in situations where the predictions take only a few values (e.g., a shallow decision tree), the current way of calculating ROC and AUC is prone to random changes depending on the order of the rows.
Moreover, the ROC is not composed of segments but has an unnaturally smooth shape.
Better to detect repeated predictions and improve AUC calculation
https://htmlpreview.github.io/?https://raw.githubusercontent.com/pbiecek/Open-Forest-Training-2021/main/rml-scripts-2021.html#Part_4:_Hyperparameter_optimization_+_Wrap-up
in situations where the predictions take only a few values (e.g., a shallow decision tree), the current way of calculating ROC and AUC is prone to random changes depending on the order of the rows.
Moreover, the ROC is not composed of segments but has an unnaturally smooth shape.
Better to detect repeated predictions and improve AUC calculation
https://htmlpreview.github.io/?https://raw.githubusercontent.com/pbiecek/Open-Forest-Training-2021/main/rml-scripts-2021.html#Part_4:_Hyperparameter_optimization_+_Wrap-up