Installation¶

gwtc_analysis runs on Linux and macOS with Python 3.10 or later. It is distributed as a Python package on conda-forge and on PyPI, and as a Docker image:

Distribution	Where	Install
Conda package `gwtc_analysis`	conda-forge/gwtc_analysis	`conda install -c conda-forge gwtc_analysis`
PyPI package `gwtc_analysis`	pypi.org/project/gwtc_analysis	`pip install gwtc_analysis`
Docker image `gwtc-tool`	Docker Hub	`docker pull danielsentenac/gwtc-tool`
Source	GitHub	`pip install -e .`

The Docker image is the recommended choice for reproducibility and for workflow systems (CI pipelines, computing clusters).

From Docker¶

docker pull danielsentenac/gwtc-tool
docker run --rm -v "$PWD":/work -w /work danielsentenac/gwtc-tool gwtc_analysis -h

From conda-forge¶

conda install -c conda-forge gwtc_analysis
gwtc_analysis -h

From PyPI¶

pip install gwtc_analysis
gwtc_analysis -h

PyPI displays the project as gwtc-analysis, the spelling of its first registration; package names are normalized, so gwtc_analysis and gwtc-analysis are the same project, and the package, its files, the import name and the command are all gwtc_analysis.

The PyPI package declares only the light dependencies (numpy, pandas, matplotlib, minio, requests). The modes that read PE files, strain and skymaps also need the gravitational-wave software stack (GWpy, PESummary, PyCBC, ligo.skymap, LALSuite, h5py, astropy): install it first, for instance in an IGWN conda environment, or use the conda-forge package or the Docker image.

From source¶

The PE and strain modes need the gravitational-wave software stack (GWpy, PESummary, PyCBC, ligo.skymap, LALSuite), which is easiest to get from the IGWN conda environments:

git clone https://github.com/danielsentenac/gwtc_analysis
cd gwtc_analysis
conda activate igwn          # or any environment with the GW stack
pip install -e .
gwtc_analysis -h

Installing the package (from conda-forge, PyPI or Docker) provides the gwtc_analysis command used throughout this documentation. From a source checkout that is not installed, python -m gwtc_analysis.cli is equivalent.

icarogw, for the `hubble_constant` mode¶

The sample and combine stages of hubble_constant need icarogw [57] and bilby [58]. icarogw requires Python ≥ 3.12 and is not on PyPI, so it usually lives in an environment of its own:

conda create -n icarogw python=3.12
conda activate icarogw
export TMPDIR=~/tmp                                                  # the torch wheels are large
pip install torch --index-url https://download.pytorch.org/whl/cpu   # CPU torch first, not the CUDA build
pip install git+https://github.com/icarogw-developers/icarogw.git

The mode is then pointed to that interpreter with --icarogw-python ~/.conda/envs/icarogw/bin/python. It runs icarogw in CPU mode and puts the environment's lib/ on LD_LIBRARY_PATH itself.

numpy version

If other packages in the icarogw environment need an older numpy (for instance ligo.skymap), pin it: pip install numpy==2.1.1 scipy==1.14.1 worked.

Caches¶

Downloads are cached so that each file is fetched once:

Directory	Content
`~/.cache_gwtc_analysis/zenodo`	Zenodo version listings (one day) and sensitivity-injection files
`~/.cache_gwtc_analysis/pe_catalog`	PE samples extracted for `hubble_constant` (`$GWTC_PE_CACHE` or `--pe-cache` to move it)
`.cache_gwosc/`	Skymap tarballs and the PE index, per Zenodo record
`~/.gwcache`	Public GWTC-1 products for GW170817, supplementary PSDs