pudu 0.1.0

A Python library for explainability of machine learinng algorithms in an agnostic, deterministic, and simple way.

A Python library for explainability of machine learinng algorithms in an agnostic, deterministic, and simple way.

• GitHub repo: https://github.com/pudu-py/pudu
• Documentation: https://pudu-py.github.io/pudu
• PyPI: https://pypi.python.org/pypi/pudu
• Conda-forge: https://anaconda.org/conda-forge/pudu
• Free software: MIT license

Introduction

pudu is a Python package that helps intepret and explore the results of machinme learning algorythms. It does this by quantifying the change in probability according to the change in the features. This library works with any case that has a probability function, which is normally available in scikit-learn and keras methods, and works for both 1-d (vectrors) and 2-d problems (images). In order to see the exact procedure and format needed, please refer to the examples in the docs.

Features

The following is a list of the main procedures that pudu package enables.

• Importance: estimates the absolute or relative importance of the features.
• Speed: calculates how fast a preditcion changes according to the changes in the features.
• Synergy: tests the synergy between features and the change in classification probability.
• Easy plotting of the results from the above.

Quickstart

1. Install this library using pip::

    pip install pudu
   

2. Install this library using conda-forge::

    conda install -c conda-forge pudu
   

3. Test it by running one of the examples in the docs.

4. If you find this library useful, please consider a reference or citation as:

    ...
   

5. Stay up-to-date by updating the library using::

   conda update pudu
   pip install --update pudu
   

6. If you encounter problems when updating, try uninstalling and then re-installing::

    pip uninstall pudu
    conda remove pudu
   

Credits

This package was created with Cookiecutter and the giswqs/pypackage project template.