Installation#

Quickstart#

The easiest and recommended way to install TJPCov is to install it via conda:

conda install -c conda-forge tjpcov

Alternatively, you may also install TJPCov via PyPi:

pip install tjpcov

will install TJPCov with minimal dependencies, and:

pip install 'tjpcov[full]'

will include all dependencies (for details, see Optional dependencies (PyPi only) section).

Note

If you plan to use TJPCov with MPI on NERSC, extra work must be done to get MPI running in your conda environment. See the NERSC docs for how to install MPI there.

Supported Python Versions#

TJPCov currently runs on python 3.8, but python 3.9, 3.10, and 3.11 are supported.

TJPCov also has a few specific software versions hardcoded. Please check the pyproject.toml file to see version requirements.

Optional dependencies (PyPi only)#

Because TJPCov relies on some packages that may not be necessary for every user, we have added different installation options to accommodate different use cases. For example, if a user has no plans to use MPI with TJPCov, they do not need mpi4py. Below we list the different installation options available on PyPi.

  • pip install tjpcov will install tjpcov and the minimal dependencies.

  • pip install tjpcov'[doc]' will install tjpcov, the minimal dependencies, and the dependencies needed to build the documentation.

  • pip install 'tjpcov[nmt]' will install tjpcov, the minimal dependencies, and the dependencies needed to use NaMaster.

  • pip install 'tjpcov[mpi4py]' will install the minimal dependencies and the mpi4py library to use MPI parallelization. Does not work on NERSC (see note above)

  • pip install 'tjpcov[full]' will install tjpcov and all dependencies.