History¶

0.3.2 (03.10.2019)¶

Fixed Discretizer
- Properly addresses the cut-offs in the transform method, when not enough split points is provided.
- Adds random_state parameter for reproducibility.
Added fit_intercept parameter to the VIFSelector
- Extends functionality through the ability to control the model intercept. Statsmodels natively do not include the intercept in the linear regression so to properly execute the VIF selection it is recommended to add the intercept and aforementioned parameter provides this.

Fixed MANIFEST.in
- Includes extras_requirements.json to distribution package.

Initiated xai subpackage
- Creates base class ExplainerMixin for all explainers with the fit & explain convention (familiar from scikit-learn API).
- Releases, in the development version, the xai.FeatureEffectExplainer implementing calculation of Partial Dependence Plot, Individual Conditional Expectation and Marginal Plot.
Optimized Discretizer
- Limits the transformations only to continuous variables by not taking into consideration categorical variables and leaving them unchanged.
Created optional dependencies
- Reduces the number of dependent packages by creating optional functionalities that require additional installation of the extra requirements.
Renamed collinearity_reduction module
- Changes the module name to feature_selection to prepare it for future expansion in this direction.
Modified utils.is_numeric
- Excludes from checking for numericity categorical pandas.Series.
- Adds the ability to disable the attempt to project on a numeric type.
Moved mathy module
- Changes the location of mathy module to utils subpackage. The new location is as follows: utils.mathy.
Fixed missing values temporary imputation inside the VIFSelector
- Imputs the missing values before checking the NaN condition.

Added PandasFeatureUnion transformer
- Creates feature_union module with PandasFeatureUnion transformer implemented that concatenates multiple transformers returning pandas.DataFrame.
Fixed preprocessing.ColumnProjector corner case handling
- Checks whether the list of columns to be projected is non-empty, because when projecting an empty list of columns onto the category dtype, ValueError was raised.
Expanded TargetEncoding docstring and unified types
- Adds notes on the use of the indicated encoder.
- Projects to make the transformed dataframe’s columns being outputted as floating point type.
Changed utils.check_column_existance name
- Repairs the typo in the function name by changing it to: utils.check_column_existence.
Expanded Sphinx documentation.
- Adds example usage of force_context_manager function.
Added utils.check_continuity function
- Creates function that asserts whether the variable is truly continuous at a given repetition threshold.

Changed the topology of the repository
- Transfers the column_parsing module to the preprocessing subpackage and changes the name to a more adequate in terms of actual functionality.
- Creates preprocessing.imputation module used for imputation of missings.
Added Sphinx documentation
- Creates HTML documentation of the package with use of the Github Pages.
- Standardizes the docstrings across the whole package.
- Adds initial example of usage for the paralytics.Discretizer transformer.
Created utilities for web scraping
- New module preprocessing.scraping presenting the base class for scraping using the Selenium package.
Fixed functionality of defective transformers
- Transfers key operations of the preprocessing.ColumnProjector transformer to the fit method to avoid errors when applying to corner cases.
- Adapts the functionality of the TargetEncoder to be able to apply its transformations to entire DataFrame without first choosing only variables of the type: category.
Initialized XAI subpackage
- Initializes directory for the Explainable Artificial Intelligence subpackage which will be developed over the 0.2.X versions.