Changelog¶
All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog and this project adheres to Semantic Versioning.
[Unreleased]¶
Added¶
- Added
create_fitted
class method to many densities and destructors to make manual creation of fitted estimators simpler and more uniform. - Added
ddl.base.create_inverse_transformer
to create inverse transformations that may not be destructors (e.g., an inverse of addl.linear.LinearProjector
). - Added bias/shift term to
ddl.linear.LinearProjector
Changed¶
- Fixed bug in
ddl.univariate.HistogramUnivariateDensity
when bin widths were not uniform. Scipy histogram do not normalize for bin widths so that is needed before calling Scipy histogram. - Updated implementation of
ddl.base.create_implicit_destructor
andddl.base.create_inverse_canonical_destructor
- Renamed
ddl.base.get_implicit_density
toddl.base.create_implicit_destructor
and renamedddl.base.get_inverse_canonical_destructor
toddl.base.create_inverse_canonical_destructor
(with previous names kept for backwards compatability but will issue aDeprecationWarning
). - Fixed various small bugs.
[0.0.2] - 2018-08-21¶
Added¶
- Significant documentation has been added including:
- New demo notebooks
- Docstrings for many classes and functions
Changed¶
- Changed
n_dim
ton_features
everywhere in the source to match with scikit-learn convention. - Moved
ddl.deep.CompositeDestructor
toddl.base.CompositeDestructor
since it seemed more like a fundamental building block rather than a deep component. - Fixed functionality of
ddl.base.CompositeDestructor
,ddl.deep.DeepDestructor
andddl.deep.DeepDestructorCV
to use therandom_state
parameter to set the global random state so that the random state for each sub destructor does not need to be set manually. - Made several univariate densities private since not used outside of module.
- Merged
ddl.deep.DeepCVMixin
intoddl.deep.DeepDestructorCV
since no useful case for mixin. - Removed explicit dependency on pre-built mlpack and updated docker/singularity images.
- Updated build system especially for circleci to make it simpler to build mlpack extension.
- Simplified
ddl.univariate.HistogramUnivariateDensity
implementation by merely using scipy’sscipy.stats.rv_histogram
Removed¶
- Removed
ddl.base.AutoregressiveMixin
because it was not needed. - Removed (by making them private) the validation functions
ddl.validation.check_destructor_interface
,ddl.validation.check_uniformability
,ddl.validation.check_invertibility
,ddl.validation.check_canonical_domain
,ddl.validation.check_identity_element
, because these are already called inddl.validation.check_destructor
and :param:`properties_to_skip` can be used to check only one property if needed.