be ignored in favour of the manually specified one.
Updating and installing specific components
Individual likelihood or theory components can be installed using
$ cobaya-install component_name --packages-path /path/to/packages
This will also work with your own or third-party likelihood classes.
To force reinstallation of a package that is already installed, e.g. because you modified it manually, you can use the
--force) option, e.g.
$ cobaya-install camb -f --packages-path /path/to/packages
To upgrade an obsolete requisite, use the
- install.install(*infos, **kwargs)
Installs the external packages required by the components mentioned in
infoscan be input dictionaries, single component names, or names of yaml files
force – force re-installation of apparently installed packages (default:
test – just check whether components are installed (default:
upgrade – force upgrade of obsolete components (default:
skip – keywords of components that will be skipped during installation.
skip_global – skip installation of already-available Python modules (default:
debug – produce verbose debug output (default:
path – optional path where to install the packages (defaults to any packages_path entry given in the info dictionaries).
code – set to
Falseto skip code packages (default:
data – set to
Falseto skip data packages (default:
no_progress_bars – no progress bars shown; use when output is saved into a text file (e.g. when running on a cluster) (default:
no_set_global – do not store the installation path for later runs (default:
Trueif all components are installed without skips,
Falseif some components skipped, None (or exception raised) otherwise