Cyclic Boosting 1.4.0

• vectorized QPD in #75

Cyclic Boosting 1.3.0

new features:

• helper functionality for automatic interaction-term selection
• extended J-QPDs
• end-to-end quantile-regression chain with QPD

Cyclic Boosting 1.2.5

fixed wrong version number in init.py

Cyclic Boosting 1.2.4

bug fix concerning empty-bins corner case for multi-dimensional features in generic-loss mode

Cyclic Boosting 1.2.3

bug fix for parameter uncertainty in additive generic models (e.g., additive quantile regression)

Cyclic Boosting 1.2.2

• fixed analysis plots of quantile regression modes (show target quantiles rather than means)
• fix for weight usage in quantile regression modes
• implemented ordered smoothing of categorical features (was simply mapped to unordered smoothing before)

Cyclic Boosting 1.2.1

fix (in generic loss modes) for potential empty bins in multi-dimensional features

Cyclic Boosting 1.2.0

• enabled optional hierarchical iterations in training process
• added bin occupancy plots in training analysis plots of one-dimensional features
• added functionality to estimate full individual probability distributions from three estimated quantiles (e.g., from quantile regression) by means of quantile-parameterized distributions

Cyclic Boosting 1.1.2

fixed some misnomers

Cyclic Boosting 1.1.1

• some minor plotting fixes
• added quantile matching functionality to estimate full individual probability distributions from a few predicted quantiles