Version 0.7.0

A major update to the BOOMER algorithm that introduces the following changes:

• L1 regularization can now be used.
• A more space-efficient data structure is now used for the sparse representation of binary predictions.
• The Python API does now allow to access the rules in a model in a programmatic way.
• It is now possible to output certain characteristics of training datasets and rule models.
• Pre-built packages for the Linux platform are now available at PyPI.
• The documentation has vastly been improved.

Version 0.6.2

A bugfix release that solves the following issues:

• Fixes a segmentation fault when a sparse feature matrix should be used for prediction that was introduced in version 0.6.0.

Version 0.6.1

A bugfix release that solves the following issues:

• Fixes a mathematical problem when calculating the quality of potential single-label rules that was introduced in version 0.6.0.

Version 0.6.0

This release comes with changes to the command line API. For brevity and consistency, some parameters and/or their values have been renamed. Moreover, some parameters have been updated to use more reasonable default values. For an updated overview of the available parameters, please refer to the documentation.

A major update to the BOOMER algorithm that introduces the following changes:

• The parameter --instance-sampling does now allow to use stratified sampling (stratified-label-wise and stratified-example-wise).
• The parameter --holdout does now allow to use stratified sampling (stratified-label-wise and stratified-example-wise).
• The parameter --recalculate-predictions does now allow to specify whether the predictions of rules should be recalculated on the entire training data, if instance sampling is used.
• An additional parameter (--prediction-format) that allows to specify whether predictions should be stored using dense or sparse matrices has been added.
• The code for the construction of rule heads has been reworked, resulting in minor performance improvements.
• The unnecessary calculation of Hessians is now avoided when used single-label rules for the minimization of a non-decomposable loss function, resulting in a significant performance improvement.
• A programmatic C++ API for configuring algorithms, including the validation of parameters, is now provided.
• A documentation is now available online.