Installation guide#

This section provides detailed information about installing ArviZ. Most ArviZ functionality is available with the basic requirements, but ArviZ also has optional dependencies to further enhance the library. This guide will cover both basic and fully-fledged ArviZ installs and several installation methods.

Stable#

ArviZ can be installed either using pip or conda-forge.

Using pip#

pip install arviz

Use the below pip command to install ArviZ with all of its Optional dependencies.

pip install "arviz[all]"

Using conda-forge#

conda install -c conda-forge arviz

Development#

If you want to install the latest development version of ArviZ, use the following command:

pip install git+https://github.com/arviz-devs/arviz

Note: It can take sometime to execute depending upon your internet connection.

Dependencies#

Required dependencies#

The required dependencies for installing ArviZ are:

setuptools>=60.0.0
matplotlib>=3.5
numpy>=1.22.0,<2.0
scipy>=1.8.0
packaging
pandas>=1.4.0
xarray>=0.21.0
h5netcdf>=1.0.2
typing_extensions>=4.1.0
xarray-einstats>=0.3

and

python>=3.8

ArviZ follows NEP 29 and SPEC 0 to choose the minimum supported versions.

Optional dependencies#

The list of optional dependencies to further enhance ArviZ is given below.

numba
netcdf4
bokeh>=1.4.0,<3.0
contourpy
ujson
dask[distributed]
zarr>=2.5.0
xarray-datatree
  • Numba

    Necessary to speed up the code computation. The installation details can be found here. Further details on enhanced functionality provided in ArviZ by Numba can be found here.

  • Bokeh

    Necessary for creating advanced interactive visualisations. The Bokeh installation guide can be found over here.

  • UltraJSON

    If available, ArviZ makes use of faster ujson when arviz.from_json() is invoked. UltraJSON can be either installed via pip or conda.

  • Dask

    Necessary to scale the packages and the surrounding ecosystem. The installation details can be found at this link.