Bugfix

Fixed a minor and rare bug related to sample_weight.

Improved the implementation of loss_function "group_mse_cycle".

Improved the implementation of loss_function "group_mse_cycle".

Added a new loss_function and a new validation_tuning_metric

Added the loss_function "group_mse_cycle" and the validation_tuning_metric "group_mse_by_prediction".

Bugfix

Bugfix related to the "group_mse" loss_function and validation_tuning_metric.

Bugfix

Fixed a minor bug regarding monotonic constraints.

Increased interpretability and added an option to ignore interactions when using monotonic_constraints

The new method, get_coefficient_shape_function(predictor_index:int), returns the coefficient shape across relevant predictor values, ignoring interactions. This increases interpretability by making it easier to analyse how main effects work in the model.

Added an option to ignore interactions when using monotonic_constraints. An use case for this can be to reduce the decline in predictiveness due to monotonic constraints when a large proportion of the predictors have monotonic constraints.

Bugfix

Fixed a small bug that could occur on seldom occasions. After the fix, the boosting procedure terminates if no change was made to the model during the previous boosting step.

Changed a default hyperparameter value and bugfix

Reverted the default hyperparameter value of boosting_steps_before_pruning_is_done to 0. Further tests indicate that pruning may more often than not slightly increase predictiveness, but this comes at a significant computational cost if the model gets many terms.

Fixed a rare bug related to the updating of term coefficients.

Bugfix

Minor bugfix related to pruning.

Set the default value for boosting_steps_before_pruning_is_done to 500

Set the default value for boosting_steps_before_pruning_is_done to 500. The reason is that empirical results so far indicate that this slightly improves predictions compared to not pruning. The drawback of pruning is somewhat increased model training time, especially when the number of terms gets large.