Release history

Version 0.2.2

Changelog

  • Bug fixes

Version 0.2.1

Changelog

  • Bug fixes

  • Rename abstract base class ModelFactory to SoftMachine

Version 0.2

Changelog

Adds core set of data descriptors, basic feature preprocessors and first regressor, thoroughly revised api.

New algorithms

  • data descriptors:

    • ALP

    • CD

    • EIF (wrapper requiring optional eif dependency

    • IF (wrapper for scikit-learn implementation)

    • LNND

    • LOF

    • MD

    • NND

    • SVM (wrapper for scikit-learn implementation)

  • feature preprocessors:

    • LinearNormaliser

      • IQRNormaliser

      • MaxAbsNormaliser

      • RangeNormaliser

      • Standardiser

    • SAE (requires optional tensorflow dependency)

    • VectorSizeNormaliser

  • regressors:

    • FRNN

API changes

  • Uniform ModelFactory pattern: callable algorithms that create callable models.

  • Preprocessors can be included at initialisation and are applied automatically.

  • Algorithms are presented no longer by submodule (neighbours, trees, etc), but by type (classifiers, feature preprocessors, etc)

  • Many changes and additions to secondary functions that can be used to parametrise the main algorithms.

Version 0.1

Changelog

Adds number of existing fuzzy rough set algorithms.

New algorithms

  • FRFS

  • FRONEC

  • FROVOCO

  • FRPS

API changes

  • neighbours.FRNNClassifier replaced by neighbours.FRNN.

  • Classifiers give confidence scores; absolute class predictions can be obtained with utility functions.

  • Classifiers follow construct/query pattern; scikit-learn fit/predict pattern can be obtained with utility class.

  • neighbours.owa_operators moved to utils.owa_operators.

  • utils.OWAOperator no longer initialised with fixed k, has to be passed to method calls instead.

  • utils.OWAOperator method calls and functions in utils.np_utils now accept fractional and None k.

Version 0

Changelog

First release, by Oliver Urs Lenz.

New algorithms

  • Fuzzy Rough Nearest Neighbour Classification with Ordered Weighted Average (OWA) operators