Installation

Quickstart

PyPI is the distribution channel for xplainable release versions. The best way to install it is with pip:

pip install xplainable

Optional dependencies

To use xplainable’s embedded GUI in jupyter, you will need to install xplainable with the gui extra:

pip install xplainable[gui]

To use xplainable’s advanced plotting functions, you will need to install xplainable with the plotting extra:

pip install xplainable[plotting]

Environment

Working in Jupyter

The xplainable GUI relies heavily on ipywidgets. Unfortunately, this requires a strict environment to ensure that it renders as expected. Ensuring that you have correctly installed the required dependencies before using the xplainable GUI is essential.

For now, we recommend using Python 3.8 when working with the GUI. This environment will yield the best visible results until we can stabilise ipywidgets for later versions of Python.

Known Tornado Issue with the xplainable GUI

Versions of tornado greater than 6.1 are known to cause issues with the jupyter kernel and ipywidgets. We recommend using tornado==6.1 for now.

Reproducible Installs

As libraries get updated, results from running your code can change, or your code can break completely. It’s essential to be able to reconstruct the set of packages and versions you’re using. Best practice is to:

  1. use a different environment per project you’re working on,

  2. record package names and versions using your package installer; each has it’s own metadata format for this:

  • Conda: conda environments and environment.yml

  • Pip: virtual environments and requirements.txt

  • Poetry: virtual environments and pyproject.toml