Added the possibility to use a custom validation_tuning_metric

Added the possibility to use a custom validation_tuning_metric in APLRRegressor by passing a custom Python function.

Reverted to updating the intercept in each boosting step

Reverted to updating the intercept in each boosting step. The reason is slightly better predictiveness on several datasets.

Deprecation of field and bugfixes

• Deprecated the constructor field intercept in APLRRegressor.
• Bugfix related to the Python wrapper that previously did not pickle correctly.
• Bugfix related to a warning when the model has not been trained yet but is attempted used.

Changed intercept estimation methodology and consequently deprecated intercept_steps

Changed intercept estimation methodology and consequently deprecated intercept_steps. The intercept is now fully estimated in the first boosting step.

Improved fitting when group_mse is used as a loss function. Fixed a minor bug.

Improved fitting when group_mse is used as a loss function. Fixed a minor bug related to handling of incorrect user input (an error is now thrown if m=0).

Added APLRClassifier, enabling two-class and multi-class classification

Added APLRClassifier, enabling two-class and multi-class classification. Also two small bugfixes in APLRRegressor and a renaming of the get_m() method to get_optimal_m().

Added the weibull loss function

Added the weibull loss function.

Added functionality and renamed parameters

Renamed the following:
"family" to "loss_function".
"gaussian" to "mse".
"tweedie_power" to "dispersion_parameter".

Added the "negative_binomial" and "cauchy" loss functions.
Added the "group_mse" validation tuning metric.

Added the possibility of specifying interaction constraints for each predictor

Added the possibility of specifying interaction constraints for each predictor. See interaction_constraints in API_REFERENCE.md for details. Also added a model term interpretation example in documentation/term_interpretation_example.xlsx.

Added family "quantile"

Added family "quantile" (for quantile regression).