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 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 a ddl.linear.LinearProjector).
Added bias/shift term to ddl.linear.LinearProjector

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 and ddl.base.create_inverse_canonical_destructor
Renamed ddl.base.get_implicit_density to ddl.base.create_implicit_destructor and renamed ddl.base.get_inverse_canonical_destructor to ddl.base.create_inverse_canonical_destructor (with previous names kept for backwards compatability but will issue a DeprecationWarning).
Fixed various small bugs.

Significant documentation has been added including:
- New demo notebooks
- Docstrings for many classes and functions

Changed n_dim to n_features everywhere in the source to match with scikit-learn convention.
Moved ddl.deep.CompositeDestructor to ddl.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 and ddl.deep.DeepDestructorCV to use the random_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 into ddl.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’s scipy.stats.rv_histogram

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 in ddl.validation.check_destructor and :param:`properties_to_skip` can be used to check only one property if needed.