Changelog

0.3.1

Fixed the documentation link in ‘setup.py’ - should now work correctly on PyPI.

0.3.0

  • Expanded support to Python 3.7 and Python 3.9 (previously only 3.8 was supported in ‘setup.py’);

  • Refactored and updated docs and README.rst file; and

  • Added initial continuous integration testing for multiple platforms on PR to dev or main branches by way of GitHub actions.

0.2.0-beta

A release that comes with added functionality + a bugfix and enhanced/updated documentation. The release comes with the following items:

  • Added functionality for computing the decision function on a given model index in the BoostedModel class;

  • Fixes intended functionality of validation_iter_stop in the BoostedModel class (#22); and

  • An example has been added to the documentation showing how a forward-propagating neural network can be built using pytorch.

0.1.1-beta

A bugfix release that comes with enhanced/updated documentation. The release comes with the following bugfixes:

  • Fixes both training and holdout stopping criteria in the main BoostedModel class (#18); and

  • Fixes internal learning rate calculation logic in the BoostedModel class (#17).

0.1.0-beta

The initial release provides boosting functionality via two main classes: BoostedModel and BoostedLinearModel.

The former is the most general class available and is meant to be used to build a boosted ensemble with any regression algorithm. The key is to provide a callable that returns a model object implementing a fit and predict method. Arguments can be passed to the callable via the model_kwargs_dict argument.

The latter class, BoostedLinearModel, is meant to be used with linear models that have coef_ and intercept_ attributes. The class provides additional methods relevant to linear models, and it streamlines prediction of the ensemble by taking advantage of the fact that the sum of linear models is simply a linear model.

Between the beta release and 1.0.0, documentation and examples will be updated. Thus, additional functionality will be exposed and usage of the package will become more transparent. The exposed API may change a little with 1.0, and will become more modular under the hood. It is also expected that additional functionality will be provided.