Installation Instructions

Installing the Package

The package releases are uploaded to PyPi, therefore allowing you to install it with pip:

$ pip install fat-forensics

If you want to install the latest (development) version of the package please see the instructions provided in the Developer Guide. Among others, it explains how to install the package from source.

Package Requirements

Hard Dependencies

The package is built upon NumPy and SciPy with the following minimal requirements:

Package

Version

NumPy

1.10.0

SciPy

0.13.3

These requirements are listed in the requirements.txt file in our GitHub repository.

Soft Dependencies

Some parts (modules) of FAT Forensics depend on other packages. These are not installed by default. The following table shows soft dependencies of fatf modules:

fatf module

Required package

fatf.transparency.lime

lime

fatf.transparency.models.lime

fatf.transparency.predictions.lime

fatf.transparency.predictions.surrogate_explainers

scikit-learn>=0.19.2

fatf.transparency.sklearn

fatf.utils.data.feature_selection.sklearn

fatf.vis

matplotlib>=3.0.0

These dependencies can either be installed manually or alongside fat-forensics via pip:

$ pip install fat-forensics[xxx]

where xxx can be replaced with any of the following to pull appropriate soft dependencies during the package installation:

all

Installs all soft dependencies: lime, scikit-learn and matplotlib.

lime

Only installs lime as a soft dependency.

ml

Only installs scikit-learn as a soft dependency.

vis

Only installs matplotlib as a soft dependency.

dev

This option installs all the development requirements. Please consult the Developer Guide for more details.

The exact versions of these soft dependencies can be found in the requirements-aux.txt file.

Supported Platforms

At the moments we only test the package on Linux-based systems. From our experience the package also works on Mac OS and Windows, however these platforms are not officially supported.