PermutationImportance

Important variables determined through data-based variable importance methods


Keywords
predictor, importance, variable, model, evaluation
License
MIT
Install
pip install PermutationImportance==1.2.1.8

Documentation

PermutationImportance

Build Status Documentation Status

PermutationImportance Logo

Welcome to the PermutationImportance library!

PermutationImportance is a Python package for Python 2.7 and 3.6+ which provides several methods for computing data-based predictor importance. The methods implemented are model-agnostic and can be used for any machine learning model in many stages of development. The complete documentation can be found at our Read The Docs.

Version History

  • 1.2.1.8: Shuffled pandas dataframes now retain the proper row indexing
  • 1.2.1.7: Fixed a bug where pandas dataframes were being unshuffled when concatenated
  • 1.2.1.5: Added documentation and examples and ensured compatibility with Python 3.5+
  • 1.2.1.4: Original scores are now also bootstrapped to match the other results
  • 1.2.1.3: Corrected an issue with multithreading deadlock when returned scores were too large
  • 1.2.1.1: Provided object to assist in constructing scoring strategies
    • Also added two new strategies with bootstrapping support
  • 1.2.1.0: Metrics can now accept kwargs and support bootstrapping
  • 1.2.0.0: Added support for Sequential Selection and completely revised backend for proper abstraction and extension
    • Return object now keeps track of (context, result) pairs
    • abstract_variable_importance enables implementation of custom variable importance methods
    • Backend is now correctly multithreaded (when specified) and is OS-independent
  • 1.1.0.0: Revised return object of Permutation Importance to support easy retrieval of Breiman- and Lakshmanan-style importances
  • 1.0.0.0: Published with pip support!