Skip to content

Latest commit

 

History

History
71 lines (57 loc) · 2.5 KB

RELEASE.rst

File metadata and controls

71 lines (57 loc) · 2.5 KB

Release 0.2.0

Import name change

The main module has been renamed from giottotime to gtime. Therefore, now all the imports should be changed in order to match the new name of the package.

Renaming

  • causality_tests to causality
  • shifted_linear_coefficient.py to linear_coefficient.py
  • shifted_pearson_correlation.py to pearson_correlation.py
  • feature_creation to feature_extraction
  • models to forecasting
  • standard_features.py to external.py
  • calendar_features.py to calendar.py
  • trend_features.py to trend.py
  • time_series_features.py to standard.py

Moving

  • calendar.py and external.py to feature_generation

Major Features and Improvements

  • Split TrendForecaster and Detrender. Before, the TrendForecaster was both a predictor and a transformer. To have a more sklearn-like interface, the classes are now splitted and the PolynomialDetrender and ExponentialDetrender classes have been grouped into the Detrender class.
  • The 'y' matrix is now created from the horizon_shit function, instead of being created by the FeatureCreation class.
  • The ShiftedPearsonCorrelation and ShitedLinearCoefficient classes now have the possibility to test shifts starting from a specified value, instead of always testing the shifts starting from 1.
  • The interface for creating new features through the FeatureCreation class has now changed and supports multiple columns as input.
  • Implementation of MovingCustomFunction class, in order to allow the user to apply custom functions with a rolling window.
  • General improvements in the documentation.
  • The topology features are no longer part of giotto-time, but will instead be implemented in giotto-tda.
  • Removed giotto-tda from the requirements.

Bug Fixes

  • Fixed a bug that was causing the transform method of the ShitedLinearCoefficient and of the ShiftedPearsonCorrelation classes to shift the time series in the wrong direction.
  • Fixed some bugs in the documentation.

Release 0.1.2

  • Documentation fixes.
  • The ShiftedLinearCoefficient and ShiftedPearsonCorrelation now can compute the p-values of the test through bootstrapping, in order to check for the significant of the test.

Release 0.1.1

Documentation fixes.

Release 0.1.0

Initial release of giotto-time.